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

[DisEnchantment]

Speculate at will

 

[Glo.]

And this 8950HS with 130/2400 pts is supposedly what? Strix Halo?

For comparison:
CB 2024
Ryzen 9 7950X -> 125/2185 pts

Not much better in ST If true.

Its 35W ES of 4 Zen5+8 Zen5C with 16 CU design.

35W ES, supposedly.

IF this is true its running circles around Zen 4 in terms of perf/watt.

 

[BorisTheBlade82]

Please don't forget that for Zen5 quite likely applies a similar difference in peak frequency, as for Zen4 Desktop vs. Mobile. There it is 5.7Ghz vs. 5.2Ghz or almost 10%.
So a lot of unknowns still even if those numbers have any linkage to reality.

 

[adroc_thurston]

IF this is true its running circles around Zen 4 in terms of perf/watt.

I mean yea, Turin is a thing.

Please don't forget that for Zen5 quite likely applies a similar difference in peak frequency, as for Zen4 Desktop vs. Mobile.

STX1 should be clocked roughly the same as PHX1 so it's a nice comparison point.

 

[H433x0n]

Damn you are quick. Anyway 7950x official score is 2919 from here: https://browser.geekbench.com/processors/amd-ryzen-9-7950x

The 8950X late ES got 3500 roughly, so about 20% ST uplift. Very disappointing to me to be honest.

I feel like he’s just guessing anyway, wouldn’t read too much into it yet.

 

[soresu]

The 8950X late ES got 3500 roughly, so about 20% ST uplift. Very disappointing to me to be honest.

Keep in mind that early BIOS releases after a major µArch change tend to need time to settle before you can really judge the performance as fully accurate to the HW potential.

 

[H433x0n]

Keep in mind that early BIOS releases after a major µArch change tend to need time to settle before you can really judge the performance as fully accurate to the HW potential.

If it truly hits 3500 ST in GB6 that’s a good result IMO. If it improves on that I’d be pretty bullish for the release.

 

[soresu]

Its 35W ES of 4 Zen5+8 Zen5C with 16 CU design.

35W ES, supposedly.

IF this is true its running circles around Zen 4 in terms of perf/watt.

True to a point, but without core clock context for comparison it doesn't mean much.

Certainly encouraging though.

 

[jpiniero]

All I'll say is that the PL1 is likely higher than 35 W (all of NBC's reviews of HS models were well above that)... but that's a real nice score if legit.

 

[inf64]

If it truly hits 3500 ST in GB6 that’s a good result IMO. If it improves on that I’d be pretty bullish for the release.

Versus 7940H it's exactly 32% faster in GB6 ST test (7940H scores 2616pts @ 5.2Ghz). If adroc is correct and ST max boost is similar to Zen 4 mobile parts (~5.2Ghz), we have a nice confirmation of IPC over 30% in some tests.

 

[igor_kavinski]

What tickles my funny bone is the lengths Intel will go, to prevent Zen 5 ending 2024 as the undisputed x86 GB6 ST champ

(without exotic cooling)

 

[Glo.]

What tickles my funny bone is the lengths Intel will go, to prevent Zen 5 ending 2024 as the undisputed x86 GB6 ST champ

(without exotic cooling)

What do you mean? UserBenchmark will declare Intel the one and only winner of CPU comparison. Regardless of reviews, and scores.

 

[TESKATLIPOKA]

Its 35W ES of 4 Zen5+8 Zen5C with 16 CU design.

35W ES, supposedly.

IF this is true its running circles around Zen 4 in terms of perf/watt.

Seriously?

4*Zen5+8*Zen5c 8950HS(>>35W) ES: 130/2400 pts
12*Zen4 R9 7900(65/88W): 107/1419 pts
Result: 21.5% better ST and 69% better MT.
I can still accept ST gain, but that MT is impossible considering It has the same amount of cores.

 

[poke01]

So I think Meteor Lake is going to beat AMD Strix and M3!

Then I woke up...

 

[Glo.]

Seriously?

