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

Page 726 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

CouncilorIrissa

Senior member
Jul 28, 2023
541
2,121
96

Rigg

Senior member
May 6, 2020
540
1,273
136
The TPU review has the 7700 using slightly less power pretty much across the board. Both are using more than 60W in CB 2024 nT.

power-applications-compare-vs-7700.png

power-games-compare-vs-7700.png
 
  • Like
Reactions: Tlh97 and poke01

Abwx

Lifer
Apr 2, 2011
11,557
4,349
136
I completely doubt that graph.

We ll have to wait for confirmations, what is odd is that Computerbase had the 9700X idle power at 27W, wich is excessive, and if the CPU use only 60W then it means that the cores are fed with only 33W.

On the other hand the 7700 use about the same power but they measured its idle power at 13-14W, wich would amount to 46W for the cores and hence 40% more cores power than the 9700X.

What is even more weird is that Hardwareluxx measured 14W idle power but got the same CB 2024 score, wich after all is possible since there s enough power headroom left for the Computerbase sample, but that would mean that Hardwareluxx sample did use only 47W overall in CB 2024, wich is dubbious, so possibly that we ll have revised reviews next week when they test the bigger boys.
 

Rheingold

Member
Aug 17, 2022
55
151
76
We ll have to wait for confirmations, what is odd is that Computerbase had the 9700X idle power at 27W, wich is excessive, and if the CPU use only 60W then it means that the cores are fed with only 33W.
No, the 60W number is for the cores, without SOC. Add the 27W and some change for the correct 88W PPT in total.

And yes, the 27W indeed seems excessive, considering they used RAM at JEDEC timings, so the memory controller should not be in OC mode.
 

poke01

Platinum Member
Mar 8, 2022
2,205
2,803
106
No, the 60W number is for the cores, without SOC. Add the 27W for 87W in total.
yep. Look at PC Watch, it adds up. No need to wait.
1723249970768.png
1723250089788.png
 
  • Like
Reactions: lightmanek

Abwx

Lifer
Apr 2, 2011
11,557
4,349
136
The TPU review has the 7700 using slightly less power pretty much across the board. Both are using more than 60W in CB 2024 nT.

power-applications-compare-vs-7700.png

power-games-compare-vs-7700.png
They measure power at the 12V CPU power rail, so that include the VRMs losses wich are about 10-12%, so their CB run is at 65W or so, that s comparable to Computerbase s 60W.
No, the 60W number is for the cores, without SOC. Add the 27W and some change for the correct 88W PPT in total.

And yes, the 27W indeed seems excessive, considering they used RAM at JEDEC timings, so the memory controller should not be in OC mode.
That s the full Soc power since they get about 90W peaks in Blender, previously they always used the CPU package power, wich include the uncore.

Indeed their methodology is curious, they measure the power with Blender at mid test, peak value and minimal value, there s something flawed here because they get the 7700 minimal power at 24W while in all their previous reviews it was about half this value.
 

Rigg

Senior member
May 6, 2020
540
1,273
136
They measure power at the 12V CPU power rail, so that include the VRMs losses wich are about 10-12%, so their CB run is at 65W or so, that s comparable to Computerbase s 60W.

That s the full Soc power since they get about 90W peaks in Blender, previously they always used the CPU package power, wich include the uncore.

Indeed their methodology is curious, they measure the power with Blender at mid test, peak value and minimal value, there s something flawed here because they get the 7700 minimal power at 24W while in all their previous reviews it was about half this value.
Screenshot from 2024-08-09 20-16-03.png
 

inquiss

Senior member
Oct 13, 2010
200
274
136
Who knows, we'll see. It's just an option, but something will be needed anyway. 4-5 years between Zen4 and Zen6 with nothing else than what we've seen so far just won't do. That is unless we'll see substantial price cuts along the way.
They won't because it doesn't have enough memory bandwidth. X3D and some APUs probably. X3D could have more memory stacks as a potential option for killing it at gaming.
 

