This thing still any good?

>>105704394 (OP)
yes but it's starting to show its limit in the most recent slop at 1440p and above

>>105704411
Shouldn't the strain on the CPU go down the further up you go in resolution?

>>105704394 (OP)
for productivity it's still well worth it

>>105704394 (OP)
in multithreaded tasks is still decent.

In single threaded tasks and gaming its nearing end of life at this point, will start to struggle and lag behind in newer bloated and unoptimized games. IPC and clocks have gone up a lot in the pass 5 years. Even worse as it is not a 3dvcache variant.

Expect lag in GTA 6, but playable.

>>105704394 (OP)
2500k is still good

>>105704394 (OP)
i still have one, it does a good job keeping my room nice and warm in the winter

For most people, I'd say it still holds up. If you don't really need DDR5 or PCIe 5.0 support, you should be alright.

>>105704394 (OP)
Perfectly fine unless you're some gamer and wanting peak FPS bro for your 240hz monitor or some other inane shit.

>>105704394 (OP)
i've never been bottlenecked by my 5800X
so you should be fine

as always, gaymers are mentally ill, OPs CPU is still a monster.

>>105704453
>In single threaded tasks and gaming its nearing end of life at this point,
lol the fuck are you talking about?

>>105704394 (OP)
I'm still using its little brother (5900x) on my desktop and yes, it's still good, great even
>>105704499
also that. tends to turn my room into the flames of hell in the summer though

>>105704394 (OP)
It's maybe like 30%-50% slower than top of the line. Literally nothing.

>>105704394 (OP)
I'm not changing my 5900x until we get second generation AM6.

AM4->AM5 improvements aren't that impressive to me.

>>105704394 (OP)
My 5600 (non X) is still doing just fine. Does everything I need, even with PS3 emulation.

>>105705083
My 2600 does everything I need, including PS3 emulation.

i am yet to be bottlenecked by my 5600G

>>105704394 (OP)
I have a Ryzen 5 7600, ask me anything

>>105705116
Can you do an in socket upgrade or are you stuck? I can still go up to a 5800X3D, although never will since it's expensive as fuck, and probably be stuck with the affordable 5800XT to prolong the life of my system even more.

>>105704394 (OP)
with higher power limits, it can touch tips with the 7900x , definitely still very good. paired with fast ddr4 and the single core boost above 5ghz, still very decent against the newer generation of chips.

real shame amd never released the planned x3d variant.

>>105705171
>real shame amd never released the planned x3d variant.
Was it because of the heat issues?

>>105705183
maybe hard to say, but I think it was mainly because it's launch would be too close to the 7000 series launch , which would cannibalize sales. the 5800x3d cannibalized sale, but to a lesser extent because it isn't good for productivity.

Never gotten my 2700X to hit 100 % outside of some file compressing. Tight memory helps.

>>105705167
why would ever get this over 7500F without the iGPU?

AM4 is so comfy brudis

5900X good enough for my needs. I'll wait for AM6, don't really like scAM5

>>105705371
had one until recently, it bottlenecks hardcore in games even though it's not hitting 100%

>>105704394 (OP)
its gonna be a long time before i upgrade my 5800x

>>105704453
You maean games not worth playing

>>105704394 (OP)
if you want mulit-core, just get a cheap epyc build (32-64 cores). You can still run a GPU and windows on the server boards if you want a workstation. if it's for gaming, go with something with better single core performance.

>>105704837
This!

>>105705183
I think they had issues with latency due to the nature of these larger chips already being on 2 chiplets glued together. Why add more l3 cache when it's going to get nerfed by having to sync between two chips? There are some server/multi-core workloads that benefit from extra l3 cache, but for the most part it was a single-core performance enhancement so it made more sense to put it on chips with just 1 chiplet, from what I remember/understand.

Moore's law died a largely-unnoticed and expected death sometime around 2005, so new computers are no longer many times more powerful than used ones. So, why on Earth would the average person who only wants to watch movies, visit Facebook, surf the Internet, and shop online want to buy a new computer when a used one can easily do all of that? The cooling loop systems are largely unnecessary, because computing, today, isn't that demanding where you need lots of power that'll produce heat enough to act as a small space heater, unless, you're gaming and AI-slopping 24/7, nonstop.

5900x and 5950x are more than enough for everyday tasks and gaming fit the next 5 to 10 years. Most games that are worth playing do not even require having a 5900x OE 5950x, so they are way more than sufficient.

Computing has been stupid efficient and overkill for over 10 years. They are merely repackaging incremental, unnecessary "upgrades" and selling FOMO. NONE OF IT is necessary for the average user who spends 99.9% of their time and usage on a computer IN A BROWSER. Phones and phablets, too, have essentially replaced the desktop and GPU features are dwarfed by the number users who do not need a GPU for 99.9% of features offered. A CPU the likes of a 5900x and 5950x are still overkill by today's standards and everyday task requirements.

and lets not forget that modern games have become benchmarks for obsessed consomers (aka "enthusiasts") and are designed to run badly purposely on perfectly capable hardware, its all a big scam chain.

>>105705170
Get a 5700x3D off aliexpress

>>105705171
>paired with fast ddr4
zen 3 cpus barely scale past 3000 mhz cl14 but retards will always imply they do (and by a lot) while showing 0 proof

>>105706909
3800mHz cl14 checking in

>>105706946
>he is either running 2:1 mode losing performance or cooking his memory controller with turbo autismo overclocking
ok, cool, how's that relevant to my post?

>>105706909
if you bother to tune the subtimings it can make a pretty big difference on am4, and most of the gain comes from the increased infinity fabric speed , not the memory, bog standard xmp won't cut it.

>>105706712
Unless I'm on the wrong aliexpress, it's like 400+ dollars on there and less than 300 on newegg. Are you trying to fool me? 5800XT is around 150-160 new.

>>105706909
picrel
>>105708440
5700X non 3D is the best choice on AM4

>>105704394 (OP)
this baby dropped to 250€ right now in europe from 320€.
still thinking to swap it with my 3700X to give my PC some longer life since I don't want to change to AM5 now which is also EOL.
My GPU is a 6750XT so the CPU won't limit.

>>105704394 (OP)
anything from zen 3 and up will hold for at least 5 years

maybe 7 if you dont give a damn about pure perfomance

>>105704421
Only relatively to the GPU. In absolute terms it would be higher with more pixels.
Put 5900x in my new rig half a year ago, gaming on 1440p is great, compiles LineageOS in about 2 hours. Don't like how AM5s look like AI generated sci-fi slop.

How's the 3900x these days? I have one lying around, was wondering if I should use it for a little build for my living room.

>>105706909
>while showing 0 proof
here you go retard
>>105707007
>cooking his memory controller with turbo autismo overclocking
board default is 1.2 V
you can do 3800 at 1.0-1.1 V
you're prolonging the life of the memory controller
>>105706946
based
>>105708700
aaand based

5700x3d is based. Upgraded from 3700X and it kicks ass. Wasn't expecting so much performance increase outside games as well, but in games of course it's kicking ass.

>tfw I spent 890 euro bucks buying this piece of shit from some scalper because I thought I NEED more cores when a 5800X would have been more than enough for anything I do

>>105709835
How do you not know how many cores you need?

