Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 337 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Ajay

Lifer
Jan 8, 2001
15,714
7,988
136
AFAIK AMD has not yet decided if Zen6 will be on AM5 or a new platform.
AMD has most definitely decided. They are just holding their cards close to their vest. Almost all of these decisions/changes occur internally a year a two b/4 they are leaked. Even board partners are getting specs far later than they did 10 years ago. Such is the state of a very competitive market.
 

adroc_thurston

Platinum Member
Jul 2, 2023
2,626
3,871
96

eek2121

Platinum Member
Aug 2, 2005
2,942
4,047
136
AMD has most definitely decided. They are just holding their cards close to their vest. Almost all of these decisions/changes occur internally a year a two b/4 they are leaked. Even board partners are getting specs far later than they did 10 years ago. Such is the state of a very competitive market.
You guys are talking about AMD killing off AM5 and AM4 isn’t even dead yet. AMD has unreleased Zen 3 stuff on AM4 still.
 

Ghostsonplanets

Senior member
Mar 1, 2024
407
693
96
Middling at best. MTL Arc already does that at elevated power levels.
Arc's excellence in benchmarks means nothing if certain games outright do not work with it, or require some fiddling to get them to work.
Doesn't Arc literally cheats in by replacing the programs shader with custom hand optimized shaders in the popular graphics benchmarks? And such results don't reflect the performance in games. So much that Intel had to add a option in their Graphics Panel to disable these?

I remember the MTL GFX IP hype due to TS leaks and people taking those results at face value and thinking it was way ahead of PHX. But in reality, MTL is basically a match or behind PHX, but with worse driver stack, incompatible games and power curve that is biased towards the higher end of the curve.

Don't take me wrong, MTL GFX is a fine piece. But it's amazing that Intel has been talking and pushing integrated graphics for 10+ years now and are still so behind.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,828
3,661
136
Doesn't Arc literally cheats in by replacing the programs shader with custom hand optimized shaders in the popular graphics benchmarks? And such results don't reflect the performance in games. So much that Intel had to add a option in their Graphics Panel to disable these?
That's literally what driver optimizations are. NVIDIA and AMD do it with games - the fact that they are ahead is because they have decades of head start.
 

uzzi38

Platinum Member
Oct 16, 2019
2,677
6,256
146
You're forgetting about Lunar Lake.
???

No, you're underestimating how arse MTL is. LNL being as good as it is at 15W relative to MTL only puts it on par with PHX at 15W.

MTL is utterly unusable at 15W, it consistently falls far short of Phoenix on average frame rates and has utterly unplayable frame times to boot at 25W. It just gets even worse at 15W.

 

Glo.

Diamond Member
Apr 25, 2015
5,742
4,626
136
Yes, even with 33% more CUs Strix Point is still scoring barely enough in TS to beat juiced up MTL. And Xe2 is supposedly twice as fast as MTL (8 Xe2 cores vs 4 Xe-LPG cores).
LNL is the same perf as MTL but in smaller(lower) thermal envelope on iGPU.

What LNL does is it increases the amount of ALUs per cluster. Thats it. In terms of performance, it will still be 1024 ALUs vs 1024 ALUs between MTL and LNL. Yes, LNL will be more efficient, but will not mean it will be TWICE AS FAST. It means it will be the same perf. at lower power.
 

Ghostsonplanets

Senior member
Mar 1, 2024
407
693
96
That's literally what driver optimizations are. NVIDIA and AMD do it with games - the fact that they are ahead is because they have decades of head start.
Well, yes. What I mean to say is that Arc Alchemist is an uArch that does exceptional well in benchmarks but these results don't reflect into real game performance. Intel literally had to add a "Disable performance optimizations" into their control panel due to Arc benchmark not reflecting real world performance and being called out for it.


LNL is the same perf as MTL but in smaller(lower) thermal envelope on iGPU.

What LNL does is it increases the amount of ALUs per cluster. Thats it. In terms of performance, it will still be 1024 ALUs vs 1024 ALUs between MTL and LNL. Yes, LNL will be more efficient, but will not mean it will be TWICE AS FAST. It means it will be the same perf. at lower power.
Well not quite, SIMDs are twice the size (PVC-style) too.
So it's a fatter GPU per Xe Core? I always thought Intel GPU arch subdivision to be too high and leading to occupancy problems and others bottleneck, not too different from past Mali GPUs(I might be totally off base here).

So they going with fatter Xe Core to try to cram on area and increase per core performance is a nice change of direction.

Edit: I totally missed this is the Zen 5 thread. Sorry for going off-topic, mods.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,381
2,880
136
Yes, even with 33% more CUs Strix Point is still scoring barely enough in TS to beat juiced up MTL. And Xe2 is supposedly twice as fast as MTL (8 Xe2 cores vs 4 Xe-LPG cores).
MTL has 8 Xe-LPG cores, not 4!
What we should expect from Lunar Lake IGP I don't know.
But I am not even sure, If AMD will have a direct competitor against It, maybe Kraken Point? 4+4 and 8CU looks pretty similar.