4*Zen5+8*Zen5c 8950HS(>>35W) ES: 130/2400 pts
12*Zen4 R9 7900(65/88W): 107/1419 pts
Result: 21.5% better ST and 69% better MT.
I can still accept ST gain, but that MT is impossible considering It has the same amount of cores.

It must be that Mahogany Crankshaft, or something.

 

[inf64]

So I think Meteor Lake is going to beat AMD Strix and M3!

Then I woke up...

Let's see MTL beat Raptor Lake and Zen 4 mobile parts in performance. That's a tall order as it is, no need to bring in Zen 5 juggernaut to the fight.

 

[Glo.]

To be honest with you guys.

I would NOT be surprised if that 2400 MT Cinebench 2024 score was actually 1400 for those ESes .

And somebody simply made a typo.

 

[Abwx]

Could be 2400 pts at something like 80-100W, the 7950X use about 160W in CB 2024.

 

[TESKATLIPOKA]

Could be 2400 pts at something like 80-100W, the 7950X use about 160W in CB 2024.

That would still be ridiculously good for something with only 12 cores.

 

[Joe NYC]

Halo was never monolithic. Strix halo is not coming to desktop ever, unless AMD caves in to the demand of all 10 people who'd actually pay premium price for this chip on AM5.

I don't think it is possible to release Strix Halo to Socket AM5.

First, if it just uses external DDR5 memory, it loses half of the bandwidth, meaning roughly half of the graphics performance.

Second, Socket AM5 is too small to house a chip with internal 4x LPDDR5 memory chips.

 

[PJVol]

So far the MT uplift doesn't look good for the desktop part, and it becomes clearer what the rumors about “prioritized ST performance” meant )

 

[soresu]

Laptops. Duhhhhhh.

AMD doesn't see a market for APUs. Hence no monolithic APU on AM5 yet.

There's plenty market for APUs in embedded SFF desktops a la NUCs, just not socketed systems.

NUCs are basically laptop minimalist physical system design for the desktop market.

Since Renoir there have been plenty of NUC type customers for AMD APUs given the number of such devices on Amazon, plainly AMD sees the market there as they almost certainly have some assistance from AMD in designing their mainboards to spec.

I'd be down for an Atari VCS like thin NUC with Strix Halo as a Steam console 😎

If rumours hold on it coming in 2025 that would coincide with 2nd gen AirJet products and we could see some interesting compact systems come out of that.

 

[soresu]

I don't think it is possible to release Strix Halo to Socket AM5.

First, if it just uses external DDR5 memory, it loses half of the bandwidth, meaning roughly half of the graphics performance.

Second, Socket AM5 is too small to house a chip with internal 4x LPDDR5 memory chips.

Current LPDDR5 specs are getting better at 9600 MT/s even without the full benefit of full fat DDR5 latencies.

Combined with the benefits of CAMM for future smol systems it should be enough to do OK with 40 CUs, especially if SH also has a little ♾ cache too.

 

[Timmah!]

To be honest with you guys.

I would NOT be surprised if that 2400 MT Cinebench 2024 score was actually 1400 for those ESes .

And somebody simply made a typo.

Except in the vid that table says 2K and not 2400 for 8950HS (35W). 2400 is supposed 8950x.

Its still curious anyway. So 4 Zen5 + 8 Zen5c cores (thus not even 12 full fat Zen5 cores) provide performance equal to 83 percent of 16 full fat Zen5 cores in CB24 MT. I mean, you would expect 12 full cores to provide 3/4 of that performance. This is not even that and still performs better.

 

[Goop_reformed]

Versus 7940H it's exactly 32% faster in GB6 ST test (7940H scores 2616pts @ 5.2Ghz). If adroc is correct and ST max boost is similar to Zen 4 mobile parts (~5.2Ghz), we have a nice confirmation of IPC over 30% in some tests.

Then how do you explain 7950x w/ 2919 vs 8950x w/ 3500 ?

 

[Abwx]

I mean, you would expect 12 full cores to provide 3/4 of that performance.

Only if power is 0.75x as well, with 12C you can clock the chip 10% higher or so at same power than a 16C.

0.75 x 1.1 = 0.825