>>105709910
I fell for the future-proofing meme

>>105709966
Live and learn

>>105708700
yeah your picrel shows that a much higher ram frequency + tuned timings (of course we have no idea how much and you never bothered to test same frequency + tuned timings vs higher frequency + same timings) barely increases performance
>>105709483
what's this supposed to prove? there's literally no comparison and no memory timings/frequency specified anywhere, plus it's getting rekt from the 5800x3d so how does it prove "it's touching tips with the 7900x"?
>you can do 3800 at 1.0-1.1 V
...if you got a golden sample or something maybe
no thanks, i'm sticking to 3200cl14 @ 0.8v
>you're prolonging the life of the memory controller
I'm not worried about that, it's probably fine up to 1.4v, it's the 30+ watts of power consumption at idle that's retarded

>>105711201
>it's getting rekt from the 5800x3d
it's not
a tuned 5800X can beat a stock 5800X3D in many games/tests
but a tuned 5800X3D can almost certainly beat a tuned 5800X in everything
>if you got a golden sample or something maybe
i don't know but this setup passes every test except P95 small FFTs at 1.0 V
1.05 V is rock solid, passes several overnight tests
i haven't touched any settings for like 6 months
>i'm sticking to 3200cl14 @ 0.8v
that's a very reasonable choice, i'm thinking of turning this rig into a home server after i upgrade and then i'll probably be running similar settings
i did a quick test with the 50k factorio benchmark and the difference between 3200C12 to 3733C14 is like 7%, most of your uplift (as always, on any platform) comes from low latency timings
>it's the 30+ watts of power consumption at idle that's retarded
makes sense

>>105711201
>>105711290
factoriobox.1au.us, Flame_Sla's 50k SPM belt map

>>105711290
>but a tuned 5800X3D can almost certainly beat a tuned 5800X in everything
Not really, there's a 10% clock speed difference between the two.

>>105712064
i don't own one so i can't give a definitive answer but 10% clock speed isn't massive
i can only compare this 5800X with tests that other people have done with stock 5800X3Ds (XMP is still stock, you didn't do it yourself)

>>105712095
10% is a pretty big difference in a cpu, it's the difference between a sandy bridge (2011) and a skylake (2016)
not to mention the big ass cache slows down memory access

>>105712124
2015*

>>105712124
>it's the difference between a sandy bridge (2011) and a skylake (2016)
also a big IPC difference between those two
>not to mention the big ass cache slows down memory access
9800X3D has a BIOS option that messes with memory latency, "Zen 5 Gaming Optimizations"
all those tests need to be done on the same BIOS version for zen 5

i've seen people get the same or better memory latency as me on 5800X3Ds
this 5800X kinda sucks ass, can't do 1900 FCLK, can't do more than -10 CO, can't do more than 4.9 GHz
but i'm sure there's a few games where non-X3D is superior

the real lesson is that you should always tune your hardware

>>105712187
Any guides for how to get started with this?

>>105712268
the DDR4 OC guide is a decent-ish starting point
https://github.com/integralfx/MemTestHelper/blob/oc-guide/DDR4%20OC%20Guide.md
if you are running 10th gen intel it's perfect but otherwise there are some inaccuracies and missing information
for example it doesn't have tertiary timings for AMD and it doesn't have any information at all for 13th/14th gen intel (anything from 12th gen is mostly applicable, but they do have updated memory controllers)

i would start by finding out
1. what ICs your memory kit has (the guide explains this)
2. what your maximum fabric clock (FCLK) is, you can decouple this from the memory during initial testing

the OCCT combined CPU+RAM test is good for FCLK testing, it should tell you fairly quickly if you're unstable by throwing WHEAs or even blackscreening and resetting the system
you might need to adjust SOC and VDDG voltages, sensible range for SOC is 1.0-1.2 V, sensible range for VDDG CCD is 0.9-1.0 V, sensible range for VDDG IOD is 0.9-1.1 V
once you know your maximum FCLK you can follow the guide according to the kit you have

good tests for memory (not memory controller or FCLK) is testmem5 with the ABSOLUT profile (free), karhu (paid) and OCCT (paid >1 hour)
for the memory controller run prime95 with the large FFTs option

sorry if that makes no sense but it's been a while since i did memory overclocking so i don't remember all the details, just follow the guide and look up what results other people have gotten with similar kits and CPUs

>>105712268
>>105712405
this video is for AM5 but most of it is still applicable to AM4
https://www.youtube.com/watch?v=Xcn_nvWGj7U
ignore anything about gear 2, or desynced fabric clock, you ALWAYS want to run your fabric synced 1:1 with memory on AM4

>>105712405
>>105712461
Awesome thanks a bunch for the helpful info. I'm going to see how much more I can eke out of my 5900x

>>105704394 (OP)
Yes.
I’m still running a 1950x to this day and occasionally gaming on it (with a 6900XT) and while it does have a loooot of bottlenecking it’s still doing the job and the cores are useful still. Might have to finally upgrade but the 9000 series didn’t look too good.

>>105712540
based
and a reminder, as always, you get 80% of the results with 20% of the effort
i've squeezed everything i can out of this memory kit but that's because i think it's fun, i could have stopped long ago and still gotten very good results
regardless you will be running circles around people who only know how to enable XMP

>>105704394 (OP)
Yeah, it's alright. Mine is in my home server now and I use it to run a Windows VM with GPU passthrough as well, performance is still quite good. That VM has a 3080 Ti in it and when using it I mostly run into GPU bottlenecks rather than CPU.

>>105704411
CPU load doesn't really increase with resolution, only GPU load does.

>>105704421
Nah, that's also not quite right. CPU load remains the same while GPU load increases with resolution. So running 100FPS at 720p or 100FPS at 4k is the same for the CPU but harder for the GPU.

>>105705716
aw heck year

>>105711302
I've spent the past half an hour dicking around with this shit and it seems it's a benchmark that was specifically crafted to only scale with memory speeds and nothing else, barely uses any cpu
...so how does that give you conclusive data on how much faster ram actually helps in gaming?
those percentage gains aren't even that impressive when considering it's a ram-only benchmark, and that 3733 mhz setup once again shows very noticeable diminishing returns
>a tuned 5800X can beat a stock 5800X3D in many games/tests
ok, show me those games/tests then?
and even if it can, it can only do so when running veeery inefficiently (like 5 ghz peak boost + insanely fast ram, twice the power consumption of a 5800x3d)
>most of your uplift (as always, on any platform) comes from low latency timings
I've always seen the opposite raccomended tho

>>105712064
you have to consider that performance doesn't scale linearly with clockspeed at all
past 4 ghz the gains from higher clockspeeds get slimmer and slimmer, the x3d chips pretty much run at the clockspeed sweetspot where going higher barely improves performance
so a 5800x3d is losing like 2% of performance due to lower clock speed, while gaining 20+% from the bigger cache
how is that a problem
>>105712124
>the big ass cache slows down memory access
how is that a problem when it also reduces the need for memory access by an insane amount?
sure there are some edge case scenarios where an x3d is slower, but only slightly, rarely, and still while being more efficient, it's all cope really.
"a turbo autismo tuned 5800x can sometimes be very slightly faster than a 5800x3d" sounds way less impressive than what you niggers seem to be implying

>>105713262
you should definitely consider getting a 5700x3d, the 5700g is dogshit (2600 tier gaming performance)
plus you have a b550 and a gen 4 x8 gpu so not being able to run gen 4 because of a gen 3 cpu is kinda retarded (realistically not a problem unless you start getting into vram issues but whatevs)

>>105713480
>it seems it's a benchmark that was specifically crafted to only scale with memory speeds and nothing else, barely uses any cpu
factorio is a real game, it's not a benchmark tool like 3dmark or unigine
it just has a very convenient benchmark mode that you can run straight from the command line
>so how does that give you conclusive data on how much faster ram actually helps in gaming?
because most CPU-intensive games scale with fast RAM
>those percentage gains aren't even that impressive when considering it's a ram-only benchmark
it's not a "RAM only" benchmark, if you ran 3733C14 on a 14900KS it would outperform the 5800X by a lot, because the cores are faster
>it can only do so when running veeery inefficiently (like 5 ghz peak boost + insanely fast ram, twice the power consumption of a 5800x3d)
this is true, i wouldn't say twice but X3D chips are very efficient
but i think you're confusing me for the other anon because i never said you avoid X3D
>I've always seen the opposite raccomended tho
no one tests anything

>>105704394 (OP)
Dunno, is my Ryzen 5 3600 and RTX 2070 Super still any good? I can still play everything I want to with no problems at all. I guess it's all about what you want to do.