Abwx

Lifer
Apr 2, 2011
11,557
4,349
136
View attachment 104969
They provide a total system power of 208W but no idle system power number, they should be discarded as they dont help at all, beside it seems that they are not that competent at measuring power :
Correction (Aug 7): There was an error in our original power graph showing how the 9700X behaves in CB (208 W is the correct figure, not 221 W)

We would need the idle power to check the delta, what seems sure is that Cinebench R23 fully load the CPU up to 88W and with a 4.4GHz frequency according to Hardwareluxx.

Cinebench 2024 could be a lighter load, this could explain Techspot s 4.8GHz but then at stock Computerbase has only 4.3-4.4GHz in CB 2024, so it s obvious that numbers are all over the place and problematic to sort out.

Also i noticed that Techopowerup "updated " their benches and "unexpectedly" removed a bench, so their averages are actually biaised, here the 14900K review with the browsing perfs :


And FI the 9600X review for the same perfs in browsing :


Hey, what happened to WebXPRT4 ??

Let see the scores in a CZ review with overclockings for all CPUs to check if there s not something too "bothering" along with Computerbase stock results :
 

Attachments

  • AMD-Ryzen-9600X-9700X-WebXPRT4.Png
    AMD-Ryzen-9600X-9700X-WebXPRT4.Png
    30.6 KB · Views: 32
  • Screenshot 2024-08-10 at 07-07-00 AMD Ryzen 5 9600X und Ryzen 7 9700X im Test Benchmarks in ne...png
    Screenshot 2024-08-10 at 07-07-00 AMD Ryzen 5 9600X und Ryzen 7 9700X im Test Benchmarks in ne...png
    103.7 KB · Views: 33
Last edited:

Rigg

Senior member
May 6, 2020
540
1,273
136
They provide a total system power of 208W but no idle system power number, they should be discarded as they dont help at all, beside it seems that they are not that competent at measuring power :


We would need the idle power to check the delta, what seems sure is that Cinebench R23 fully load the CPU up to 88W and with a 4.4GHz frequency according to Hardwareluxx.

Cinebench 2024 could be a lighter load, this could explain Techspot s 4.8GHz but then at stock Computerbase has only 4.3-4.4GHz in CB 2024, so it s obvious that numbers are all over the place and problematic to sort out.

Also i noticed that Techopowerup "updated " their benches and "unexpectedly" removed a bench, so their averages are actually biaised, here the 14900K review with the browsing perfs :


And FI the 9600X review for the same perfs in browsing :


Hey, what happened to WebXPRT4 ??

Let see the scores in a CZ review with overclockings for all CPUs to check if there s not something too "bothering" along with Computerbase stock results :
Bruh....... The 9700X should pull its 88W PPT limit (as indicated by HWI64) in CB 2024. This is entirely consistent with all of the 65W TDP AM5 CPU's. This isn't a mystery that requires autistic levels of investigation.

136607.png


AMD%20Ryzen%207%209700X%20vs%20Ryzen%207%207700%20Power%20analysis_575px.png
 

sl0519

Junior Member
Aug 10, 2024
21
54
51
Zen 5 isn't what we had hoped for from the gaming perspective, but to say "Zen 5" is a flop suggested by some journalists (HUB) is a bit of a stretch.

Sure if you compare it directly with the non X 7000, the efficiency claim is no longer the case. But what if AMD decides to put out a non X 9000 version with let's say, 45w? Wouldn't that "reshuffle" the whole lineup and thus making the claim AMD is trying to hide reviewers from making comparisons with the non X 7000 totally invalid?

