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

[DisEnchantment]

Speculate at will

 

[RnR_au]

Maybe AMD is simply acknowledging that the 3D vcache models are the gaming CPUs and non are primarily work focused.

The interesting thing is that there are multiple reviews now that calls the 9600X the best budget gaming cpu. Its weird how some reviewers are calling the release a flop. According to some, AMD will have the best value gaming cpu and once the X3D parts arrive, the best gaming cpu bar none.

 

[DisEnchantment]

I think and still do that STX Halo is the best thing to come out Zen5 family. Zen5 core is no longer stuck to dual channel but quad channel for client and yeah RAM is soldered but its fast RAM.

I would be inclined to think that way too.

Z5 L3 is fairly big, so being too mem BW limited in many scenarios would indicate prefetching is not working very well or at least prefetching is tailored to address other workloads. So it would need fatter caches.
One evidence of this is that 7800X3D has higher SIR2017 1T score than 7700X at same frequency, around 6% faster (then it hits some other bottleneck likely)

I would imagine having a bigger L2/SLC would alleviate some of this.
However for Strix Halo, having LLC which can prefetch and keep some of the data ready, overhauling the SDF for improved latency and throughput and the Octa channels of LPDDR5X would really help with bandwidth and memory parallelism.

Strix Halo is going to be one interesting part.

 

[poke01]

Nice, reviews are up. But where is the efficiency?

View attachment 104773

thats what HUB also tweeted.

https://twitter.com/x/status/1821436981328654462

Zen 5 is a crazy good server core and thats where AMD's focus is

 

[Evleos]

The part you all should read:

 

[CouncilorIrissa]

Nice, reviews are up. But where is the efficiency? What's up with "ground-up" re-design... wait for Zen6? )

View attachment 104773

I guess it depends on the workload selection and the OS used.
Phoenix's review has by far the largest workload selection of all reviews and when comparing the 7700 to 9700X, the latter has a 1W higher average power (73W vs 72W) and a +20% higher geomean across all benchmarks (90 vs 75).

 

[Joe NYC]

Now reading through so might be repeating.

Will be interesting to see how Strix Halo with it's 256b memory bus work on CPU tasks?

comparing 4 channels of 8533 vs 2 channels of 5600 is 3x the bandwidth. And the Strix Halo packaging should be able to give each CCD full bandwidth.

 

[HurleyBird]

Yes, and some of the ones they throw at desktop CPUs are hella obscure with peculiar performance profiles. You really have to know your stuff to read a full Phoronix review of a desktop CPU and understand what information is on display. Their server CPU tests are much more concise.

Maybe you need to know your stuff to interpret what any one particular result means, but if you're just looking at the average it really doesn't matter. When it comes to distilling performance to a single number, regression to the mean is real, and sample size is king.

 

[coercitiv]

Zen 5 is a crazy good server core and thats where AMD's focus is

Every time I open the thread this sentence comes up in one shape or another.

All Zen cores were made for servers. No Zen core was made for consumers and just happened to work well in servers too.

 

[Joe NYC]

There were rumours that V-cache in Ryzen 9000X3D is different.

But nothing concrete in what way it would be different. No clock regression would be the most obvious guess, but that does not make the V-Cache different, it makes the CPU with V-Cache different.

The biggest game changer would be if V-Cache covered the whole CPU, and AMD switched to Wafer over Wafer packaging. This would turn V-Cache models into mainstream, high volume parts. Also, covering the whole CPU die could double the V-Cache size.

 

[Joe NYC]

At this point my biggest worry is that AMD already hyped up Zen5 vcache a bit saying it had some “interesting” features

Based on what we’ve seen so far of the core I’d expect it to have even more uplift than Zen4X3D vs Zen4, but if we apply the “reverse marketing lingo” filter, “interesting” doesn’t sound too good….

It would be disappointing if after 4 years, 3rd generation, if AMD still remained stalled at the same place with V-Cache, which is in one of AMD's most promising and differentiating technologies.

 

[misuspita]

MinisForum has a brand new MiniPC using HawkPoint CPU, and adds some nice IO features.

The advantage of HawkPoint vs. either StrixPoint or Kraken is that it has 8 full cores and there is no need for Windows 11 for improved scheduling of dense cores.

Best one is the beelink, looking nice and very quiet. But yeah, if they manage to cram a Strix Halo in a 15cmx15cm device, I'd instabuy if it's under 1500€ barebone... Sort of, as I understand there's gonna be soldered memory, no CAMM

 

[Timmah!]