>>105713575
>factorio is a real game, it's not a benchmark tool like 3dmark or unigine
yeah, but again, the map you provided seem to have been prurposely built for benchmarking, and it barely uses any cpu while scaling more with memory than any other game I've ever seen, it's still a forced result.
>because most CPU-intensive games scale with fast RAM
most cpu-intensive games are actually cpu-intensive, you know...
I wouldn't call something that uses 5-10% of my cpu "cpu-intensive", it's barely maxing out a thread like 30% of the time, by comparison cities skylines 2 will use 100% of my cpu, beamng drive at least 50%, this thing is just ram intensive.
>no one tests anything
says the retard claiming a tuned 5800x is faster than a 5800x3d with absolutely 0 data to back it up

the x3d chips are good for non tech-savvy people who just want a fast computer and are willing to pay double for it

>>105713721
>the map you provided seem to have been prurposely built for benchmarking
yeah, it's supposed to be an intensive test
if you were getting only 50 UPS in-game you would feel it
>and it barely uses any cpu while scaling more with memory than any other game I've ever seen
i hope you understand that most games don't use all your cores and therefore never hit 100% CPU usage
and MMOs scale more than factorio, but most MMOs don't have a dedicated benchmark mode
>says the retard claiming a tuned 5800x is faster than a 5800x3d with absolutely 0 data to back it up
there's been several data points posted in this thread, what do you want me to run?

>>105713566
I'm planning to do exactly that and use the 5700G for an HTPC, I just have had other expenses lately.

>>105713721
if big cache were important the i7-5775c would wipe the floor with everything
there is more to cpus.

>>105713753
>i hope you understand that most games don't use all your cores
there's a big difference between "not using all your cores" and "using a single thread but only half the time"
this benchmark is clearly a very edge case scenario, I've always played cpu-bound simulation games and this is the first time I see something like this
>there's been several data points posted in this thread
I already explained a thousand times why they are all very inconclusive
>what do you want me to run?
3200cl16 5800x3d vs 3200cl16 5800x
+
turboautismo 5800x3d vs turboautismo 5800x
this way we can prove/disprove the following claims:
1. a 5800x on turboautismo ram is faster than a 5800x3d running regular ram
2. a 5800x3d scales far less with faster ram than a 5800x does
my prediction is that point 1 is complete bullshit coping and 2 is also not really the case (better scaling sure, but not by as much as some people are claiming)
also use gta 4 or something like that which people actually play and which resembles a real video game (and also the best and only game worth playing desu) rather than that factorio retardation

>>105713877
>i7-5775c
only has 6 mb of l3 cache, am i missing something here?
>there is more to cpus.
sure... but here we're literally talking about SAME SILICON with cache and clock speed as the only determining factors, no one here has ever claimed that cache is the only thing that matters, yeah I know that a 7600x is faster than a 5800x3d no shit?

>>105713919
pic
>>105713937
128MB

>>105713877
cache is important, fast RAM is important, IPC is important, frequency is important, core count is important
you have to put in the most work to get your RAM fast and stable so redditors love to put their head in the sand and pretend it doesn't matter
>>105713919
>I already explained a thousand times why they are all very inconclusive
because you want them to be
i used to play a lot of wow and RAM tuning made a huge difference there, not a synthetic workload at all
>3200cl16 5800x3d vs 3200cl16 5800x
>turboautismo 5800x3d vs turboautismo 5800x
>this way we can prove/disprove the following claims:
>1. a 5800x on turboautismo ram is faster than a 5800x3d running regular ram
>2. a 5800x3d scales far less with faster ram than a 5800x does
i'm not gonna go out and buy a 5800X3D, it's discontinued and there's better options on the market
i think you care about this more than me
do whatever you want, go out and build your dream system with some shitty JEDEC sticks, no one cares
>>105713962
benchmarks from 1.x.x are not relevant, the game was optimized with the 2.0 release

>>105713991
it shows a well tuned zen 3 is as fast as a non tuned zen 5

>>105714023
How do I tune a 5800X3D

>>105714055
CO, BCLK, RAM

>>105714081
but you don't want to get hog wild with the ram because the power and current limits include the io die

>>105713098
it depends on if you bottleneck or not, if you are fully pegging your GPU then CPU load *will* go down

>>105714143
maybe not the current limit but the power limit definitely

>>105704499
>105W
rookie numbers.

>>105714081
>>105714143
I'm running this memory kit: https://geizhals.de/g-skill-ripjaws-v-schwarz-dimm-kit-128gb-f4-3600c18q-128gvk-a2271234.html
And I don't intent on buying anything faster.
I've already undervolted Vcore according to this:
https://www.youtube.com/watch?v=BOdolaIDADk

Is the only thing left BCLK? Because I don't want to fuck up my hard drives with that.

>>105714218
tighten the timings and pull your vSoC down
cl18 is what you run at 4000 with mid ram, at 3600-3800 you're looking to run cas 16
https://www.overclock.net/threads/a-guide-to-ram-overclocking-on-zen-3.1798093/

>>105714143
>the power and current limits include the io die
easy solution
>>105714218
>And I don't intent on buying anything faster
it's not about buying anything faster
first impressions says that's not a great memory kit but it doesn't matter, you can still increase the frequency and tune the subtimings
and who knows maybe you got a good bin
refer to >>105712405
just make sure you stay within safe voltage limits because not all ICs can tolerate 1.5+ V
>>105714251
that guide is terrible, just saying
don't leave anything on auto
don't blindly copy settings from other people

>>105714285
i don't think you can adjust pbo settings on 5800x3d, the chips would burn up

>>105704421
No, the strain doesn't go down because CPU things are still CPU things regardless of resolution. The reason why people claim "gpu matters more at higher resolutions" is because the higher resolution you go, the higher the load on the gpu. The gpu often becomes the bottleneck.
The reason you do not see this scaling with cpu is because cpu things are as noted, cpu things. Stuff like AI, NPCS, draw calls, computational things, physics, etc none of that stuff scales with resolution. Its the same regardless the resolution. While things like textures, lightning, all of that, does scale with resolution because the higher the resolution, the more pixels from those things need to be rendered.

A lot of people mask a CPU bottleneck with a GPU bottleneck because they slap a really shitty GPU into their computer and buy the biggest monitor (resolution) they can.

>>105714321
oh right
but i heard it was unlocked at some point?
1000/1000/1000 is just a meme, it's not gonna hit that, ever

>>105714285
>that guide is terrible, just saying
>don't leave anything on auto
>don't blindly copy settings from other people
the guide doesn't tell you to set anything to auto except maybe trfc2,4 tcke

>>105714338
>draw calls
Forgot to note, if you have a really powerful gpu, you can run into the issue where the cpu cannot keep up with all the frames being shitted out by the gpu. You can "mitigate" this with a higher resolution, only because you are reducing your performance (less frames).

>>105714321
>>105714345
You can on Asus boards.
But I can't just turn it on in Ryzen Master because I'm on Linux.
>>105714285
>first impressions says that's not a great memory kit but it doesn't matter
It's alright, the 14-14-14-34 kit wasn't available anymore and the 16-22-22-42 kit is the same as my 18-22-22-42 with some lipstick on it.
>you can still increase the frequency and tune the subtimings
No clue how to do that.
>and who knows maybe you got a good bin
I'll need to see if that's the case
>refer to >>105712405
Will do
>just make sure you stay within safe voltage limits because not all ICs can tolerate 1.5+ V
And those safe limits would be?

>>105714375
there are reports of cjr degrading over 1.4V
I think you have djr

As long games will need to run on a base PS5, anything higher than Zen 2 will be enough.

That's the reason why the Q6600 and the 2500K were so famous for their longevity, because they were faster than the Xbox 360 and the PS4 CPUs respectively.

