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

[DisEnchantment]

Speculate at will

 

[Tuna-Fish]

Will the chipset actually make a difference for memory speeds? I've heard both, but I thought the CPU memory controller was the limitation.

The chipset isn't really involved at all, but the quality of the traces on the motherboard matter a lot for achieving high memory speeds, and historically the minimum acceptable spec for them has been improved with new chipsets.

 

[9949asd]

That's the sad state of tech journalism today... Everything is sensationalized and WTFtech led the way in the 2010s.

That explains the really good temps, then. There's a bunch of people on Xitter who are reacting to the WTFtech/Videocardz articles about the info you shared, but those articles don't do a good job of clarifying why the temps are so good. It's leading people to incorrectly believe that Zen 5 is super easy to cool and/or super power efficient.

For example, from the Videocardz article, just no mention of cooling method whatsoever.
View attachment 102915

Same with WTFtech:
View attachment 102916

So often authors just steal whatever info is eye-catching for their article but fail to represent the source faithfully because they don't bother reading the rest of the thread... SMDH.
View attachment 102918

Unbelievable is right.

230w PPT oc idle 100w?! And core temp only 23c? That’s definitely a sub ambient temperature cooling.

 

[tsamolotoff]

Dissapointing climax.

Custom loop is not bounds and leaps better than typical AIO at keeping temperatures low for such a short benchmark, so it'd be the same with any other cooling solution (provided its coldplate is not U shaped like most Intel-optimized setups are). Also, you don't have to buy expensive german parts, I make do with full 100% chinese setup (a knockoff mora, some cheap threadripper-adapted waterblock with a finned coldplate that covers whole CPU etc), and it wouldn't be that much worse as compared to legit MORA/ XSPC Raystorm etc.

 

[tsamolotoff]

230w PPT oc idle 100w?!

If you tweak powerplan a bit and there is browser open in the background, idle cpu power consumption is around 60W for me personally (20W of which is SoC)

 

[gaav87]

Custom loop is not bounds and leaps better than typical AIO at keeping temperatures low for such a short benchmark, so it'd be the same with any other cooling solution (provided its coldplate is not U shaped like most Intel-optimized setups are). Also, you don't have to buy expensive german parts, I make do with full 100% chinese setup (a knockoff mora, some cheap threadripper-adapted waterblock with a finned coldplate that covers whole CPU etc), and it wouldn't be that much worse as compared to legit MORA/ XSPC Raystorm etc.

Next we will hear that the "daily OC/PBO+ CO tunning" is -30CO max fclk with high vsoc/vddg and mem oc+tertiary timings tuned.
The high W on the soc was suspicious.

 

[Hail The Brain Slug]

Next we will hear that the "daily OC/PBO+ CO tunning" is -30CO max fclk with high vsoc/vddg and mem oc+tertiary timings tuned.
The high W on the soc was suspicious.

It's higher than my increased voltages and tuned tertiary 6400/2133fclk setup on the 7950X, so he's done something really bad.

 

[Det0x]

It's higher than my increased voltages and tuned tertiary 6400/2133fclk setup on the 7950X, so he's done something really bad.

You guys know both soc voltage and soc powerusage is showing in the screenshots ?
ES is ES for a reason i guess

 

[Hail The Brain Slug]

You guys know both soc voltage and soc powerusage is showing in the screenshots ?
ES is ES for a reason i guess

There's no reason the IOD would be ES quality silicon since it's reused from Raphael. His VSOC looks decent. I imagine it's other voltages HWINFO isn't showing he has cranked.

 

[soresu]

I think it's safely bet that Zen5 has larger realworld IPC uplift than Zen2-Zen3 iteration at least in FP performance

Given they just doubled FP/SIMD throughput for Zen2 they wouldn't have done hslf so much for Zen3.

Zen5 is much like Zen2 in that regard at least.

 

[PJVol]

Looks like 9950X does not have any sweet spot for perf/watt. It keeps gobbling up any additional Watts given and turns it into perf.

Blender 3.3 is a relatively moderate Cac workload to draw any conclusions

Hmm. Still don't really know what you mean.

Global limit is 5.85 GHz, same as the 7950x. The L3 cache runs at the same frequency as the cores for Ryzen CPUs, so there's no separate limit.

