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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

We gonna have Zen5X3D on shelves in March 2024?

No, just vanilla Zen5.
X3D gonna be later since very weak comp environment.

No. Desktop doesn't matter. Shelves don't matter.

So true, Turin and especially -Dense gonna be $$$.

 

[TESKATLIPOKA]

I would rather know when we will see first Strix Point and Strix Halo laptops.

 

[jpiniero]

No. Desktop doesn't matter. Shelves don't matter.

Intel made more revenue on desktop last quarter than any segment AMD has...

 

[adroc_thurston]

Intel made more revenue on desktop last quarter than any segment AMD has...

Revenue isn't profit.
Profit is what matters.

 

[gdansk]

Intel made more revenue on desktop last quarter than any segment AMD has...

And even if Zen 5 is the messiah it won't change that by targeting desktop.

 

[jpiniero]

Revenue isn't profit.
Profit is what matters.

They did have a billion in Operating Income in Client (they don't break that down by Desktop/Notebook).

 

[Abwx]

Lol ... somehow he managed to throw three different types of VR together and put it in one Zen package

You post schematics without even knowing what it is about to make people think that you are in the know, and you are saying things that are false, i never said that there 3 kind of regulators in Zen, but a single kind, namely capacitors that are charged by high speed switching mosfets and without inductances, learn to read before keeping trolling.

Indeed you dont even know how those systems works, FTR i can design any of these kind of regulators for whatever power up to 1000W if you want.

Case closed for me since you are just cluless and trolling all the way, ignorance being your argument.

 

[Saylick]

Anyone have an idea when Turin will be available to the general public via Dell, HP, etc? H2 2024?

 

[adroc_thurston]

They did have a billion in Operating Income in Client (they don't break that down by Desktop/Notebook).

That's with some pretty aggressive cost-cutting.

H2 2024?

Yea.

 

[igor_kavinski]

No, just vanilla Zen5.
X3D gonna be later since very weak comp environment.

 

[Saylick]

Yea.

Dang, my company would miss the upgrade cycle by 6 months then. No matter, Genoa it is. Would be a sweet upgrade over these first gen Xeon Scalable processors that we've had for years now.

 

[adroc_thurston]

If only ARL wasn't so memey.

Dang, my company would miss the upgrade cycle by 6 months then

Yea.
I would generally wait for Turin since it's a lot better as general-purpose CPU stick of doom, but that assumes you can defer the procurement.

 

[igor_kavinski]

No matter, Genoa it is.

How many you need to replace? Get just one to show the bosses how great your technical acumen is so they would be willing to wait and put down $$$ to get Turin to replace everything.

 

[Saylick]

Yea.
I would generally wait for Turin since it's a lot better as general-purpose CPU stick of doom, but that assumes you can defer the procurement.

Remind me again, are the Turin vanilla CCDs on TSMC N4P? Just the Turin-Cloud CCDs are N3E? If so, and if there's not much perf/W gains for the N4P vanilla variant, then perhaps I'm not missing out on much since server processors are going to be power limited anyways.

 

[Saylick]

How many you need to replace? Get just one to show the bosses how great your technical acumen is so they would be willing to wait and put down $$$ to get Turin to replace everything.

We're a small engineering firm in the grand scheme of things. Not Mom and Pop small, but definitely far from an international design conglomerate. I think our budget is at least a hundred thousand but definitely less than a million dollars, so you're looking at a dozen or two blades. I'm just an engineer in the company and am not part of the IT Department who makes the final decision, but FWIW I am the chair of an internal committee that weighs in on our computational needs.

 

[adroc_thurston]

Remind me again, are the Turin vanilla CCDs on TSMC N4P?

Yes.

Just the Turin-Cloud CCDs are N3E?

They're Dense CCD, also used for edge part aka Sorano.

If so, and if there's not much perf/W gains for the N4P vanilla variant

Nodes aren't the only way to get ppw.

then perhaps I'm not missing out on much since server processors are going to be power limited anyways.