>>105714375
>the 14-14-14-34 kit wasn't available anymore
yeah, that's B-die and those kits have all been discontinued long ago, sadly
>the 16-22-22-42 kit is the same as my 18-22-22-42 with some lipstick on it
for sure, if it was 16-18-18 it would be a better bin, low tRCD-tRP is the main indicator of RAM quality/potential
>And those safe limits would be?
check the guide
i have micron rev E so i could put in 2 V if i feel like i need some more excitement in my life, but yours might die at 1.5 V

>>105704453
>NOOOOOOOOOOOO, NOT MY HECKIN' AAAAA SLOPERINOOOOOOO

>>105714406
Q6600 was faster than PS4
https://browser.geekbench.com/v6/cpu/12511183 not even OCd
https://browser.geekbench.com/v6/cpu/3218520

>>105714388
Aren't those 16GB modules only?
>yeah, that's B-die and those kits have all been discontinued long ago, sadly
Yep, I was still running a 4790K when they killed off B-Dies.
>for sure, if it was 16-18-18 it would be a better bin
The only 16-18-18 G.Skill was 3200 as far as I know. 3600 starts at 16-22-22-42 for 128GB kits at 3600.
>low tRCD-tRP is the main indicator of RAM quality/potential
Yes, but none were available when I put this thing together at a reasonable price.
>i have micron rev E so i could put in 2 V if i feel like i need some more excitement in my life
Wouldn't that fuck with the io die of the CPU?

>>105714493
Everything from the second line on was meant for >>105714423

>>105714493
the first thing you have to do is boot windows and get thaiphoon burner to identify your sticks

>>105714540
Hold on, lemme reboot.

>>105714493
>>105714503
>Yes, but none were available when I put this thing together at a reasonable price
yes it's fine, i'm not trying to shit on your setup
there are two reasons to overclock
1 is trying to get the most out of your setup
2 is trying to get a world record or get first place in an OC competition
if you aiming for 2, then yes you do need to bin a million different kits and you need the very best RAM, CPU and board you can find
but for a daily setup, who cares? once you've gotten everything you can out of the system and you're satisfied with the performance, don't worry about it
even just getting it down to say 3600 16-20-20 with tuned subtimings would be a big improvement
>Wouldn't that fuck with the io die of the CPU?
lol maybe
i would obviously never do that, but the memory sticks would survive

>>105714540
Just booted up on Windows, where the fuck can I even get thaiphoon anymore?
This site https://www.thaiphoonburner.com/ gives me FREE CANDY van vibes and I don't know if I can trust this https://www.softpedia.com/get/Tweak/Memory-Tweak/Thaiphoon-Burner.shtml site either.
Also, holy shit windows is so fucking bad, I feel like I have lead in my shoes.
>>105714624
>1 is trying to get the most out of your setup
That's what I'm after.
>i would obviously never do that, but the memory sticks would survive
Gotcha

>>105714669
try the softpedia link, the other one is more suspicios

And here I am, still living comfortably with my i5-4690, having no issues.

>>105714733
ok

>>105714669
it works but if you get a virus i'm very sorry
the file name they gave me is the same I got from the official website a few years ago I think

>>105714725
>>105714802
I think it's C-Die?
Also BRB, gonna go back to Linux.

>>105704394 (OP)
lack of AVX-512 is crippling its potential longevity but it's more than enough for most everyday and even some non-everyday tasks.
I use that with 128GB RAM to run multiple containers, VMs, games, browsers, code compilation, whatever at the same time. I cannot sense any slowdown.

>>105714802
>>105714832
>using windows
LOL

>>105714840
>AVX-512
Why is that so important for longevity?

>>105714845
I use linux normally but you don't get low level hardware access with it so I have to boot windows sometimes whether I like it or not

>>105714832
It's MJR. It does 16-20 etc at 3600
I don't know what trfc it does

>>105714845
I'm back on Linux, wtf is even your problem?
>>105714867
Cause games are going to use it soon.
>>105714922
>MJR
Thanks.
>I don't know what trfc it does
Is this what you're looking for?

>>105714832
it's 16gbit AJR or MJR
won't have the same OC characteristics as 8gbit AJR or MJR, FYI
>>105714922
here
it's pretty bad

>>105714956
Explains why it was so cheap. 220€ for 128GB was just too good to pass up on.

>>105714956
huh, that's not very good but not the worst I've seen.
>>105714950
it's pretty impressive that they nailed it down that close

>>105714976
tRFC is just a static performance nerf, there are other timings that have a bigger performance impact
350 ns tRFC, 7800 ns tREFI means you are losing (350/7800) cycles to refresh
that's 4.4%, which is objectively pretty bad, but you can still tighten other timings and get good performance

I am running 2 sticks of this ram with a 3700X running in bios with XMP pre-set settings.
3200 at CL14 should be a pretty good deal
Is it worth it to try and getting 3600 at CL14?
or is it even possible?
any suggestions?

DDR4-3200 CL14-14-14-34 1.35V
https://www.gskill.com/product/165/170/1535961634/F4-3200C14D-16GFX-EOL

>>105715028
that's magic bdie ram

>>105715028
>or is it even possible?
yes
you have B-die
with your sticks you can get even better than that
3200 14-14-14 at 1.35 V is a good bin

here >>105712405

ah sorry, it is 2x 16GB sticks not 8 GB

>>105715028
>>105715043
that's what was recommended back than as the must-have ram
thanks will check the guide
<3

>>105715076
>thanks will check the guide
good luck anon
i would aim for maybe 3733 14-15-15 or 3800 14-15-15 at 1.5-1.55 V
ryzen 3000 in particular loves high memory/fabric speed because it speeds up core-to-core communication as well

>>105714986
>it's pretty impressive that they nailed it down that close
They wouldn't be called G[punkt]skill if they didn't know how to nail some things kek.
>>105715002
I'll see.
Maybe I'll just grab some 2nd hand b-die off of e-gay when DDR6 starts rolling in and DDR4 becomes as cheap as DDR3 is now.

>>105708440
I got a 5700x3D off aliexpress for $200 CAD a few months ago. Maybe the price fluctuates but 400 doesn't seem right

>>105715169
>Maybe I'll just grab some 2nd hand b-die off of e-gay when DDR6 starts rolling in and DDR4 becomes as cheap as DDR3 is now.
you can't run 128 GB of B-die on AM4
64 GB is the max (and that will be painful to stabilize)

>>105715188
>you can't run 128 GB of B-die on AM4
I am aware, but thanks for the heads-up.
>64 GB is the max (and that will be painful to stabilize)
Because of memtesting?

>>105715197
>Because of memtesting?
no, because B-die is temperature sensitive and hard on the memory controller
running B-die in quad rank is a patience test (according to buildzoid, i've never done it myself)

>>105705170
>>105706712
>>105708440
>>105715177
I got my 5800X3D for the same price brand spanking new off of mindfactory back when it was still available ...

Still hoping for a 5950X3D for next year as AMDs "final" AM4 CPU for a round ten years of AM4 CPU releases. That would be GOATed.

I use a similar slower one in a nas I built

>>105704453
>Expect lag in GTA 6, but playable.
lol nigger that shit is meant to run on a laptop 4800H equivalent CPU

>>105708700
>GTA IV
Kek all that autismo mem tuning and I still have a better experience plopping the disc into my Xbox 360.
PC port sucks ASS.

>>105704394 (OP)
it's still more than fine for building code and running VMs. you likely need less.

>>105712124
clockspeed means precisely jack shit since zen existed
watched motherfuckers overclocking the 1600 to 4ghz to gain 3fps in games and 3 points in niggerslopmark kinda funny
mine proceeded to die slowly after 4.5 years on stock settings

>>105715781
I'm running my 5900x at lower TDP, like 120W. still pretty fast and cool af

>>105714388
when I saw my shit going 1.4v and reading from some AMD employee that ryzen is safe for 1.5v I found it insane however it's the amperage that really matters, which is why allcore load drops it to fucking 1.1v... so they say, at least

>>105714493
>Wouldn't that fuck with the io die of the CPU?
probably not , in ddr4 the io voltage is very low.