66% IPC gain!

666 percent in fact. Threadripper renamed to Soulripper, your soul being the currency you have to pay to obtain one.

 

[Joe NYC]

Best one is the beelink, looking nice and very quiet. But yeah, if they manage to cram a Strix Halo in a 15cmx15cm device, I'd instabuy if it's under 1500€ barebone... Sort of, as I understand there's gonna be soldered memory, no CAMM

Strix Halo would be fantastic part for Mini PCs, except the price will be a bit higher than current Mini PCs, but hopefully more like $1,000 for pricing with ~32 to 64 MB of memory on board.

 

[HurleyBird]

The part you all should read:

View attachment 104774

It's not a technical thing, but this snippet is an interesting peek into AMD infosec:

The full capability of these 2 extra ALUs was not revealed in the GCC patch which says that multiply is still 1/cycle. When asked about this, AMD told me this error in the GCC patch was intentional to avoid leaking the true capability of Zen5 before they were ready to reveal it.

 

[misuspita]

I'm in no rush, I have a 5700G completely silent now and I need something big to get me to move from it, no 30% improvements, I need much more...

And I reckon STH with 16 cores and 4070m perf is what doctor ordered...

 

[Hans Gruber]

It would be disappointing if after 4 years, 3rd generation, if AMD still remained stalled at the same place with V-Cache, which is in one of AMD's most promising and differentiating technologies.

Well said. Intel said we can just add cache to our chips as well. It's not hard to do.

 

[yottabit]

The part you all should read:

View attachment 104774

But what should our takeaway be? That’s it’s terribly membw starved and can’t see those gains in practical application? Yay?

 

[FlameTail]

The biggest game changer would be if V-Cache covered the whole CPU, and AMD switched to Wafer over Wafer packaging. This would turn V-Cache models into mainstream, high volume parts. Also, covering the whole CPU die could double the V-Cache size

That would be incredible!

 

[yuri69]

It would be disappointing if after 4 years, 3rd generation, if AMD still remained stalled at the same place with V-Cache, which is in one of AMD's most promising and differentiating technologies.

AMD still uses the good 'ol substrate from their K10 MCM days. Their SerDes has been more or less the same since Zen 1. The IF topology is the same since Zen 2. The IF link bandwidth stagnates since Zen 3... :/

 

[poke01]

Strix halo will be great for creators, specifically video editors but does one do encoding on a CPU?

edit: added extra context

 

[yottabit]

I don’t think anyone can be saltier than Zen3 owners like me, who abstained from upgrading their 5950x because of all the rumors Zen5 would be such a monster and the most exciting dream Mike Clark ever had. I remember thinking “wow a ~40% performance upgrade with Zen4 would be great, but combined with another 32% from Zen5 would be awesome, almost doubling my current performance so I should just wait”

Part of my brain knew it couldn’t be such a big jump 2 generations in a row since Zen4 really knocked it out of the park, and Zen5 would be a similar node. But the stupid neanderthal part of my brain gave in to the rumor hype. I wish I could take a time machine back and buy a 7950x on launch

From a pure techie perspective though the new architecture is exciting, and I’m interested to see what it can do with the memory bw uncorked (Turin, STX Halo, Zen5 Threadripper if there is one) and we’ll see regarding the X3D chips but my expectations are pretty low

 

[yottabit]

There’s a few additional things I’d like to highlight:

1) Gamers Nexus said their first 9600x sample just… didn’t work. in a variety of boards they tested. They were later sent a second one that worked fine. Should we be worried about that?

2) Roman found on his channel that with unlimited PBO the 9700x drew similar power and hit similar temps to Zen4. So were there really IHS/solder improvements or was it just the reduced TDP? We’ll see with 9950X reviews I guess

 

[PJVol]

thats what HUB also tweeted

It would be nice if TPU added the 9700X configured as 105W TDP (105W cTDP or 142W PPT) to the comparison instead of "PBO maximum" to see the CPU V/F behavior over a wider range. May be some reviewers did that (haven't gone through all the reviews yet).
But even beyond that, it feels like something is wrong with the desktop Zen5's internal power/performance management.

 

[carancho]

MLID absolutely nails the analysis

And also has by far the best thumbnail of the lot

 

[tsamolotoff]

From Scatterbencher OC guide, another very weird decision of AMD that complicates non-static OC:

The difference isn’t just the factory-fused voltage-frequency curve, but also the Fmax: CCD0 has a programmed Fmax of 5750 MHz whereas CCD1 has a programmed Fmax of 5450 MHz.