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

Page 679 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

jdubs03

Golden Member
Oct 1, 2013
1,280
902
136
Last edited:

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
NBC has their review up of the Zenbook model. Battery life looks good, not really an improvement over what was already available, but is better than a similarly equipped Snapdragon Galaxy Book *(roughly equal when normalized for battery capacity)*.

1722177145285.png

Their Zenbook scores slightly higher than Anandtech's in single core tests and they have Proart numbers that score slightly higher than that. I think the Zenbook doesn't sustain single core boost and for some reason, Anandtech's laptop is hitting less frequency than others, which is decreasing their 1T Spec numbers a bit. Even with that, IPC is disappointing at < 10% increase, but it seems like they tried to balance performance and efficiency with STX. Desktop and server should have a bit higher IPC.

Overall it seems like STX is a minor performance bump, good efficiency increase, really good graphics bump, and good battery life. Nothing revolutionary, but an all around really good performer. It clearly leads the x86 space and doesn't have the issues that the Arm on Windows laptops have to deal with. Having the fastest NPU allows for laptop vendors to market it as the AI leader. Even if that doesn't really mean anything right now, with the AI fever going around the world, it's a great marketing point for them to advertise.
 

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
If you want to know how STX performs with higher power, you can go to the Zenbook review and look for the Asus Proart numbers. The review isn't up yet, but they have the numbers in the Zenbook charts. Not sure what the power consumption in the Proart is though.

1722179151597.png
1722178248214.png
1722178322490.png
1722178351926.png
 

Attachments

  • 1722178219347.png
    1722178219347.png
    20 KB · Views: 44
Last edited:

CouncilorIrissa

Senior member
Jul 28, 2023
723
2,676
106
  • Like
Reactions: Joe NYC

gdansk

Diamond Member
Feb 8, 2011
4,568
7,681
136
Strix Point is much worse than I expected.

No point waiting for cheap Kraken Point when they'll only be ~10% better compared to existing 7/8000 laptops. Actually since it's 4+4 it might be worse in MT.
 
  • Like
Reactions: controlflow

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
Strix Point is much worse than I expected.

No point waiting for cheap Kraken Point when they'll only be ~10% better compared to existing 7/8000 laptops. Actually since it's 4+4 it might be worse in MT.

It may be able to offer better efficiency. That seems to have been a focus for Zen 5 on mobile.
 
  • Like
Reactions: Joe NYC

Josh128

Golden Member
Oct 14, 2022
1,319
1,984
106
From Phoronix's 100+ Linux benchmark test:

These initial Ryzen AI 9 HX 370 results have me super excited for Zen 5. It's different having AMD shipping their next-gen laptop SoCs ahead of the desktop processors, but in a few weeks attention will turn to the Ryzen 9000 series. I've been pushing the limits of this SoC in being eager to test Zen 5, so stay tuned for more articles ahead of the Ryzen 9000 series launch. The generational performance uplift is nice with Zen 5 but really captivating my interest has been the large power efficiency gains at least among these laptop SoCs. The Ryzen AI 9 HX 370 also furthers its lead over Intel Meteor Lake that makes an even larger hill for Lunar Lake to climb, which will be interesting to see how that battle goes later in the quarter.
 

Abwx

Lifer
Apr 2, 2011
11,884
4,873
136
Bit strange. Perhaps it's M3 Max mislabelled as M3.

Whatever it is that s still an impossible curve, there s no CPU whose perf increase linearly with power, that s against the laws of physics.

This and the fact that the 8945H gain only 10-12% perf from 35W to 70W lead me to think that this guy has no clue about power mesurements methodologies.

Same as NBC who put the 370 at same perf/watt in ST than a 185H, yet Computerbase measure 40-50% more power for the 185H in Cinebench ST at 28W while the 370 hoover at 18-20W.

Edit : There s a pic of the die at Bilibili, die size is 225.64mm2 :


 
Last edited:
  • Like
Reactions: lightmanek

ryanjagtap

Member
Sep 25, 2021
146
214
126
Splitting core to 2 different CCXs really hurt performance. Still don't understand why they did it, it seems that doing one 8-core hybrid CCX solution would have been better alternative.
Yup. I thought the same. They even made a 7545U/8540U with 2+4c config with unified cache. Could have done something like that for efficiency rather than doing this.
 

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
Whatever it is that s still an impossible curve, there s no CPU whose perf increase linearly with power, that s against the laws of physics.