>>105704394 (OP)
it better be. I paid a fucking fortune for that fucker during coof.

>>105714423
>yeah, that's B-die and those kits have all been discontinued long ago, sadly
I remember B-die being memed to shit
>i have micron rev E so i could put in 2 V if i feel like i need some more excitement in my life
kek what the fuck
what's hynix shit like cause I have only ever got that from 2 ddr4 kits I bought over the years, I seem to gather MFR was a bit of a legend of DDR3?

>>105715865
>I remember B-die being memed to shit
for good reason
go here
https://www.techpowerup.com/dram-latency
put in 3600 14-14-14
then change data rate to 6000
do you know of any DDR5 which is even close to approaching those latency numbers?
>kek what the fuck
https://www.reddit.com/r/overclocking/comments/ex2q40/cheap_reve_vs_21v
>what's hynix shit like cause I have only ever got that from 2 ddr4 kits I bought over the years
as with any other manufacturer it ranges from great to absolute trash
for DDR5 hynix is best, the only sensible option at the moment
for DDR4 hynix DJR is good, i may be a bit biased but i would put it in third place behind micron rev E and samsung B-die
the issue is DJR does really bad primaries, and the secondaries don't go any lower than rev E (except tRFC)
CJR is also OK i guess
i had an AFR kit once and it was fucking terrible

>>105715745
>he thinks 480p lowest settings is a better experience
runs great with dxvk, you must be retarded

>>105715865
>>105715938
like this
never ever happening on DDR5
it's only gonna get worse from now on

>>105704394 (OP)
It's fine, I'm using a 5900X (kinda wish I got the 5950X instead though) and I have no plan on upgrading until the last generation of DDR5 CPUs, but I have DDR autism and want that shit as stable as possible before I put it in my system. If you're worried about gaming just get whatever X3D CPU is current and you should be fine.

Means nothing and isn't necessary

>>105715953
its been getting worse pretty well since the debut of DDR memory, you used to be able to buy cl3 kits of ddr2 800, though I will say the changes to ddr5 that allow for more simultaneous accesses does drop the average latency a fair bit, even if the best case is worse.

>>105716398
>though I will say the changes to ddr5 that allow for more simultaneous accesses does drop the average latency a fair bit
this is just an increase in bank groups, you get the same improvement going from single rank DDR4 to dual rank DDR4

>>105708700
>5700X non 3D is the best choice on AM4
Based retard spouting nonsense

>>105704411
Bullshit. I was using a 3900X for 4K gaming and getting at or slightly over 100fps on medium-high settings. Recently upgraded to a 9800X3D and gained 20-50fps across the various games I play. Ram also played a role, went from DDR4 2x16 3600mhz to DDR5 2x16 6400mhz.

>>105704421
No. CPU usage remains the same because most of the work for games is done GPU side when it comes to different resolutions. 1080p vs 4K will see minor differences in CPU load but massive differences in GPU load.

>>105716454
its also the move from 2x 64bit channels to 4 x 32bit, i wouldnt under sell the increase in bank groups either, makes single rank dimms much faster and makes dual rank dimms as good as quad rank.

I have a 5900X on a Crosshair VIII Dark Hero mobo with 32GB of Samsung B-Die RAM chips (Patriot Viper Steel) and I could never get the advertised 4400mhz. No matter how much I try.

>>105716660
I just got 3600mhz and like never ever gave any fucks about anything related to timing and shit, I just like...use my PC, and it's fine.

>>105716074
what's your build around the 5900x?

>>105716674
Well, I'm a tinkerer and not a pedophile. So, I'm reading this thread and testing recommendations and suggestions.

>>105716687
MSI X570-A PRO
G.Skill 4x16GB 3200MHz CL14 b-die (can't remember the exact model)
Sapphire RX 6900 XT Nitro+ SE
Corsair MP600 500GB system disk (pcie4)
Samsung 990 EVO Plus 4TB extra disk (pcie3)
3x4TB Samsung 870 EVO (RAIDZ1)
3x16TB WD Ultrastar HC555 (RAIDZ1)

>>105716660
you have to run zen3 in 2:1 mode to go above 3600-4000mhz, to which there is literally no point in doing so unless you are pushing >5000mt/s which very few motherboards can actually do, also getting dual rank sticks to work at higher speeds requires screwing around with the ODT values, takes fucking forever.

checking your motherboard, there is a total of 1 4x8gb kit validated at 4000mhz, and there are no 2x16gb kits rated validated above 3600 , so I dunno what you were expecting, looks like the board is t-topology, which is rare on am4 boards, and means it works better with 4 single rank sticks instead of 2 dual rank sticks.

>>105716709
you seem insane tho

>>105716781
Forgot
Noctua NH-D15 CH.BK
Fractal Design Define 7 Solid
3x Noctua NF-A14 2000RPM

>>105716784
4 dual RAM 8GB sticks. FCLK 2000mhz 1.395v on SOC timing is 15, 15, 15, 15 Auto everything else.

>>105716805
yeah quad rank is pretty rough on the memory controller significantly limits the maximum frequency, afaik on ryzen 3000/5000 , getting dual rank above 4000-4266 is difficult, and getting quad rank above 3800 is nearly impossible.

>>105716842
Got it thanks for the info

>>105716800
probably, just not a pedophile like you

>>105716871
normal people don't keep mentioning it in every single comment they make. something you want to tell us anon?

>>105716893
what are you talking about? I only mentioned my specs once and there's nothing normal about you being a pedophile, dunce faggot.

>>105716936
https://en.wikipedia.org/wiki/Accusation_in_a_mirror

>>105716784
anon, can you kindly explain what is meant by 1:1 and 2:1 and how to make the necessary changes? thanks

>>105716957
>no u
you can't be serious. you fucking girly-mouthed little bitch

>>105716982
seek help anon

>>105716958
1:1 mode is where the dram ,memory controller clock and data fabric clock are equal, technically 1:1:1
eg. 3600mt/s(1800mhz) = uclk 1800mhz = fclk 1800mhz
2:1 mode , is where the memory controller speed is halved,
eg 3600mt/s(1800mhz) = uclk 900mhz = fclk (asynchronous)

when in 2:1 mode with the uclk halved and the fclk is asynchronous, a significant latency penalty is incured ~15 ns , which is not worth it at all, as it requires extreme ddr4 speeds to overcome ( >5000mt/s)

>>105716958
the setting in the bios is called UCLK DIV1 MODE if I recall correctly.

>>105704421
no, it stays the same. like if the cpu is maxed out doing 50fps at 1080p, then it'll just be maxed out at 50fps at 1440p as well (assuming the gpu doesn't become the bottleneck and assuming only the resolution changed)
the only time the cpu load will go down is if you increase the resolution enough to max out the gpu, then the cpu load will go down as it's now the gpu which can't keep up

>>105705170
i think he's talking about the i7-2600 meme, not the ryzen 5 2600

>>105706462
>Moore's law died a largely-unnoticed and expected death sometime around 2005, so new computers are no longer many times more powerful than used ones.
moore's law was never about performance (directly), it was about transistor count

I just bought a 5800XT for $120 and I'm feeling smug about it
Huge upgrade from my 1700X

>>105705716
Those fuckers need to keep making the 5800x 3D.

>>105715745
works great in linux

>>105704394 (OP)
I have it since launch
still kicking ass but my only regret is not going for the integrated graphics version so I cant play around with GPU passthrough
and also I don't do the kind of work that require that much parallel CPU raw power that I used to do anymore so its overpowered and becoming irrelevant

>>105713566
>the 5700g is dogshit (2600 tier gaming performance)
Wrong, 5700G is considerably faster than 2600 (~40%) for gaming. I can tell you since I made that exact same jump. I personally use the 5700G for productivity and I love it

>>105717360
you can do gpu passthrough with one gpu, i've done it before. however it does mean you need to close your linux gui and completely unload the gpu driver before you can start the vm (automated with scripts of course). it's certainly more convenient with two gpus and i use two now

Still rocking AM4 without any plans of upgrade
>Loonix pornstation: 5700G + iGPU + 3060 12G + RX570
>Windows gayming machine: 5600X + 6700XT

>>105717360
You aren't missing that much, GPU passthrough can be a pita

Is DRAM calculator still recommended for memory OC'ing with the 5900x? that software hasn't been updated on over 5 years it seems.