Doubt you'll be buying 96/128c SKUs and in mainstream-ish SP5 parts Turin is king due to gigantic per-core perf uplifts.

 

[igor_kavinski]

so you're looking at a dozen or two blades. I'm just an engineer in the company and am not part of the IT Department who makes the final decision, but FWIW I am the chair of an internal committee that weighs in on our computational needs.

That's a good number of blades. Tell them the more you wait, the sweeter the hardware

Buy half Genoa, half Turin. Better for the financials too.

 

[Saylick]

Doubt you'll be buying 96/128c SKUs and in mainstream-ish SP5 parts Turin is king due to gigantic per-core perf uplifts.

That does sound tempting, especially since the key workload we use these servers for is 100% single threaded with emphasis on matrix math but non-AVX, but we run multiple parallel processes. Naturally, my preferred processor at the moment are the Genoa F SKUs that are frequency optimized. Turin in the 48 core range sounds like it would be killer.

 

[adroc_thurston]

Turin in the 48 core range sounds like it would be killer.

Yea defer the procurement as hard as you can.
lol.

 

[igor_kavinski]

That does sound tempting, especially since the key workload we use these servers for is 100% single threaded with emphasis on matrix math

Shouldn't a GPU be a better fit for such workloads?

Using Performance Libraries

Programming oneAPI projects to maximize hardware abilities.

www.intel.com

Look at the crazy low completion times on the GPU:

 

[Hitman928]

That does sound tempting, especially since the key workload we use these servers for is 100% single threaded with emphasis on matrix math but non-AVX, but we run multiple parallel processes. Naturally, my preferred processor at the moment are the Genoa F SKUs that are frequency optimized. Turin in the 48 core range sounds like it would be killer.

Are you running civil/mechanical engineering software?

 

[Saylick]

Yea defer the procurement as hard as you can.
lol.

FFFFFF. I'll do what I can to defer.

Shouldn't a GPU be a better fit for such workloads?

Using Performance Libraries

Programming oneAPI projects to maximize hardware abilities.

www.intel.com

Look at the crazy low completion times on the GPU:

View attachment 84263

Yeah, you would think, but the software was written in the 90s and is archaic by today's standard. The only reason why we use it is because the analysis engine is robust and it's considered "tried and true".

Are you running civil/mechanical engineering software?

Yes, it's building nonlinear time-history analysis. Earthquake simulations.

 

[igor_kavinski]

Yeah, you would think, but the software was written in the 90s and is archaic by today's standard. The only reason why we use it is because the analysis engine is robust and it's considered "tried and true".

Long term, your company should really be looking at rewriting the software. That's insanely inefficient to depend on bruteforce computing power.

 

[adroc_thurston]

Long term, your company should really be looking at rewriting the software. That's insanely inefficient to depend on bruteforce computing power.

Tech debts are hard to get rid of.

 

[Hitman928]

FFFFFF. I'll do what I can to defer.

Yeah, you would think, but the software was written in the 90s and is archaic by today's standard. The only reason why we use it is because the analysis engine is robust and it's considered "tried and true".

Yes, it's building nonlinear time-history analysis. Earthquake simulations.

What OS does it support? Buying bleeding edge hardware means you need to use an up to date OS for proper hardware support, but many times these engineering CAD programs are slow to support OS updates. I was in a similar position as you a few years ago and we couldn’t go with the latest hardware because it required an OS version to support the hardware that the software wouldn’t support. Granted, the software probably could have been made to work on the updated OS, but when you are talking multi million dollar projects, people get real antsy at the words, “not officially supported “ showing up in the design resources plan. This is especially true because most CAD software support contracts are contingent on you running on officially supported configurations.

Just something to think about when planning upgrades. We actually came up with a stepped approach that got us the much better hardware eventually, just later than we had anticipated.