AMD 3D V-Cache for Ryzen CPUs is Impressive: Up to 192MB L3 Cache
video description
Date: 2021-06-01
Related videos
Comments and reviews: 10
Zorba
Steve, you made the point AMD can work with TSMC and leverage that relationship compared to Intel who have to do it in house. Frankly on the basis of a 2 vs 1 fist fight it is a valid argument however as a comparison of business scale AMD & TSMC vs Intel on the basis of research capability I think you got that totally out of balance. That is not to say Intel hasn't been asleep at the wheel on CPUs in recent years but even the most biased reports recognise the Intel giant is waking up. So far no comment by any CES presenter on how they are fixing supply issues for gamers or industry.
reply
Steve, you made the point AMD can work with TSMC and leverage that relationship compared to Intel who have to do it in house. Frankly on the basis of a 2 vs 1 fist fight it is a valid argument however as a comparison of business scale AMD & TSMC vs Intel on the basis of research capability I think you got that totally out of balance. That is not to say Intel hasn't been asleep at the wheel on CPUs in recent years but even the most biased reports recognise the Intel giant is waking up. So far no comment by any CES presenter on how they are fixing supply issues for gamers or industry.
reply
Robert
Structural silicon -- basically AMD have got two slabs of undeveloped real estate positioned either side of the new die-on-die cache to make things level physically for contacting the integrated heat spreader. I can't see a technical problem with them making a V2.0 cache chip which is the full width of the underlying die but with twice or even three times this expanded cache. The result would by Ryzens with, say, 768MB of L3 cache and Threadrippers with a couple of GB of L3 cache. The price would be up there in lights though.
Interesting times.
reply
Structural silicon -- basically AMD have got two slabs of undeveloped real estate positioned either side of the new die-on-die cache to make things level physically for contacting the integrated heat spreader. I can't see a technical problem with them making a V2.0 cache chip which is the full width of the underlying die but with twice or even three times this expanded cache. The result would by Ryzens with, say, 768MB of L3 cache and Threadrippers with a couple of GB of L3 cache. The price would be up there in lights though.
Interesting times.
reply
Wereweeb
I wonder what will be the cost implications of this. On the one hand, they can increase yields by orders of magnitude, but that's not as important today. On the other hand, it's still a 45% larger die for a 15% increase in gaming performance. And perfectly stacking two dies with microscopic copper bumps is no easy feat.
I can see this making sense if they can offset some of the costs by reducing the process steps per die, especially that of the SRAM die. But I'm skeptical of that. Zen 4 might end up being pricey.
reply
I wonder what will be the cost implications of this. On the one hand, they can increase yields by orders of magnitude, but that's not as important today. On the other hand, it's still a 45% larger die for a 15% increase in gaming performance. And perfectly stacking two dies with microscopic copper bumps is no easy feat.
I can see this making sense if they can offset some of the costs by reducing the process steps per die, especially that of the SRAM die. But I'm skeptical of that. Zen 4 might end up being pricey.
reply
liaminwales
Nvidia have AMD FSR with there lack of backward compatibility, even if it's no where near as good as DLSS 2 if it just looks better than non native resolution and helps older cards it's a big PR win.
Suspect if it's on PS5/Xbox it may get good support in the next few years.
DLSS is amazing but the only cards that have it have less need than the older ones.
reply
Nvidia have AMD FSR with there lack of backward compatibility, even if it's no where near as good as DLSS 2 if it just looks better than non native resolution and helps older cards it's a big PR win.
Suspect if it's on PS5/Xbox it may get good support in the next few years.
DLSS is amazing but the only cards that have it have less need than the older ones.
reply
spodule6000
Is this really going to be L3 cache or actually L4? If it is L3 then does that mean there's no L3 cache on the CCX?
L4 cache has been tried on Intel's Crystal Well chips a few years ago and performed very well. those were 128MB.
Down the road, I see that this could be a real bonus for AMD's APUs, mitigating the bandwidth constraints of the RAM.
reply
Is this really going to be L3 cache or actually L4? If it is L3 then does that mean there's no L3 cache on the CCX?
L4 cache has been tried on Intel's Crystal Well chips a few years ago and performed very well. those were 128MB.
Down the road, I see that this could be a real bonus for AMD's APUs, mitigating the bandwidth constraints of the RAM.
reply
Charles
So 96MB cache per core complex. Assuming the core complex on zen 4 still the same at 8 cores and rumor has it that zen 4 epyc will have 12 compute dies of 8 cores each for total 96 cores. So .... the cache could be .... 12 ccx X 96MB = 1,152 MB or 1.1GB cache!!! in comparison the latest Epyc in the market has 256MB cache for 64 cores.
reply
So 96MB cache per core complex. Assuming the core complex on zen 4 still the same at 8 cores and rumor has it that zen 4 epyc will have 12 compute dies of 8 cores each for total 96 cores. So .... the cache could be .... 12 ccx X 96MB = 1,152 MB or 1.1GB cache!!! in comparison the latest Epyc in the market has 256MB cache for 64 cores.
reply
SteelSkin667
Considering that FSR is hardware-agnostic, a crucial difference between it and DLSS is that it doesn't need to be specifically supported by display drivers. If/when Nvidia decides to drop DLSS, it'll stop being supported on new hardware, whereas games releasing with an implementation of FSR will retain the functionality forever.
reply
Considering that FSR is hardware-agnostic, a crucial difference between it and DLSS is that it doesn't need to be specifically supported by display drivers. If/when Nvidia decides to drop DLSS, it'll stop being supported on new hardware, whereas games releasing with an implementation of FSR will retain the functionality forever.
reply
MrBarlien
I don't get the caginess when it comes to FSR, it's something that many old and new gpus may benefit from. If it only manages to work good on like 5 games and dies after two years it's still a win. It's not like it was a feature locked to and sold as part of a gpu architecture like dlss. Anyhow, great videos as always!
reply
I don't get the caginess when it comes to FSR, it's something that many old and new gpus may benefit from. If it only manages to work good on like 5 games and dies after two years it's still a win. It's not like it was a feature locked to and sold as part of a gpu architecture like dlss. Anyhow, great videos as always!
reply
Penny
I would be interested to get some kind of deep dive on the thermal implications and trade-offs of all of these silicon stacking solutions.
I also don't understand why they're using silicon as structural components when its thermal conductivity is so poor. Does that have to do with the thermal expansion characteristics?
reply
I would be interested to get some kind of deep dive on the thermal implications and trade-offs of all of these silicon stacking solutions.
I also don't understand why they're using silicon as structural components when its thermal conductivity is so poor. Does that have to do with the thermal expansion characteristics?
reply
Phoenix
I had a thought about 3D stacked cache a few years ago back when HBM was a little more relevant in the regular consumer market. Half dismissed it out of thermal concerns. Crazy to think they actually managed it. Amazing.
Now I wonder if it s possible to do with cores too. Or some novel cooling solution.
reply
I had a thought about 3D stacked cache a few years ago back when HBM was a little more relevant in the regular consumer market. Half dismissed it out of thermal concerns. Crazy to think they actually managed it. Amazing.
Now I wonder if it s possible to do with cores too. Or some novel cooling solution.
reply
Add a review, comment
Other channel videos