>>105717417
its still useful for older memory like samsun B-die micron rev E, but completely useless for newer ics like micro 16Gbit revB and ect.

>>105717360
>not going for the integrated graphics version
They had an iGPU version of the 5950x?

>>105717384
mind sharing ur recommended setup? what linux specifically

>>105717584
Gotta be lower tier but I thought I would need all dem cores

>>105717619
What's the highest they went back then with the APUs?

>>105717403
Damn that sucks. I used to dream of GPU enabled airgapped VM fortress

>>105717305
why vsync'd at 72 fps?
you aren't running a 10 year old 1080p tn panel monitor with turboautismo overclocking are you?

>>105704411
Fortunately, new games suck.

>>105717584
not only is an iGPU version of the 5950x not a thing, but the only ryzen 5000 apus (5600g, 5700g) are all monolithic and 16 mb of cache compared to their regular counterparts and actually perform WAAAY worse
the 5600g is on average slower than a 3600
so yeah no idea what this nigger is smoking, he regrets not buying one of those crusty ass cpus?

>>105717305
lol

>>105717286
5000 are kind of a secret - they are higher quality than they were supposed to be. ie they are better than consumer grade.

I'm a lot more quality minded, now that I'll be putting my car in the shop for the 3rd time this year.

>>105717943
the fuck does that even mean

>>105717828
Don't we all.

Have you considered installing Jesus OS? Jesus OS is not an operating system but an ethos. Basically, you setup a separate computer for each main task you do. There's more to it, and I think there's a lot of potential for this, especially with 16-bit machines getting more popular.

If you've ever used WordStar you know that the myth of improvement is only true in certain areas. Yeah, games look way more realistic. Yeah, desktop publishing is much more accurate now. But some stuff is worse. Like did you know DOS had way nicer note taking TSR options than on modern machines?

>>105717988
Failure rates are supposed to be higher on consumer equipment. But AMD just had really good yields with 5000.

And to be clear, I will never again get any Intel consumer cpu, no matter how great the deals are right now etc. Their server stuff is ok.

>>105718038
apart from the recent intel 13 and 14 series, pretty much no series of cpus has ever had high failure rates or anything like that, plenty of people still rocking second and fourth gen intels, early ryzens, even amd fx cpus, all fine, it's one of the most reliable components you can get, none of what you are saying makes any sense, cpus just don't die.

>>105718062
Intel is the great genius at passing off cpu faults as the fault of other components, or users, or software. Anyone but Intel.

>>105705083
Are you sure because my 5800X sure as fuck couldn't emulate Gran Turismo PS3

>>105715177
>I got a 5700x3D off aliexpress
What are the security implications

>>105717274
I think he meant denard scaling but he's still half right as CPU performance hasn't increased as much with just density scaling alone as it had in the past with clock & density scaling.

>>105713522
>past 4 gbz
Ryzen (even zen4/5) in general is so fucking weird. It feels like a really old phenom CPU with outdated IO and busses that's turbocharged with the latest bleeding edge cores making it really imbalanced.
Because on intel CPUs, performance scales nearly linearly with clock speed.

Still good for another 5 years

>>105713744
I'm savvy enough but would always buy x3d since the silicon lottery is a pretty shit game to play

>>105714840
>Dropped since 11th gen
>Only 10% of CPUs on steam support it
It's alright, intel has got you covered

>>105715028
>2x16gb bdie
That's best in slot ram
3600cl14 is probably doable at 1.45-1.5v as long as you have the cooling for it

>>105718047
Fucking E-marketing cores man... Their upcoming raptorlake refresh 12 p-core finally gets rid of it but it still won't have avx512

>>105717613
i think i used this guide;
https://github.com/martinopiaggi/Single-GPU-Passthrough-for-Dummies
with some adjustments as i went
tested in arch and gentoo. there's several guides out there so it shouldn't be difficult to find answers. i was using an RX560 (now i'm using an RX580 + RX6600, it has been a while since i did single gpu passthrough)

>>105717866
when i made that webm, yes, though the 72hz was mainly for videos

>>105718984
tx
I might be looking into single gpu as well since my hardware apparently has bad iommu grouping and cost me zero to try

>>105718062
>>105717943
>>105718038

Failure rate is the wrong word to use here, What he means is the quality of the chips was higher than expected, aka they draw less power and clock higher than expected.

>>105718062
Intel 10th gen onwards (except 12th gen) have had increased failure rate and it coincides with intel starting to program their cpus with 1.5-1.7v stock voltages

>>105718282
The Game I'm playing on RPCS3 gets a constant 60fps during actual gameplay, so it works for me. Not all games are equal and some will perform worse on any CPU just because it is emulation.

>>105718984
How was the performance compared by bare metal tho

>>105715222
So just rev-e and be done with it?

>>105719841
i only use it for solidworks, so i didn't run any benchmarks. didn't feel any different at least

>>105720324
I would recommend micron rev B over rev E, easier to get and comes in 32gb dimms does better timings and higher speeds , only trade off is the higher trfc, which wont matter anyway since quad rank will eat it.

Im on a 4k60Hz screen. My 5800X3D will be fine for probably 10 other years. Transistors spent on CPUs IPC is diminishing returns. On GPUs it works better, especially on high res.

>>105718592
yes, i agree that cpu performance hasn't gone up much in the last 20 years relative to the years prior, all i'm saying is that that's not moore's law

>>105717417
no
stop trying to take shortcuts, it will only waste more of your time
>>105717469
it's useless for rev E and overly conservative for B-die (but it will probably "work" for B-die, because just like hynix DDR5 it's popular enough that people have figured out baseline timings which work on almost every kit)
>>105720324
>>105720353
16gbit rev B is what you want
the best dies are in 16GB ballistix sticks, those are pretty cheap on ebay but there's a chance you'll get dual rank rev E sticks instead

>>105715714
>lol nigger that shit is meant to run on a laptop 4800H equivalent CPU
consoles are way more optimized for gaming than windows, a 4800H laptop won't run gta 6 as good as a console. The only OS who is truly optimized for gaming is steam OS.

>>105717866
>why vsync'd at 72 fps?
not him, but GTA 4 is one of those games where the physics get funny if you have too high fps

I have a 5900X and it's ok but I bet I can at least get something noticeably better
I just don't want to spend $500 purely on a CPU upgrade (because I need a mobo and RAM)

>>105717943

>>105717943
this trash post sounds so "the wife". Cringe, retarded and try-hard.

>>105718678
4x8GB B-die achieving cl15 at 3800mhz stable

>>105718692
>avx512
Wtf does the average person need this for?

Anybody extreme OC'd a 5900x?

How am I doing guys?

>>105718984
Dunno why, but I can't get this guide to work for my 7800xt.

>>105713098
*Unless it's a Bethesda engine which does most of the shadows on the CPU.

>>105721498
1. nah it really doesn't apart from the final mission which was fixed in the definitive edition or with silent patch anyways
2. that starts at like 30 fps
3. the webm is also encoded at 72 fps, why would he switch his recording settings for a single webm?

>>105722845
those are some funky XMP timings but this doesn't tell us anything
if you are running tRRDS 9, tRRDL 11 you need to fix that ASAP

>>105722713
For fine wining and free a performance speedup of course
Haswell and skylake have aged terribly well with avx2 support in contrast to sandy bridge and ivy bridge

>>105722845
I don't know, all I see is you're running 3800 which is pretty good
>>105723354
the physics are iffy even at 60 fps without fusionfix but it doesn't fix npc pathfinding

>>105721498
Skyrim as well

>>105722845
>>105723682
Can you explain why and what I should run them at? also what info should I share that'll be useful and helpful?

