What went wrong?

Nothing, the non-HT ones ran cool and quiet compared to amd kshit housefire.
With the same cooler i could run P4 without a fan, meanwhile the Barton was idling at 60C with a fan somehow foreshadowing the upcoming decade of pure trash from amd until Ryzen was born

>>105783669 (OP)

>>105783669 (OP)
the oldfags of the early days when frequency scaling was trivial thought that the x86 uArch-fags were a burden for the company so their new shinny arch was optimized for higher clock at the cost of IPC. And that was a stopgap for the Itanium future...
>intel goes turboretard and tries kill x86 and replace it with a GPU-like piece of shit (VLIW).

>>105783669 (OP)
put the cart ahead of the horse, many of the design features of it are actually quite forward thinking, however certain technologies were not up to the task to support such a design, namely the process node and the relative infancy of techniques like power gating.

Everybody rags on the long pipeline, but extending the pipeline to 20 stages has persisted to this day, and will likely be extended again as transistor budgets allow for more advanced branch prediction.

>>105783669 (OP)
Too many house fires is my guess.

>>105783897
You can't blame them for trying, if this shit worked they would have been kings for 20 years powering every high end system.

But yeah this anon got it, Intel R&D was focused on the worst project that Intel ever started; as it was a trillion dollar plan.

>>105783669 (OP)
Putting raw clockspeed over everything else. They realized the error of their ways with the Pentium M, somewhat, but it took a while for them to use its design philosophy on desktop.

mage a longer bibeline :DDD xDDDD

>>105783669 (OP)
certainly was better than whatever amd was doing at the time

>>105783669 (OP)
long pipelines. think of pipelines like having a road that is a mile long vs a road that is two miles long. intel went with really long roads. they couldn't get clock rates high enough.
while amd had shorter roads. lower speed, but shorter roads meant amd got things done faster.

>>105783897
KWAB

The hilarious thing is that AMD tried to replicate Netburst with Bulldozer, and failed just like Intel did. It was worse, though, because they didn't have the marketshare and respect that Intel did to weather the storm.

>>105784382
Their stocks never recovered.

>>105784524
The worst part is that they did it after beating Intel with Athlon even if AMD had worse nodes than intel. they began working on Bulldozer around 2003-2004 iirc and they failed to launch it in time, around 2008, Phenom II-III were kinda improvised. Meanwhile the small cats arch family were a far better foundation and they could delivered it as a product in less time, after some improvements they "renamed" it as Zen.

>>105784540
Fucking kek.

>>105784576
Did AMD ever state why they decided to copy Intel Netburst even though it was AMD was better with athlon / athlon 64?

>>105784078
Could it ever be made to work?

>>105784714
No.

>>105784682
We don't even know why Lisa Su didn't allow a new high end cpu launch between 2014 and 2017, excavator was decent and Puma was very similar to Zen 1.

>>105784714
yes, But it would require recompilation and re optimization of everything, and a very good compiler.

>>105783669 (OP)
Too many pipelines

>>105784682
they didn't copy netburst, they attempted a reverse smt design, but it wasn't very good so they pushed the frequency up.
>>105784743
that one is easy, amd's financials were horrendous at the time, and taping out more big chips would take resources away from building a new core(zen), I remember before zen launched amd's share price was like 7 dollars,

>>105784774
it was a $1 a share in 2015

>>105783669 (OP)
They put Intel inside instead of an actual processor.

>>105783669 (OP)
YOTD, faaggoyt.. NOT! FUCK YOU! GET FUCKED HA

>>105784714
Could they make it compute at a levels that it could push out all of the RISC processors and be force all of the unix vendors to bend the knee in the face of Intel's ability to produce at scale? Yes

Could it do this in real life with branching code? No, it would require an omnipotant compiler. I am curious if Intel or HP is dusting off old Itaniums and feeding code through it through some sort of AI cluster to compile to dynamically as a test to see if they could get it to work properly by throwing astronomical amounts of compute at it.

>>105784714
if they had done what transmeta did , it probably couldve worked.

>>105783917
ipc is actually better on all subsequent revisions of netburst, only the first one willamete was lower ipc than its predecessor, the misconception peristed because older software ran better on northwood than prescott

but if software was compiled with optimizations turned on prescott is faster per clock than northwood, which should be no surprise because the branch predictor is absolutely massive on prescott, so its more like same ipc , higher clocks and oh boy do they clock high.

90nm prescott does 4-4.2ghz easily, the catch being that 90nm is a pile of garbage, and it consumes insane amounts of power at those speeds, 65nm netburst is actually pretty good , they clock to 4.5 to 5ghz wihout insane voltage , but again power consumption skyrockets past 4.5ghz.

and finally , the platform itself really hurt netburst, having to use a super slow aging fsb and cache structure made the latency insane, inorder for netburst to perform it needs huge amounts of bandwidth too. on the one i tested at 266fsb the latency was something insane over 200ns in sane cases, pushing the fsb up to 400 reduced it to like 120ns,

little known fact is that intel actually fixed netburst, there is a chip known as tulsa , which has a totally revised cache structure and is actually the first intel chip to be a true dual core and have an uncore, carrying a hefty 16mb of l3 cache, compensates for the slow bus, and it results in a up to a +70% uplift according to slides from a hotchips presentation on the cpu.

>>105783669 (OP)
high GHz CPUs can't melt steel beams

>>105784295
sure

>>105787006
AMD's lowest end Athlon chip was beating Intel's EE housefire chip lol. I don't think Bulldozer was ever that bad.

>>105787006
>I don't think Bulldozer was ever that bad.
It wasn't, the speed was ok and it was cheaper.

>>105787006
the main issue for intel and games at the time was the aging fsb based platform, amd had already moved the memory controller on-die, i tested a cedarmill once and just changing the cas latency of the memory down 1 tick increased the 3d mark score by 10% which is ridiculous, horribly memory bottlenecked.

>>105791563
Conroe didn't have it on die and still BTFOd the Athlons

>>105791563
>the main issue for intel and games at the time was the aging fsb based platform,
Their core architecture was ok using FSB... Intel only removed the northbridge with nahalem.

>>105783669 (OP)
my first processor, learned programming and played Morrowind on that shit
10/10

>>105791589
conroe has a 4mb shared l2 , 12 way associative cache , which is huge for the time and penryn 6mb l2 cache, the socket 604 xeons got even bigger caches.

by comparison the prescott based chips only had a 2x1 or 2 mb 8 way associative l2 cache, that was not shared meaning all comminucation occured over the slow fsb. l1 was also half the size