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

[DisEnchantment]

Speculate at will

 

[Hitman928]

Normally, yes, but when Ian Cutress says "due to a typo" that would be box packaging and labeling, not what you are speaking about.

It’s a joke. . .

 

[yottabit]

Is there a chance Strix Halo would outperform 9950X for membw bound workloads?

Does anyone happen to have numbers for theoretical membw for Strix Halo vs 2:1 DDR5 ~8000 on Zen5 DT?

 

[Rheingold]

My best guess is that the Yields just aren't there. 5.4 and 5.5GHz kinda works, so the 2 small SKUs can Release on 8th, but the number of 5.6 and 5.7GHz CCDs is so low that it would've been a paper launch.

Going from how omnipresent the 9900X is in marketing and product listing leaks, and the first early review also being on a 9900X, I very much doubt that.

 

[Makaveli]

Sadly, the only ones rejoicing at AMD's recall are Intel users who get more time to NOT get depressed upon seeing Zen 5 benchmarks.

With all the 13 and 14th gen degradation stories that are going on they have enough to worry about with their own rigs to be concerned with AMD.

 

[CouncilorIrissa]

Is there a chance Strix Halo would outperform 9950X for membw bound workloads?

Does anyone happen to have numbers for theoretical membw for Strix Halo vs 2:1 DDR5 ~8000 on Zen5 DT?

LPDDR5x @ 8533 MHz with 256-bit interface is 273 GB/s.

 

[Rheingold]

That’s not what hardwareluxx is reporting. It’s also a hardware issue.

They are guessing. They see AMD recalling all units and announcing that they're replacing it with a fresh batch and jump to the conclusion that it must be a hardware issue.

But that's not necessarily the case. The "didn't go through QA" explanation works just as well. It's simply faster to recall and send out the next batch of products, which this time already went through QA. They can then without pressure reopen the recalled units and do the QA for them, which certainly takes more time.

 

[adroc_thurston]

Is there a chance Strix Halo would outperform 9950X for membw bound workloads?

Yeah, by a lot, each CCD gets 2*32B fabric uplink (aka the new norm going forward).

LPDDR5x @ 8533 MHz with 256-bit interface is 273 GB/s.

That's not the limiting factor at all.

 

[CouncilorIrissa]

That's not the limiting factor at all.

I'm aware. He explicitly asked about memory bandwidth, not the throughput between the I/O die and the CCD (for GNR).

What's your best guess on how high the STXH will clock anyway?

 

[Det0x]

Is 3600uclk and 2400fclk possible or is the italian guy delusional ?

I already gave my opinion in post #16459

I can accept he thinks he is running 2400mhz FCLK and/or buggy bios is showing him 2400mhz FCLK, but i dont think the system is actually running that in reality..
But its also very easy to prove me wrong, he just could post some read/write/copy numbers from aida or even better clam cache/mem benchmark @ 2000mhz, 2200mhz and 2400mhz FCLK so we could see the scaling (he have no NDA right?)
(Zen4 and now Zen5 is pretty much all bandwidth limited by FCLK)

He can use this is a comparison for what maxed out 2200mhz FCLK on a dual CCD cpu should look like (12 vs 16 core count show not matter in this regard)

6600MT/s 1:1


8000MT/s 2:1

 

[Josh128]

So Italian dude with 9900X runs a few gaming tests with a "questionable" memory setup and thats it? Radio silence since then? At least throw us some more crumbs, bro, even stale ones, lol. People better hope I never manage to snag such an early unit and not be under NDA like him. There would be a lot of pissed off tech news outlets out there...

 

[Hans Gruber]

At least AMD owes customers an explanation and a side by side comparison of the non certified chips vs. the certified AMD Zen 5 chips they are sending out to testers. We need to see what is going on and why the 1-2 week release delay is happening.

 

[poke01]

N3E is enterprise/server, N3 is efficient junk. N3P is the value play performance/efficiency mainstream 3nm. N3X is for Nvidia, high end performance and increased silicon density.