The global frequency limit is not a static metric*, look at any of the 7950X/X3D HWInfo screenshots running Blender. IIRC L3 limits are not taken into account when calculating it, so the actual effective clocks are often below the global frequency limit.
But my point was just that HWInfo reporting the actual Global Frequency limit wrong on that Zen5 PC (in fact it's reporting CCLK limit instead)
------------
* it's aggregated by PID controller from a set of actual individual infrastructure and other limits

 

[Hitman928]

Blender 3.3 is a relatively moderate Cac workload to draw any conclusions

The global frequency limit is not a static metric*, look at any of the 7950X/X3D HWInfo screenshots running Blender. IIRC L3 limits are not taken into account when calculating it, so the actual effective clocks are often below the global frequency limit.
But my point was just that HWInfo reporting the actual Global Frequency limit wrong on that Zen5 PC (in fact it's reporting CCLK limit instead)
------------
* it's aggregated by PID controller from a set of actual individual infrastructure and other limits

Ok, I get what you are saying.

 

[SmokSmog]

 

[tsamolotoff]

I initially thought those clocks / voltages were screenshotted at idle, but now i've looked at the picture again and realized it was taken while the test was running. Keeping the CPU around 60C at 1.1V with good custom loop isn't that hard, so no need for exotic cooling solutions, be it 50w or 300w. Also, as cores became wider and bigger in area, it's probably easier to sustain good thermal energy transfer as compared to Z4 (especially if they've replaced whatever thing they've used with proper solder)

 

[adroc_thurston]

especially if they've replaced whatever thing they've used with proper solder

It was already proper solder.

 

[cebri1]

a very large custom cooling rig with a optimus signature v3 block together with a MO-RA3 radiator

I feel humiliated right now

 

[tsamolotoff]

It was already proper solder.

With 10-15C variance between cores on the same CCD? It seems there were big bubbles or its application to the chiplet is very uneven.

 

[therealmongo]

I am jealous. If I enable PBO Windows stops booting, even if i touch no other settings.

Changing anything except curve optimizer is futile.

AMD said the chips run cooler. Zen 4 will run cooler if you tweak it. Enabling CO alone dropped my chip 10 degrees with a slight performance uplift.

Zen 4 can do it as well, you just need a recent BIOS. Many on OCN have achieved it. If I had the money to burn I would pick up a better kit that could handle 8000, because mine only does DDR5-6200. Alas I currently have other priorities.

Hopefully, im one of the guys on OCN who had ddr4 running at 4133/2066 mhz "stable" (not that I used it for 24/7 settings, lol) and have a number of entries in the Google sheet for DDR4 overclocking (mongoled), so hope so see 2:1 running over spec on Zen5

 

[PJVol]

With 10-15C variance between cores on the same CCD? It seems there were big bubbles or its application to the chiplet is very uneven.

It's just the thermal superposition effect typical of multi-hotspot packages.
In your case cores #0 and #9 (or #15) are physically located at the CCD corners on the far side from the IOD and are less susceptible to overheating due to interaction between hot spots.

 

[igor_kavinski]

60W: https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41246800
90W: https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41248191
120W: https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41248591
160W: https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41249476
230W: https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41250270

 

[tsamolotoff]

It's just the thermal superposition effect typical of multi-hotspot packages.

It disappears if the substance between IHS and silicon is removed and replaced either with direct die cooler or liquid metal, so I doubt it.

 

[PJVol]

@igor_kavinski
Why 9999W and not 256W?

It disappears if the substance between IHS and silicon is removed and replaced either with direct die cooler or liquid metal

No.

 

[gaav87]

230W: https://forums.anandtech.com/thread...ranite-ridge-ryzen-9000.2607350/post-41250270

Delided ?

 

[Hail The Brain Slug]

I had a dream I was reading Zen 5 reviews and they weren't even at the top of the chart for application performance, much less gaming. Zen -5% confirmed.

 

[Josh128]

So now the biggest question is, how are the final tuned /launch BIOS voltages and powers compared to what we see here. If "daily CO" = max neg offset, thats going to be very disappointing.

 

[igor_kavinski]

Delided ?

Why would someone delid an ES, especially if they are required to give it back to AMD?