What if you overclock the 9000 to 105w TDP then compare them again in terms of performance and efficiency? Wouldn't you gain back the 10 ~ 15% performance uplift people were expecting to see? (applications not gaming I know)

Why haven't I seen any reviewers plotting performance graphs from different wattages to really see where the efficiency gains from Zen 5 is really at? Yes efficiency is certainly thrown out of the window the higher the wattages but maybe at higher TDP is where Zen 5 really excels at. We need EFFICIENCY CURVE against zen 4 parts to know for sure if zen 5 is a flop or not. Those clickbait YT thumbnails is too quick of a judgement because we haven't seen how the higher end Zen 5 performs.
 

poke01

Platinum Member
Mar 8, 2022
2,205
2,803
106
What if you overclock the 9000 to 105w TDP then compare them again in terms of performance and efficiency? Wouldn't you gain back the 10 ~ 15% performance uplift people were expecting to see? (applications not gaming I know)
Yes but then power consumption increases massively. Efficiency goes out the window.
 

jdubs03

Senior member
Oct 1, 2013
880
523
136
I feel like a computerbase did a good job comparing the different skus.
I set the 7700 as a baseline because that was a 65W TDP part and what HUB was using as a comparison. And while yes the 9700X at 65/88 looks good next to the 7700X at 105/142. When compared to the 7700 it’s not overly impressive. 9% better in multi-core. 19% better in single (which is by far its best quality), and 12% better in gaming.

I did read earlier that adjusting the memory timings and a special feature on MSI boards helps gain some FPS. So that helps. But just going by stock, the gains are quite a bit less than the ~24% price premium to the 7700X.

And most people are looking through this at the lens of gaming performance when the X3Ds are more suitable for that. I’d argue that the X versions are more productivity focused. And from that angle it’s an even tougher sell.

$330 would be a fair, more appropriate price for the performance being provided.
 

Attachments

  • IMG_0836.jpeg
    IMG_0836.jpeg
    465.7 KB · Views: 25
  • IMG_0837.jpeg
    IMG_0837.jpeg
    537.1 KB · Views: 18
  • IMG_0835.jpeg
    IMG_0835.jpeg
    510 KB · Views: 33

Rigg

Senior member
May 6, 2020
540
1,273
136
I feel like a computerbase did a good job comparing the different skus.
I set the 7700 as a baseline because that was a 65W TDP part and what HUB was using as a comparison. And while yes the 9700X at 65/88 looks good next to the 7700X at 105/142. When compared to the 7700 it’s not overly impressive. 9% better in multi-core. 19% better in single (which is by far its best quality), and 12% better in gaming.

The computerbase results are skewed by their insistence on using JEDEC memory at "officially supported" speeds. It doesn't invalidate their results, but the gaming results in particular aren't consistent with how most reviewers compare CPU's. It's also not consistent with how most DIY PC builders run their systems. In reality, The gap in gaming between 7700(X) and 9700X is more like 5% (depending on the games tested) with both running @ 6000 MT/s.
 

B-Riz

Golden Member
Feb 15, 2011
1,595
761
136
I pointed that out earlier in the thread as well:





Overall, I'm inclined to blame AMD for not giving motherboard makers and/or reviewers better guidance regarding vSOC and other uncore voltages as it appears there is more than the expected variance between mobo makers. 10% MT perf difference is a lot... that's basically minus one generation in performance.

Obviously, I'm expecting less impact on 9950X results simply because an extra 10W uncore usage will have much less impact when your TDP is 170W vs 65W.

So the funny thing with SOC voltage coming up in this thread is that I don't recall many of us on the forum actively talking about tweaking that in the Zen3 and Zen4 builders threads; more something that was mentioned to lower if your boards BIOS was pushing too much.

I just lowered the 5900X and 7800X3D SOC volts in the past few months and it has made a huge difference in temp and power usage according to HWiNFO64; I am under 1.1 on both and can pass a prime95 24hr stress test. Any recent instability from tweaking has come from lowering curve offset too much...
 
  • Like
Reactions: lightmanek

biostud

Lifer
Feb 27, 2003
18,701
5,432
136
It will be interesting to see how well 9900X and 9950X move, but this is shaping up to be a terrible launch with regards to volume.

Really striking that Zen 4 sold that well despite $500 minimum motherboard, $300 minimum ram, and a brand new unstable platform sure to cause you hair loss.

Zen 5's poor sales are solely due to the pricing and product delivered.
Even if the 9700X cost the same as the 7700X would you even bother swapping it?