AMD EPYC-CPUs based on 3D V-Cache and teases Zen 4 as well as 128-core Bergamo

Published by

Click here to post a comment for AMD EPYC-CPUs based on 3D V-Cache and teases Zen 4 as well as 128-core Bergamo on our message forum
data/avatar/default/avatar13.webp
MegaFalloutFan:

Its not the same thing but its close enough, go check Intel Broadwell 5775C Benchmarks vs otehr CPUs, it had 128MB of Cache used for iGPU, BUT when you disabled iGPU, all that cache was used toas L4 CPU cache Yes it was winning in some fringe cases by a lot, but on the general note it wasn't mind blowing.
Checkout this graph from AMD @ 4 GHZ with and without 3DVC on Ryzen 5900X: https://static.techspot.com/images2/news/bigimage/2021/10/2021-10-12-image-31-j_1100.webp I agree it will depend but when you hit a CPU limited situation or a moment if the game AI overloads the CPU, it looks like it will help out tremendously then. Still it would not HURT to have it on standby.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
Valken:

Checkout this graph from AMD @ 4 GHZ with and without 3DVC on Ryzen 5900X: https://static.techspot.com/images2/news/bigimage/2021/10/2021-10-12-image-31-j_1100.webp I agree it will depend but when you hit a CPU limited situation or a moment if the game AI overloads the CPU, it looks like it will help out tremendously then. Still it would not HURT to have it on standby.
zen3+ should also allow 1:1 fclk for 4133/4266mhz sticks hope it offers pci-e 5.0 too at competitive prices like intel's 12600
https://forums.guru3d.com/data/avatars/m/164/164033.jpg
Valken:

Checkout this graph from AMD @ 4 GHZ with and without 3DVC on Ryzen 5900X: https://static.techspot.com/images2/news/bigimage/2021/10/2021-10-12-image-31-j_1100.webp I agree it will depend but when you hit a CPU limited situation or a moment if the game AI overloads the CPU, it looks like it will help out tremendously then. Still it would not HURT to have it on standby.
Yeah I think my 3900x will change into the 3D one. But most likely the 16-core one and then possibly change my 32gigs of ram into 64. And go with it until Zen 6 or something.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
Astyanax:

Key words "From" and "AMD"
true,take every slide with a ton of salt they marketed SAM the same way, it turned out to only help in very limited number of games/scenarios
https://forums.guru3d.com/data/avatars/m/262/262208.jpg
cucaulay malkin:

zen3+ should also allow 1:1 fclk for 4133/4266mhz sticks hope it offers pci-e 5.0 too at competitive prices like intel's 12600
I don't think so we will see above 4000mhz RAM on current ZEN with 3D V-Cache or not, if yes then maybe with very loose timings or unless they change that but let's see Do we need PCI_E 5.0? I still think we don't need it on X570 or even on cheaper platforms like B550 etc, maybe with ZEN 4 or 5 PCI_E 5.0 will make sense with DDR5 Hope this helps Thanks, Jura
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
jura11:

I don't think so we will see above 4000mhz RAM on current ZEN with 3D V-Cache or not, if yes then maybe with very loose timings or unless they change that but let's see Do we need PCI_E 5.0? I still think we don't need it on X570 or even on cheaper platforms like B550 etc, maybe with ZEN 4 or 5 PCI_E 5.0 will make sense with DDR5 Hope this helps Thanks, Jura
nah 4.0 is fine
https://forums.guru3d.com/data/avatars/m/90/90667.jpg
With this new Raptorium cryptocurrency on the rise, we might see the 3D cache Ryzen fade away from existence and we won't be able to use them.
https://forums.guru3d.com/data/avatars/m/266/266726.jpg
cucaulay malkin:

zen3+ should also allow 1:1 fclk for 4133/4266mhz sticks hope it offers pci-e 5.0 too at competitive prices like intel's 12600
I don't think that will happen unless they change the process node of the io die , Though it is possible they could widen the data path, so that they can push more bandwidth at a lower frequency. currently the if links can do 32byte reads per cycle, so technically if they widened the data path to 512bit from 256, in a dual die config they could basically double their read performance in multithreaded workloads , writes would stay the same though since its only 16bytes per cycle with zen2/3. However I wouldn't expect that much, mainly because its a stopgap until zen4.
https://forums.guru3d.com/data/avatars/m/282/282473.jpg
user1:

