Anonymous
6/14/2025, 2:19:48 PM No.105590922
Will 6TB VRAM be enough in 2035?
Replies:
Thread 105590922 - /g/ [Archived: 1106 hours ago]
Anonymous
6/14/2025, 2:36:12 PM No.105590998
In 2035 your GPU will still have 24 gigabytes of VRAM. HBM is for people who do something useful with their computer.
Anonymous
6/14/2025, 4:41:37 PM No.105591796
>>105591631
Lol
Lol
Anonymous
6/14/2025, 4:48:20 PM No.105591847
in 10 years common setups will have 24-32gb and flagships will have 64gb
Replies:
Anonymous
6/14/2025, 5:25:22 PM No.105592127
>>105590922 (OP)
whatever happened to those biological ram chips that they developed in the 2000s? Shit got memory holed quick.
whatever happened to those biological ram chips that they developed in the 2000s? Shit got memory holed quick.
Anonymous
6/14/2025, 5:54:25 PM No.105592397
>>105590922 (OP)
i bet on paradigm shift by 2030
i bet on paradigm shift by 2030
Anonymous
6/14/2025, 11:05:10 PM No.105594910
>>105590922 (OP)
I hope by 2035 we will have shit like Ryzen AI Max with 64CU and 256GB HBM inside a mini-PC.
I hope by 2035 we will have shit like Ryzen AI Max with 64CU and 256GB HBM inside a mini-PC.
Anonymous
6/14/2025, 11:05:57 PM No.105594921
Anonymous
6/14/2025, 11:51:23 PM No.105595386
>>105592457
we are gonna see the rtx 9060 ti 16gb vs 32gb with the decisive conclusion that 16gb is not enoughin 2035, same thing with rx 480/gtx 1060 in 2016 and 9060/5060 today
we are gonna see the rtx 9060 ti 16gb vs 32gb with the decisive conclusion that 16gb is not enoughin 2035, same thing with rx 480/gtx 1060 in 2016 and 9060/5060 today
Anonymous
6/15/2025, 1:12:39 AM No.105596057
>>105590922 (OP)
>absolute size of that die
won't they run into propagation delay issues at some point? like the speed of light it takes for a signal to get from one part of the die to another
>absolute size of that die
won't they run into propagation delay issues at some point? like the speed of light it takes for a signal to get from one part of the die to another
Replies:
Anonymous
6/15/2025, 1:19:08 AM No.105596092
>>105596057
Only if you run
>a single thread
and
>at mach fuck speeds
if you have multiples threads that run (somewhat independently) like a GPU it's not that much of an issue. The reason why they're building so many dies on such a small footprint is literally so they can utilize more spave, i.e. more dies per rack because silicon is reaching it's limits in performance density per die.
Only if you run
>a single thread
and
>at mach fuck speeds
if you have multiples threads that run (somewhat independently) like a GPU it's not that much of an issue. The reason why they're building so many dies on such a small footprint is literally so they can utilize more spave, i.e. more dies per rack because silicon is reaching it's limits in performance density per die.
Anonymous
6/15/2025, 6:06:06 AM No.105597838
>>105590922 (OP)
you're an idiot if you think they're going to glue a bunch of gpus together for you just so you can enjoy prettier games
you're an idiot if you think they're going to glue a bunch of gpus together for you just so you can enjoy prettier games
Anonymous
6/15/2025, 7:17:34 AM No.105598214
>>105590922 (OP)
>>105591847
>in 2035
For consumer GPUs:
>base 10060 SKU: 16GB
>10060 Super / Ti: 24GB
>base 10070 SKU: 16GB or 24GB
>10070 Super / Ti and base 10080: 32GB
>10090 might have 48GB or 64GB
AMD's lineup will be no different unless they put in a new CEO that isn't Huang's relative.
I'm sure that's what we're going to see in 10 years unless AMD, Intel and/or new chink players force Nvidia to step up their game and stop the vram jewry on consumer GPUs
>>105591847
>in 2035
For consumer GPUs:
>base 10060 SKU: 16GB
>10060 Super / Ti: 24GB
>base 10070 SKU: 16GB or 24GB
>10070 Super / Ti and base 10080: 32GB
>10090 might have 48GB or 64GB
AMD's lineup will be no different unless they put in a new CEO that isn't Huang's relative.
I'm sure that's what we're going to see in 10 years unless AMD, Intel and/or new chink players force Nvidia to step up their game and stop the vram jewry on consumer GPUs
Anonymous
6/15/2025, 10:56:24 AM No.105599215
>>105590922 (OP)
Trust me, we'll still be whining about 8gb not being enough.
Trust me, we'll still be whining about 8gb not being enough.
Anonymous
6/15/2025, 11:27:53 AM No.105599390
>>105597867
i mean okay i admit i don't know much about this stuff, but let's say you have a frequency of 4 gigahertz or whatever, that means a clock cycle lasts 0.25 nanoseconds, which means light can travel 7.5 centimeters in that time
surely this limits the chip size physically somehow at some point? maybe not at the level in OP graphic but eventually
i mean okay i admit i don't know much about this stuff, but let's say you have a frequency of 4 gigahertz or whatever, that means a clock cycle lasts 0.25 nanoseconds, which means light can travel 7.5 centimeters in that time
surely this limits the chip size physically somehow at some point? maybe not at the level in OP graphic but eventually
Replies:
Anonymous
6/15/2025, 1:01:56 PM No.105599883
>>105590922 (OP)
so pretty much they are expecting that almost all gains will be from size and power and no actual ipc gains
we are fucked
so pretty much they are expecting that almost all gains will be from size and power and no actual ipc gains
we are fucked
Anonymous
6/15/2025, 5:21:42 PM No.105601524
>>105599390
what you're thinking of affects pcie which goes up to 8ghz in 4.0.
not only are the pcie traces hundreds of times longer than anything inside a chip, and not only are the frequencies higher than anything inside the chips, but you can use a cable to go several feet out with pcie and it still works at full speed.
Now these gpus are a bunch of normal gpus connected together with what is more or less equivalent to pcie bus, but the length isn't a significant factor.
what you're thinking of affects pcie which goes up to 8ghz in 4.0.
not only are the pcie traces hundreds of times longer than anything inside a chip, and not only are the frequencies higher than anything inside the chips, but you can use a cable to go several feet out with pcie and it still works at full speed.
Now these gpus are a bunch of normal gpus connected together with what is more or less equivalent to pcie bus, but the length isn't a significant factor.