Will Alder Lake be good enough to compete with Zen 3 XT and its 3D cache?

Well, my 12700K arrives tomorrow and I've decided to keep it over the 5800X. In the end Alder Lake is a bit faster, and I'm not coming from an older AM4 setup so a new motherboard is a must anyway. I would not mind going DDR5, but I do have a 32GB 3600 CL16 kit waiting to be used.

2600K>12700K should be a decent upgrade. :LOL:
 
Well, my 12700K arrives tomorrow and I've decided to keep it over the 5800X. In the end Alder Lake is a bit faster, and I'm not coming from an older AM4 setup so a new motherboard is a must anyway. I would not mind going DDR5, but I do have a 32GB 3600 CL16 kit waiting to be used.

2600K>12700K should be a decent upgrade. :LOL:
12700k is another class compared to 5800x. It’s a great choice.
 

Will Alder Lake be good enough to compete with Zen 3 XT and its 3D cache?

Current game performance between Zen 3 and Alder Lake at 1080p is roughly equivalent except for very single threaded dominated games where Alder Lake is better. If Zen with 3D cache is 15% faster in games (talking about 1080p, fastest card as in 6900XT or 3090) then Zen with 3D cache will beat Alder Lake in general with most recent games. Alder lake is no where close to being 15% faster than Zen 3 in gaming.

Other workloads Alder lake looks extremely good performance wise, how 3D cache enhances those other useful workloads is to be seen. Will Zen 3D Cache catch up to Alder lake in something like Cinebench R23, where it already runs in Zen's 3 cache as it is? I don't think so but could be wrong. Looks like Zen 3D will be faster in games at ridicules settings for high end video cards but may not be for much else. Will be interesting but we may have to wait till next year.

AMD can also play with their boost settings, number of cores boosted plus any other things to increase performance such as using more power but still being less than Intel. Intel is pushing above 250w with the 12900k at times, that does give AMD some playing room if their CPU can handle it. Frankly I think AMD was caught off guard, their current pricing is way too high now with Alder Lake out. Even a $100 cut to me would fall short.
 

Will Alder Lake be good enough to compete with Zen 3 XT and its 3D cache?

Current game performance between Zen 3 and Alder Lake at 1080p is roughly equivalent except for very single threaded dominated games where Alder Lake is better. If Zen with 3D cache is 15% faster in games (talking about 1080p, fastest card as in 6900XT or 3090) then Zen with 3D cache will beat Alder Lake in general with most recent games. Alder lake is no where close to being 15% faster than Zen 3 in gaming.

Other workloads Alder lake looks extremely good performance wise, how 3D cache enhances those other useful workloads is to be seen. Will Zen 3D Cache catch up to Alder lake in something like Cinebench R23, where it already runs in Zen's 3 cache as it is? I don't think so but could be wrong. Looks like Zen 3D will be faster in games at ridicules settings for high end video cards but may not be for much else. Will be interesting but we may have to wait till next year.

AMD can also play with their boost settings, number of cores boosted plus any other things to increase performance such as using more power but still being less than Intel. Intel is pushing above 250w with the 12900k at times, that does give AMD some playing room if their CPU can handle it. Frankly I think AMD was caught off guard, their current pricing is way too high now with Alder Lake out. Even a $100 cut to me would fall short.
AMD has said that Zen3D won’t have any big changes to core speeds, etc.

I could see them doing some refinements to curve optimizer. Or maybe rolling some of that into PBO.

I actually think the extra cache will help in cinibench, because the 5600g and 5700g take a near generational hit in R23 performance, with halved cache. It seems clear that zen performance is very cache dependent. Whereas rocketlake got a lot of mileage from relatively little cache. And while Alderlake has more cache, it’s still half of zen.

One wonders what a 60mb cache 12700k would be like.
 
Last edited:
One wonders what a 60mb cache 12700k would be like.

An $800 12700X HEDT CPU....

In all fairness, I think Intel was running out of die space with 14nm on RKL, so the cache took a hit. I'm sure Intel could squeeze more cache on there if they wanted to.
 
AMD has said that Zen3D won’t have any big changes to core speeds, etc.