>>105714950
>Cause games are going to use it soon.
Feels good to be a retro gamer. Don't have to bother then.

>>105714950
>Cause games are going to use it soon.
games still routinely ship with non-avx2 exes for that <20% of the population still using 10 year old cpus, avx512 isn't becoming mandatory anytime soon, especially considering that plenty of recently released processors don't support it, you are smoking crack.

>>105728159
most games still don't even require/make use of avx 1
...and that's been available since the first/second gen intel core processors, maybe even earlier if i'm not mistaken

>>105726936
https://zentimings.com

rate timings
>inb4 stop using 4 sticks
never

rate timings
>inb4 stop using 4 sticks
never

>>105719770
Maybe the game you're emulating is sub-PS3 quality

>>105704394 (OP)
Plenty good. Will probably be fine for casual use till the next decade.

>>105728323
pretty good

some suggestions:
turn off power down mode
lower ProcOdt and higher ClkDrvStr/AddrCmdDrvStr might give you better results with 4 sticks (see my settings: >>105711290)
can do you tRRDS 4 and tFAW 16?
tRDRDSD/tRDRDDD could be 4
tWRWRSD/tWRWRDD could be 6

You know you can get one of those ghetto binned AM5 APUs sans IGP and increase fclk, OC memory better than normal desktop CPUs, and tighten timings, and the decreased L3 size basically doesn't matter at all
They're great. Better than great, entirely slept on. 5mins of tinkering and they're incredible.

>>105729021
AM5 RAM overclocking is nice because you don't have to buy an expensive motherboard and clockmaxx like on intel (6400 1:1 gets you basically the same performance as 8000 2:1)
you also have buildzoid easy timings which is a very good starting point

>>105704394 (OP)
It's fine if you already have one or are stuck with an AM4 board.

>>105704394 (OP)
You kind of fucked with DDR4 just doubled in price.

>>105728909
are mine any good? bought this computer prebuilt, don't know much about this stuff

>>105729422
XMP isn't even enabled

>>105729422
>worried about timings
>4400 MT/s RAM is running at 2133 MT/s
lol you you're about to get a pretty big performance boost when you finally enable XMP. How long have you been running that system?

>>105728909
>lower ProcOdt and higher ClkDrvStr/AddrCmdDrvStr might give you better results with 4 sticks
thanks for the suggestions (actually not really I will now waste another bijillion hours on this shit instead of actually using my pc i'm fucked) that seems to have boosted stability a bunch, I was even able to boot 3133 but that was highly unstable even with the previous timings
>can do you tRRDS 4 and tFAW 16 tRDRDSD/tRDRDDD could be 4 tWRWRSD/tWRWRDD could be 6
I guess it boots, it didn't immediately acck itself at 3066 mhz, in case it's unstable (it likely is), do I tighten timings and stick to 3000 or attempt to run 3066 or maybe even 3133 while loosening some (and if so, which ones?)
you seem to be one of the few with the idea that lower timings > higher speed so idk
oh fuck I was testing some settings and forgot vSoC at 900mV, i wonder if it will still run at 800mV like it did before

>>105729612
>actually not really I will now waste another bijillion hours on this shit instead of actually using my pc i'm fucked
yep
>I was even able to boot 3133 but that was highly unstable even with the previous timings
try 40/40 or 40/20 instead of 30/30
way better on my setup but different motherboard manufacturer, topology, kits, etc
YMMV
>and if so, which ones?
obviously primaries if you are trying for substantially higher speeds, general rule is +1 for every 200 MHz (gives equal absolute latency)
tRFC might need to be loosened (seems high already)
tRAS/tRC might need to be loosened
>you seem to be one of the few with the idea that lower timings > higher speed so idk
i mean, really you want both
more clock speed gives you more bandwidth AND speeds up both memory controller and infinity fabric
but that's what i found when testing 3200 12-16-12 vs 3733 14-18-14, 3200C12 was surprisingly competitive
most of your gains seem to come from timings
>oh fuck I was testing some settings and forgot vSoC at 900mV, i wonder if it will still run at 800mV like it did before
do you really need to run the SOC at such low voltage? personally i think saving a few watts of idle power on a desktop is stupid
but it's your rig, your rules
you have a budget B450 board so ~3200 might be the end of the road with 4 sticks, and you might need more VDDP to get there
VDDP is clipped to VSOC - 50mv, so you need at least 1.05 V SOC for 1.0 V VDDP

Used DRAM calculator when I set up this PC 5 years ago. I'm aware that's no longer recommended, but I've never had any issues. Any reason I should change any of this? Dual stick 16gbs

>>105729719
>Any reason I should change any of this?
not really

tRCDWR 8
tRRDS 4
tFAW 16
tWR 16
tRTP 8
tRDRDSD/tRDRDDD 4
tWRWRSD/tWRWRDD 6

should all work on micron if you feel like tinkering

>>105729769
Not really. I rather doubt I'd get much performance boost anymore. Nice to know the calculator worked for me.

>>105729708
>try 40/40 or 40/20 instead of 30/30
am i not doing 30/20 at the moment?
>general rule is +1 for every 200 MHz (gives equal absolute latency)
well if I have to increase primaries by one to have the same performance compared to sticking to a 200 mhz lower speed what's the point tho
>tRFC (seems high already)
entire mobo has an aneurysm and will require a cmos reset if it's any lower yeah
>do you really need to run the SOC at such low voltage?
no but I like that it werks and 13w idle power draw is comfy, it's like i'm not running a chiplet cpu at all
>you have a budget B450 board so ~3200 might be the end of the road with 4 sticks
I'd be perfectly happy with 3200 already, but I had trouble getting that to work even at xmp, this retarded e-waste tier hynix/micron combination is definitely the limiting factor rather than the mobo

>>105728323
Why stop using 4 DIMMs?

>>105729806
a wise choice
>>105729816
>am i not doing 30/20 at the moment?
picrel
>well if I have to increase primaries by one to have the same performance compared to sticking to a 200 mhz lower speed what's the point tho
that's not how timings work, 3000C14 ~= 3200C15 ~= 3600C17
https://www.techpowerup.com/dram-latency
that's why i used 3200 12-16-12 and 3733 14-18-14, similar absolute latency (which means i didn't have to mess with VDIMM either)
>entire mobo has an aneurysm and will require a cmos reset if it's any lower yeah
then don't touch it
i do tRFC*(new_speed/old_speed)
so if i'm going from 3733 to 4133 i do 540*(4133/3733) ~= 600
>I'd be perfectly happy with 3200 already
start by seeing if you can boot 3200 with loose timings, 3200 22-22-22 with auto subtimings
mess with the drive strengths until you find the most stable setup

>>105723682
tRRDS = 4
tRRDL = 6

>>105704394 (OP)
>4x8GB 3600mhz 18 CAS
>5800X
>MSI B550 Pro-A

What can I do and can I somehow get it to 3733 that is stable and not atrocious latency? Never overclocked AMD or ram ever in my life

>>105729865
>hat's not how timings work, 3000C14 ~= 3200C15
I'm confused... that's literally what i said, if i have to increase primaries to C15 to get 3200 stable, and you are telling me the performance is about equal, what do i gain from doing that
>start by seeing if you can boot 3200 with loose timings
xmp is already quite loose tho
>mess with the drive strengths until you find the most stable setup
I honestly doubt that could be enough to bring me to a stable 3200 with decent timings, but even if it can, I'd be spending hours if not days pretty much doing it all over... for what? 3200cl15 which is the same as 3000cl14?
and btw, doesn't lower timing/lower frequency on average require more vSoC/vdimm than higher timing/higher frequency? because I suspect that that's the case, so I'd loose the little bonus of running 800mV SoC

>>105729990
>>105730131
what more can I do with these settings, anons? or is this pretty good as it is?

Realistically cpus from 2009 are still good enough for a lot. Mostly video game addicts need to buy new crap every couple years.