I don't think that will happen unless they change the process node of the io die , Though it is possible they could widen the data path, so that they can push more bandwidth at a lower frequency. currently the if links can do 32byte reads per cycle, so technically if they widened the data path to 512bit from 256, in a dual die config they could basically double their read performance in multithreaded workloads , writes would stay the same though since its only 16bytes per cycle with zen2/3. However I wouldn't expect that much, mainly because its a stopgap until zen4.
i think i've seen people saying the g-versions can do higher fclk than x,and they are later production samples. hence my assumption.
https://forums.guru3d.com/data/avatars/m/266/266726.jpg
cucaulay malkin:

i think i've seen people saying the g-versions can do higher fclk than x,and they are later production samples. hence my assumption.
the g series ryzen 4000/5000 chips are a monolithic 7nm die, thats why the fabric clocks higher, actually have one of those, was able to run some benchmarks at 4733mt/s:2366fclk:D. bandwidth is pretty good despite being single rank. 68gb/s~ , It could probably go higher if I didn't have the igpu enabled.
data/avatar/default/avatar02.webp
Valken:

Checkout this graph from AMD @ 4 GHZ with and without 3DVC on Ryzen 5900X: https://static.techspot.com/images2/news/bigimage/2021/10/2021-10-12-image-31-j_1100.webp I agree it will depend but when you hit a CPU limited situation or a moment if the game AI overloads the CPU, it looks like it will help out tremendously then. Still it would not HURT to have it on standby.
Here someone tested how L3 cache affects gamed on ADL, 12600K vs 12900K, 4Ghz locked, 6 P cores 30Mb vs 6 P cores 20Mb https://www.capframex.com/tests/Alder%20Lake-S%2012900K%20is%20the%20new%20Gaming%20King https://cms.capframex.com/api/assets/capframex-com/7a945c16-6cf9-4b21-b035-fb545dfe0fff/grafik.png
data/avatar/default/avatar09.webp
MegaFalloutFan:

Here someone tested how L3 cache affects gamed on ADL, 12600K vs 12900K, 4Ghz locked, 6 P cores 30Mb vs 6 P cores 20Mb https://www.capframex.com/tests/Alder Lake-S 12900K is the new Gaming King https://cms.capframex.com/api/assets/capframex-com/7a945c16-6cf9-4b21-b035-fb545dfe0fff/grafik.png
"Star Wars Jedi: Fallen Order shows a clear advantage of the additional 10MB L3 cache with almost 8%. Cyberpunk 2077 on the other hand runs only 3.3% faster. On average, the advantage of the larger L3 caches is slightly less than 6%." Star Wars was a single threaded game and reached up to 8% gain. I wish they tested more CPU dependent games like Civilization in that mode but it is obvious more cache does not hurt one bit.
data/avatar/default/avatar12.webp
Valken:

"Star Wars Jedi: Fallen Order shows a clear advantage of the additional 10MB L3 cache with almost 8%. Cyberpunk 2077 on the other hand runs only 3.3% faster. On average, the advantage of the larger L3 caches is slightly less than 6%." Star Wars was a single threaded game and reached up to 8% gain. I wish they tested more CPU dependent games like Civilization in that mode but it is obvious more cache does not hurt one bit.
More doesn't hurt but how much is enough? 30Mb is the sweet spot? 40Mb? Does it have ANY effect in 4K? I game in 4K/120hz
data/avatar/default/avatar08.webp
MegaFalloutFan:

More doesn't hurt but how much is enough? 30Mb is the sweet spot? 40Mb? Does it have ANY effect in 4K? I game in 4K/120hz
All very good questions so would be interesting to see something like Ryzen 5 vs ADL with cache differences locked at a fixed frequency then measured across different games vs resolution to determine CPU IPC. Ryzen 5 and ADL seems to be nearly on par in terms of IPC so it would be a good academic test.
https://forums.guru3d.com/data/avatars/m/266/266726.jpg
MegaFalloutFan:

More doesn't hurt but how much is enough? 30Mb is the sweet spot? 40Mb? Does it have ANY effect in 4K? I game in 4K/120hz
are you fpu limited? are you int limited? if the answer is "no" , then more cache will generally help you, basically caches exist purely to compensate for the slow access times and lack of bandwidth needed to feed the core, if ram was fast enough, then cache wouldn't be needed. to give an idea of how far behind ram is , l1 cache of a lowly intel core 2 cpu gets >200gb/s , and has like 1ns of total latency(only 32kb). cache of course has diminishing returns, but it will continue to scale for a while, before it becomes negligible , probably >256mb would be needed to reach that point depending on the cpu, and how smart the cache is about what data to keep.
data/avatar/default/avatar40.webp
user1:

are you fpu limited? are you int limited? if the answer is "no" , then more cache will generally help you, basically caches exist purely to compensate for the slow access times and lack of bandwidth needed to feed the core, if ram was fast enough, then cache wouldn't be needed. to give an idea of how far behind ram is , l1 cache of a lowly intel core 2 cpu gets >200gb/s , and has like 1ns of total latency(only 32kb). cache of course has diminishing returns, but it will continue to scale for a while, before it becomes negligible , probably >256mb would be needed to reach that point depending on the cpu, and how smart the cache is about what data to keep.
Im always GPU limited, 4K/120 even with 3090 i cant achive that in most new games