AMD Radeon RX 7900

firejuggler · 2022-11-17, 14:27

Tomshardware give here some specifiction:
https://www.tomshardware.com/news/am...ries-deep-dive

for the RX 7900 XT
FP64 (double) performance 3.226 TFLOPS (1:16) and
FP64 (double) performance 3.848 TFLOPS (1:16) for the XTX

LaurV · 2022-11-18, 05:31

They are still memory limited, both. If you look to Rvii vs. Rvii Pro, the last has doubled the DP flops, but when you look here, they do the same amount of work per time unit, because the limiting factor is not the DP performance, but how much data you can bring in to play with it.

M344587487 · 2022-11-18, 16:05

But they do have infinity cache, 80/96MB(MiB?) of L3 cache for XT/XTX. It's probably why the 6000 series gets close to R7's performance at the top end, for the 7000 series there's less of it but much quicker (something like 2-3TB/s aggregate). That cache is split across the memory dies, each die getting 16MB, so it might not be that simple.

https://mersenneforum.org/showthread.php?t=28217

In my opinion the XTX is a no brainer out of the two. That seems to be the consensus for gaming too, don't be surprised if the XTX quickly runs out of stock. They may have priced the XT so high as the booby prize if they know XTX has inadequate supply. Nvidia 4000's are still priced sky high and they may not budge like they should (supposedly high stock of 3000 series to get through), market forces that should be in our favour may not be for a while.

2022-11-17, 14:27	#1
firejuggler "Vincent" Apr 2010 Over the rainbow 2884₁₀ Posts	AMD Radeon RX 7900 Tomshardware give here some specifiction: https://www.tomshardware.com/news/am...ries-deep-dive for the RX 7900 XT FP64 (double) performance 3.226 TFLOPS (1:16) and FP64 (double) performance 3.848 TFLOPS (1:16) for the XTX Last fiddled with by firejuggler on 2022-11-17 at 14:27

2022-11-18, 05:31	#2
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 10,273 Posts	They are still memory limited, both. If you look to Rvii vs. Rvii Pro, the last has doubled the DP flops, but when you look here, they do the same amount of work per time unit, because the limiting factor is not the DP performance, but how much data you can bring in to play with it. Last fiddled with by LaurV on 2022-11-18 at 05:32

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Radeon VII, gpuowl @ 12V?	mrh	GPU Computing	3	2022-10-28 18:46
Radeon VII GPU dead? (why?)	preda	GPU Computing	3	2022-09-24 20:26
Radeon RX 5500	James Heinrich	GPU Computing	11	2021-02-01 16:07
AMD Radeon Pro WX 3200	ET_	GPU Computing	1	2019-07-04 11:02
Radeon Pro Duo	0PolarBearsHere	GPU Computing	0	2016-03-15 01:32

2022-11-18, 16:05	#3
M344587487 "Composite as Heck" Oct 2017 3×311 Posts	But they do have infinity cache, 80/96MB(MiB?) of L3 cache for XT/XTX. It's probably why the 6000 series gets close to R7's performance at the top end, for the 7000 series there's less of it but much quicker (something like 2-3TB/s aggregate). That cache is split across the memory dies, each die getting 16MB, so it might not be that simple. https://mersenneforum.org/showthread.php?t=28217 In my opinion the XTX is a no brainer out of the two. That seems to be the consensus for gaming too, don't be surprised if the XTX quickly runs out of stock. They may have priced the XT so high as the booby prize if they know XTX has inadequate supply. Nvidia 4000's are still priced sky high and they may not budge like they should (supposedly high stock of 3000 series to get through), market forces that should be in our favour may not be for a while.