>>105730257
well yea, this is all about gaming and gaming emulation.

>>105730197
you get more bandwidth, latency can be the same , but because the memory is operating at a higher speed more data can be r/w during the same period of time.

>>105730195
post zentimings
this thread has accumulated a lot of useful information, go through some of the earlier posts

>>105730197
>I'm confused... that's literally what i said
yeah just making sure, a lot of people don't understand that timings are measured in clock cycles
>and you are telling me the performance is about equal
performance is better
you're overclocking the memory controller and IF as well
going from 3200 with XMP timings to 3200 with tight timings gets me 14%
going from 3200 to 3733 with equal absolute timings gets me another 7%
(only one test, but a real world scenario)
but you're already in diminishing returns territory, if you're happy with the performance already stick to 3000C15
>I honestly doubt that could be enough to bring me to a stable 3200 with decent timings
you'd be surprised
with default drive strengths i can't run 3600 XMP stable on this board
with "fixed" drive strengths booting and stress testing up to 4066 is no problem (infinity fabric completely gives up after that point)
>doesn't lower timing/lower frequency on average require more vSoC/vdimm than higher timing/higher frequency?
VDIMM yes
VSOC no, you need more voltage specifically to run higher UCLK and FCLK
for low idle power you want low memory frequency with tight timings (what you're doing right now)

i think you already have something good figured out for your usecase (low idle power and weird memory setup) so don't worry about it too much

>>105730223
you're close to the limit so any change you make is for aesthetic/tinkering purposes

that's B-die so tRFC could be a lot lower
tCL will go lower with more VDIMM
tRC might go lower
tRRDL could be 4
tWTRL could be 8
SCLs might do 2/2 or 3/3
if those are 4 DIMMs RDRDSD/RDRDDD should be 4 and WRWRSD/WRWRDD should be 6

>>105730327

>>105730327
I put tRFC to 60 and it won't post. Just runs with 1F q-code error. What should it be set to?

>>105730327
>but you're already in diminishing returns territory, if you're happy with the performance already stick to 3000C15
it's actually 3000C14
but yeah, thanks for all the advice, I might try playing with the drive strengths to maybe squeeze 3066/3133 or at least improve stability in case these new timings turn out to be unstable
btw, is it normal that my latency in memory benchmark stays around 75 ns no matter what I do? pretty much no decrease at all compared to cl16-20-20-38 xmp

>>105729454
>>105729527
>they took the bait
it was a meme config that was sent to buildzoid, no memory kit comes with such tight optimized timings out of the factory on the basic 2133 profile, should've been obvious

>>105730327
How much more voltage is recommended? 1.45 or 1.5?

These are my new timings after following the suggestions in this post:
>>105730327

>>105730986

>>105704837

Same here too. My 5900x will be more than enough for my needs for a long time. No pint upgrading to until AM6.

>>105730772
>>105730999
see pic >>105714956

A520M DS3H V2
Ryzen 5 5500
16gb DDR4 3200Mhz
Nvme 500go
1To HDD
RX 6600 8go VRAM
AIWA Monitor 1080p 100Hz FreeSync

What do, bros?

>>105731193
enjoy your computer
overclock your ram if you're bored

>>105731160
maybe it's bcuz I'm a bit sleepy, but, I don't understand that chart.

>>105731249
trfc

>>105731261
i think i understand thanks.

i love my 5950x!

Anyone here try the underclocking meme? I went ahead and set that little setting in the UEFI to -25, and I don't feel like there's any difference. I think my highs aren't as high but I haven't actually done anything strenuous yet.
Maybe I should go to -30.

>>105731385
that is not underclocking, it's undervolting

>>105731538
Fuck I keep doing that lately and I'm not sure why. But yeah I meant undervolting.

>>105731385
I've kept mine at stock, don't think undervolting is worth

>>105731758
That's what I've been doing for the past 4 or so year since I got my 5900x near launch. So far it seems like it doesn't do much, but it doesn't seem to hurt anything either. Maybe it saves me some energy.

>>105731385
Running mine at 825mv @ -42 curve
Temps went from 75° to 40° \ 90w to 55w @ 100% load
Performance wise it only fucks benchmark scores by 20% but for gaming it doesn't have any real impact on my 1080p 60fps budget system

>>105717305
dude you missed like 4 jews there

>>105731385
undervolting only really helps when you are hitting the power or thermal limits, the 5950/5900x both have pretty high boosts and are pretty efficient out of the box, if you enable PBO and have a semi decent cooler, you won't be hitting either of those limits, so the effect of undervolting is diminished, If you have a really good chip, it can help achieve >5ghz boost under lightly threaded workloads with the +200mhz pbo frequency limit from what I've seen, but not all chips are capable of doing that.

>>105731855
it's actually a complete coincidence that they were jews, i only posted that one because i felt i did decent enough with the turns, especially that first one which was really tight

>>105704394 (OP)
It doesn't work for all games, but I got a massive boost to my minimums by disabling SMT on my 5950x. Other than a small additional stutter right when opening programs, the rest of my desktop experience is basically the same compared to having SMT on.

>>105704411
>and above
shouldn't that be the opposite?

>>105731318
>core 0 2nd preferred core
lucky bastard

>>105704411
What games are you playing that benefit from anything more powerful than a 5950X?

>>105705716
Is that AI generated?
The pixels are not uniform.

>>105732514
seems legit.
>pixels are not uniform
probably due to fractional scaling.

>>105704394 (OP)
bruh that's still a recent cpu bruuuh

i5 2500k is more than enough to play any game

CPU technology stagnated with sandy bridge and now GPU technology is doing the same thing, anything past a 2500k+3060 system is a placebo

>>105732545
Ah, that might be the case.
It ruins an otherwise very beautiful pixel art for me.

>>105729054
Very true

>>105731261
Managed to get my tRFC down to 342, but there's tRFC 2 and tRFC 4. do I leave those on auto or do I have to dial them in, too?

>>105731837
there's no way that's stable
>Performance wise it only fucks benchmark scores by 20%
you should be seeing a performance increase not decrease, a 20% reduction means you are getting an insane amount of clock stretching, so your cpu is insanely unstable and a miracle it works at all, what the fuck are you doing??? literally just set a lower power limit nigga

>>105732394
isn't the whole point of preferred cores that these get prioritized by the windows scheduler? why would a core 0 preferred core be better than let's say a core 6 preferred core?

>>105730804
>is it normal that my latency in memory benchmark stays around 75 ns no matter what I do? pretty much no decrease at all compared to cl16-20-20-38 xmp
no, that's not normal
but the AIDA latency can be affected by lots of factors like your windows configuration (virtualization, core isolation, etc) and it's not represenatitive of real world performance because it doesn't stress back-to-back timings like tRC, tRRDS/tRRDL, tRDRDSCL, etc
>>105730820
lol good one
honestly i just saw prebuilt and 2133 and noped out of there
2133 9-9-9 is pretty tight but the memory controller and IF are so slow at 2133 that it's still useless
>>105730898
1.5 is fine
if you have temperature sensors on your DIMMs they should show up in hwinfo, you want to stay below ~50 C on B-die for maximum stability

this was a good thread

>>105734435
I found stability at CL14 at 1.48v. thanks again. what do you recommend for testing RAM stability?

>>105734544
hope you see this before the thread dies

https://github.com/CoolCmd/TestMem5
ABSOLUT or anta777 extreme profile is what i use
https://www.ocbase.com
paid if you need to run it for >1 hour, but you can combine this with the other tests
CPU+RAM test is good for testing stability at 1900 FCLK
https://www.karhusoftware.com/ramtest/#introduction
paid, but i've heard this is the best for DDR5
https://www.mersenne.org/download
prime95 large FFTs is good for catching memory controller instability (too low VSOC)

>>105734732
how much time do you run absolut for

>>105734750
like 10-30 minutes while i'm testing different settings
once i think i'm done i run a test overnight or while at work (so ~8 hours)