Page 4 of 4

Re: Why are CPU projects worth so few points relative to GPU

Posted: Tue Jun 16, 2020 4:24 pm
by MeeLee
The_Bad_Penguin wrote:Would love to find an affordable 3990x, but "rumors" are that the next gen Ryzens (4Q2020?) will be ~20% more efficient. And certainly, Ryzen/Threadripper 5nm would be a sight to behold.

Don't know that I would agree that "But with high end next gen CPUs, we'll definitely see a closing between CPU and GPU on FAH," because as there will be a new gen high-end cpu's, there will also be a new gen of high-end gpu's.

Starting to see web articles/blogs/videos that "NVIDIA GeForce RTX 3090 rumors: up to 60-90% faster than RTX 2080 Ti"

I would assume that a rumored nvidia 3090 ($2000?) [4m ppd?] would cost less than a 3990x ($3000?) [3m ppd?] or next gen threadripper 64t/128t ($???) [4m+ ppd?].
You can't deny that Ryzen already closed the gap between CPU and GPU?
GPUs also started having high ppd with the introduction of the RTX GPUs 2 years ago.
CPUs are just 2 years behind.
Sure, GPUs are faster, and more efficient.
However, CPUs still have a long way to go. Especially if we're talking about x86 CPU cores mixed with ARM cores.
GPUs on the other hand, are running out of space.
The TDP you can fit on a PCIE card, can not exceed 300W, without causing serious heat issues (I already experience heat issues at 225W).
The only way GPUs will increase in performance, is, if they lower boost frequencies when they increase core count, or with splitting the die into 2 pieces, spreading the heat across the heat sink in 2 locations for better cooling.
And of course, going to 5nm, or 3nm.

Intel CPUs still need to hit 7nm. Their 10nm are reported to be as efficient as AMD's 7nm design.
Aside from die size reduction, they could move away from a full x86 core design, and their size limit is nearly the entire motherboard, with a socket size increase, which can result in better cooling solutions than possible on GPUs.

I would be estimating that we're near to the maximum possible performance we can get out of GPUs (within 25%).
CPUs still have quite a while to go!

Re: Why are CPU projects worth so few points relative to GPU

Posted: Tue Jun 16, 2020 7:10 pm
by PantherX
The_Bad_Penguin wrote:Would love to find an affordable 3990x, but "rumors" are that the next gen Ryzens (4Q2020?) will be ~20% more efficient. And certainly, Ryzen/Threadripper 5nm would be a sight to behold...
There's strong speculations that Zen 3 will be 7nm which will be released later this year while Zen 4 will be 5 nm which will be released next year (https://www.techpowerup.com/review/futu ... -releases/)

Re: Why are CPU projects worth so few points relative to GPU

Posted: Tue Jun 16, 2020 7:14 pm
by PantherX
MeeLee wrote:...I would be estimating that we're near to the maximum possible performance we can get out of GPUs (within 25%)...
For a monolithic die, I would think so. However, there are speculations that Nvidia will move to MCM (multi-chip module) GPU packages with Hooper so that could provide more performance. Of course, I am not sure if existing code would be fully compatible or not but that's potentially a future problem.

Nonetheless, it is a good time to look forward to new CPU/GPU technologies over the next few years.

Re: Why are CPU projects worth so few points relative to GPU

Posted: Tue Jun 16, 2020 8:40 pm
by bruce
PantherX wrote:Of course, I am not sure if existing code would be fully compatible or not but that's potentially a future problem.

Nonetheless, it is a good time to look forward to new CPU/GPU technologies over the next few years.
It's all in the drivers.

Re: Why are CPU projects worth so few points relative to GPU

Posted: Fri Jun 19, 2020 1:01 am
by MeeLee
Zen 3 will be 7nm++.
Current Zen 2, is 7nm; however some parts of the chip are believed to be 14nm.
There also are a few architectural improvements coming. 20% is within the realm of possibilities, however most rumors based on leaked benchmarks show just about 100mHz higher GPU frequency.
Which means instead of 3,5Ghz it'll have a 3650-3700Mhz base clock, and instead of a 4,4hz boost, it hits 4,5Ghz.
Mind you, I've never been able to exceed 4,1050Mhz on my boards (with only 2x 4-pin CPU connectors).
So in reality I might see a 4,2Ghz on all cores.