N3E is not just for server. N3P is the evolution of N3E.

 

[poke01]

Yes because then most "cheap" AMD users wouldn't want the extra cost being passed onto them. And the cost got driven high due to Apple.

TSMC dictates the prices not Apple. Anyway, if you believe so show some proof that that Apple is increasing the price of TSMC nodes.

 

[Hitman928]

At least AMD owes customers an explanation and a side by side comparison of the non certified chips vs. the certified AMD Zen 5 chips they are sending out to testers. We need to see what is going on and why the 1-2 week release delay is happening.

. . . Why?

 

[gaav87]

So Italian dude with 9900X runs a few gaming tests with a "questionable" memory setup and thats it? Radio silence since then? At least throw us some more crumbs, bro, even stale ones, lol. People better hope I never manage to snag such an early unit and not be under NDA like him. There would be a lot of pissed off tech news outlets out there...

Funny thing i got hands on 5600x on 31th oct 6days before official release. And litteraly noone was interested back then xD I think i only got a mention on VC or WCFKEK
Look at the date https://valid.x86.fr/rsf5p1
I asked the italian guy if he can show aida64 with 2400fclk or come here on the forum xD

 

[Hans Gruber]

. . . Why?

It's called transparency.

 

[Hail The Brain Slug]

It's called transparency.

With no official reviews performed with potentially defective units and no retail sales to consumers of potentially defective units, there's very little reason for transparency. AMD has not "wronged" anyone, only delayed a product release a short time. They very well could have delayed it with no statement at all.

 

[Hans Gruber]

With no official reviews performed with potentially defective units and no retail sales to consumers of potentially defective units, there's very little reason for transparency. AMD has not "wronged" anyone, only delayed a product release a short time. They very well could have delayed it with no statement at all.

Is AMD paying you for your statement? Or are you an AMD voluntary public service agent here on the forum?

Hans, you are not supposed to JUMP out of Nakatomi Plaza. One week vacation. Aloha!

Mod DAPUNISHER

 

[Hitman928]

It's called transparency.

Should every company always put out a statement and compare measurements every time they find an issue with an unreleased product? What a weird request. No one is owed "transparency" for all the work, debugging, and validation that happens with unreleased products. AMD said they discovered an issue (in testing), it's been resolved, and good product will be delivered with a very small delay. As long as the released products are good, there will be nothing to add to the story.

 

[igor_kavinski]

TSMC dictates the prices not Apple. Anyway, if you believe so show some proof that that Apple is increasing the price of TSMC nodes.

It's common sense logic. Who buys up capacity in bulk, years in advance and who has loads of cash and still makes loads more through their uniquely demented userbase? Out of all of TSMC's customers, that's Apple. They may have done a service to mankind by enabling TSMC to do accelerated R&D for the invention of advanced process technologies because of Apple's needs but they are also the ones who promoted TSMC to becoming a monopoly. Apple is like oxygen. A necessary evil. Too much leads to oxidation and accelerated aging and not enough halts future progress. I love *some* things about them but I hate way more about them than I love about them.

 

[maddie]

Is AMD paying you for your statement? Or are you an AMD voluntary public service agent here on the forum?

Fantastic reply. I always like hearing from strange new worlds.

 

[Thibsie]

Ian said that Zen 5c would see a much smaller drop in frequencies compared to what Z4c had relative to the classic cores.
Interesting.

Well, this expected if they talk about Epyc : 4nm vs 3 nm

 

[igor_kavinski]

AMD said they discovered an issue (in testing), it's been resolved, and good product will be delivered with a very small delay. As long as the released products are good, there will be nothing to add to the story.

I don't really believe AMD. This sounds too suspicious, TBH.

 

[Hitman928]

I don't really believe AMD. This sounds too suspicious, TBH.

Ok. . .

What do you think is the issue then?

 

[desrever]

Ok. . .

What do you think is the issue then?

AMD said there was issues and he doesn't believe AMD so obviously theres no issues and the delay is because AMD felt like it.