I could see them doing some refinements to curve optimizer. Or maybe rolling some of that into PBO.

I actually think the extra cache will help in cinibench, because the 5600g and 5700g take a near generational hit in R23 performance, with halved cache. It seems clear that zen performance is very cache dependent. Whereas rocketlake got a lot of mileage from relatively little cache. And while Alderlake has more cache, it’s still half of zen.

One wonders what a 60mb cache 12700k would be like.
I thought AMD indicated more advanced power management features depending upon load. This could drive faster cores more in carrying certain type of loads more intelligently at faster/longer durations. Zen 2 to Zen 3, as in 3950x vs 5950x had the same cache size overall but much improved SC and MC performance since the l3 Cache was more local to 8 cores vice split between two 4 cores. So you may have an excellent point. The more uniform L3 Cache woke up the sleeping giant.

In the end to be competitive, AMD, if they can will push as needed to get the performance as high as possible. The 12600K/LF kills the 5600x and even the 5800x for the most part, the 12400 maybe even faster than the 5600x and under hopefully $200. I can see in the short term AMD doing short term sells for the season, rebates and other promotions while Intel is constantly sold out, DDR 5 not available for the higher end boards etc. Next year AMD will stay as a market leader or will not on the desktop/mobile. Much would be riding on AMD to perform well for a number of reasons. I do not see AMD not pushing power higher if they need to to compete so as to maintain that leadership perception and pricing.
 
Most of the Alder lake differences between i7 and i9 in non core-maxed workloads is down to cache, so it’s definitely important in that uarch.
 
Epyc with VCache, Milan to MilanX (Zen 3 to Zen 3 with VCache) shows phenomenal performance gains on these workloads for HPC performance. Can't help but think that it will carry over for the most part to the desktop on a number of programs.

s-AMD-EPYC-Milan-X-Performance-Gains-Gen-3-696x475.jpg
 
768m of L3 cache is getting a bit crazy, when I became aware of CPU cache amount my computer had way less ram than that.....
 
Epyc with VCache, Milan to MilanX (Zen 3 to Zen 3 with VCache) shows phenomenal performance gains on these workloads for HPC performance. Can't help but think that it will carry over for the most part to the desktop on a number of programs.

View attachment 411039
Wishful thinking but who knows. It would be interesting to see the cache help out in dGPU performance akin to the 5775c.

I am guessing this will be reserved for TR and, hopefully, available down to 8 cores.
TR in zen 3 with 3d stacked cache and quad pumped ddr4 would be a great match for the higher end ADL parts.

https://www.techpowerup.com/287439/amd-ryzen-threadripper-5000-series-delayed-to-2022?cp=2
 
Wishful thinking but who knows. It would be interesting to see the cache help out in dGPU performance akin to the 5775c.

I am guessing this will be reserved for TR and, hopefully, available down to 8 cores.
TR in zen 3 with 3d stacked cache and quad pumped ddr4 would be a great match for the higher end ADL parts.

https://www.techpowerup.com/287439/amd-ryzen-threadripper-5000-series-delayed-to-2022?cp=2
TR 24 core took the same power as the TR 64 core version, meaning there was a lot of variance on the dies. While the Epyc version was probably even better for chip quality over the TR. I do not see AMD just throwing away the higher power, lower quality Zen 3+ (Maybe VCache as well versions). Unless TSMC is much more consistent on the chip quality. Now of course the desktop versions could take all of the lower quality dies but will AMD just dump the HDET desktop? I doubt it but what do I know. I expect at least a Zen 3+ version for TR and probably later a VCache version (milking it even further if AMD goes that route). If AMD just goes full VCache, I would definitely be happy as well as many customers if they can actually provide enough quantity of chips. Having a single design, a.k.a VCache only version die does make it simpler, well sorta, if one can provide the numbers needed. The design I am assuming is flexible where VCache is more of an option vice a requirement except die height has to be the same or one thick heat spreader for the Zen3+ dies. Since the height of the VCache dies are suppose to be the same as Zen 3 -> how does one make just Zen3+ dies only and VCache ones at the same height? I hope, AMD surprises us more than we expect.
 
Back
Top