This and the fact that the 8945H gain only 10-12% perf from 35W to 70W lead me to think that this guy has no clue about power mesurements methodologies.

Same as NBC who put the 370 at same perf/watt in ST than a 185H, yet Computerbase measure 40-50% more power for the 185H in Cinebench ST at 28W while the 370 hoover at 18-20W.

Edit : There s a pic of the die at Bilibili, die size is 225.64mm2 :



NBC does power and perf/watt measurements for the whole system, which greatly obscures the differences in SoC power/efficiency.
 

mikk

Diamond Member
May 15, 2012
4,296
2,382
136
Barely faster than a 780M with same LPDDR5x-7500? What happened? Bandwidth limitation too big?
 

FlameTail

Diamond Member
Dec 15, 2021
4,384
2,761
106
The fact that a single CCX can only support upto 8 cores seems to be a problem for AMD. More cores would require another CCX, and a seperate L3 block. Perhaps they should work on larger CCXes, or even do a rework of their core cluster hierarchy?

For comparison, the ARM DSU-120 can support upto 14 cores in a single cluster, with a single block of L3 cache.
 
Last edited:

FlameTail

Diamond Member
Dec 15, 2021
4,384
2,761
106
Barely faster than a 780M with same LPDDR5x-7500? What happened? Bandwidth limitation too big?
RAM bandwidth and GPU L2 size are both unchanged... meaning that it's essentially the same memory subsystem for the GPU as Phoenix.
 

gdansk

Diamond Member
Feb 8, 2011
4,568
7,681
136
The fact that a single CCX can only support upto 8 cores seems to be a problem for AMD. More cores would require another CCX, and a seperaye L3 block. Perhaps they should work on larger CCXes, or even do a rework of their core cluster hierarchy?
That is what Zen 6 is to solve, I guess.
In Bergamo/TurinD it isn't a significant competitive disadvantage but it seems to be a problem here.
 
  • Like
Reactions: igor_kavinski

Hitman928

Diamond Member
Apr 15, 2012
6,695
12,370
136
The fact that a single CCX can only support upto 8 cores seems to be a problem for AMD. More cores would require another CCX, and a seperate L3 block. Perhaps they should work on larger CCXes, or even do a rework of their core cluster hierarchy?

For comparison, the ARM DSU-120 can support upto 14 cores in a single cluster, with a single block of L3 cache.

Increasing the cores per CCX has very real latency, complexity, and power consequences. X Elite cores are clustered in groups of 4 for a reason.
 

Abwx

Lifer
Apr 2, 2011
11,884
4,873
136
NBC does power and perf/watt measurements for the whole system, which greatly obscures the differences in SoC power/efficiency.

Dubbious methodology, beside the 370H has an OLED screen while the 185H has an IPS, in ST where power is lower this has some impact, as well as in battery life, so all in all Strix numbers are quite good.
 

MarkPost

Senior member
Mar 1, 2017
378
794
136
From Phoronix's 100+ Linux benchmark test:

These initial Ryzen AI 9 HX 370 results have me super excited for Zen 5. It's different having AMD shipping their next-gen laptop SoCs ahead of the desktop processors, but in a few weeks attention will turn to the Ryzen 9000 series. I've been pushing the limits of this SoC in being eager to test Zen 5, so stay tuned for more articles ahead of the Ryzen 9000 series launch. The generational performance uplift is nice with Zen 5 but really captivating my interest has been the large power efficiency gains at least among these laptop SoCs. The Ryzen AI 9 HX 370 also furthers its lead over Intel Meteor Lake that makes an even larger hill for Lunar Lake to climb, which will be interesting to see how that battle goes later in the quarter.
From the same review, I think this is a great achievement:

Where things got really interesting with the Ryzen AI 9 HX 370 was the power efficiency of this Zen 5 laptop SoC. Across the span of all the benchmarks, the Ryzen AI 9 HX 370 SoC was pulling about 20.4 Watts with a peak of 34.2 Watts.... Meanwhile the Ryzen 7 7840HS had an average of 35 Watts and a peak of 60 Watts. The Ryzen 7 7840U had a 27 Watt average and a peak of 51 Watts. The Ryzen AI 9 HX 370 came out faster than those parts while consuming significantly less power. This Zen 5 power efficiency is very exciting and should carry over for desktop and server parts too. Meanwhile the Core Ultra 7 155H was consuming 29.6 Watts on average with a peak of 65 Watts.
 
  • Like
Reactions: lightmanek