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

[DisEnchantment]

Speculate at will

 

[gdansk]

Well I was very optimistic on clock rates making up for some of the hype train derailing but it didn't happen.
At 6GHz I think 16% would be fine.

 

[adroc_thurston]

I think 96c Genoa is at least somewhat gimped by memory performance,

naa, that's only for client parts.
sIOD does very well with mem, you've seen one of the mandarin-speakers on Twitter post his Z4 TR results with 2 channels active.

Does Turin still use the same IOD?

No, a new one.
But not many changes there.

Well I was very optimistic on clock rates making up for some of the hype train derailing but it didn't happen.
At 6GHz I think 16% would be fine.

Still slow.
And doesn't work out with AMD's own Turin numbers.
They're like, public domain knowledge. Your mes and dolans of the world know it.

 

[Ghostsonplanets]

AMD introduces Zen5 "Strix Point" APUs and new Ryzen AI 300 branding - VideoCardz.com

AMD Strix Point: Zen5 with RDNA3.5 now official AMD is changing the branding for its new AI processors. AMD is revamping its naming scheme just 20 months after introducing a new one, this time for the upcoming mobile SKUs dubbed “Strix Point,” featuring the Zen5 architecture. These Strix Point...

videocardz.com

 

[H433x0n]

Wasn't >15% the original claim from AMD? And not long after they said that, weren't people here and elsewhere starting to run wild with speculation about how they were lowballing? So then we'd hear >20%, then 25%, then 40% and some going even higher than that! I laughed every time I saw people pushing the 40% thing, it was wildly unrealistic but people with an ax to grind against Intel or maybe Apple wanted it to be true so badly they somehow fooled themselves into believing it.

Claiming 16% at launch is right in line with the original claims of ">15%", and if it turns out they had a fudge a little to get there that's understandable given their promise of >15% and the knowledge that they had allowed the Zen 5 hype to get so unrealistic that even a very conservative 16% would be disappointing for a lot of AMD fans.

If AMD was smart they would have tamped down the expectations to avoid all the crazy claims of 25, 30, 40% and more. All that does is create disappointment - and probably reduced their sales as no doubt some people who believed Zen 5 was going to be a massive jump (or at least thought there was a chance) held off buying. Some of those buyers might decide to hold off a bit longer to see what Intel can do, and if they can deliver (which remains to be seen, the inflated expectations thing seems to be starting on that side lately) those are sales AMD loses because they were content to let the speculation and hype run wild and uncorrected.

They either a) fudged the numbers to get to 16% or b) are clinically insane and want to lowball their own product?

 

[trivik12]

16% is really good. its not as if Zen 4 cucked or anything.

 

[adroc_thurston]

16% is really good. its not as if Zen 4 cucked or anything.

It's bad because it doesn't work out with Turin perf.
For like 7% moar CCD area it's fine, but again, this doesn't really math out with semi-public Turin perf.

 

[Joe NYC]

AMD introduces Zen5 "Strix Point" APUs and new Ryzen AI 300 branding - VideoCardz.com

AMD Strix Point: Zen5 with RDNA3.5 now official AMD is changing the branding for its new AI processors. AMD is revamping its naming scheme just 20 months after introducing a new one, this time for the upcoming mobile SKUs dubbed “Strix Point,” featuring the Zen5 architecture. These Strix Point...

videocardz.com

View attachment 100268

View attachment 100269

That's pretty good for Zen 5 + RDNA 3.5. The numbers were mostly tied with Phoenix / Hawk Point.

 

[gdansk]

16% is really good. its not as if Zen 4 cucked or anything.

16% is average, not really compatible with hype trains.

 

[Ghostsonplanets]

16% IPC is a good achievement. But it will fall flat due to the hype train.

I do commend AMD for having consistent execution and delivering double digit IPC gains every generation. Will be waiting for the Kraken goodness.

 

[turtile]

Still slow.
And doesn't work out with AMD's own Turin numbers.
They're like, public domain knowledge. Your mes and dolans of the world know it.

I wonder if they will release 3nm versions next year with higher performance. The slides from a while back say 4nm and 3nm for Zen 5. I thought 5c would be 3nm, but they built both on 4nm.

 

[Joe NYC]

16% IPC is a good achievement. But it will fall flat due to the hype train.

I do commend AMD for having consistent execution and delivering double digit IPC gains every generation. Will be waiting for the Kraken goodness.

I think the gamers will wait for V-Cache.

 

[Goop_reformed]

16% IPC is a good achievement. But it will fall flat due to the hype train.

I do commend AMD for having consistent execution and delivering double digit IPC gains every generation. Will be waiting for the Kraken goodness.

16% is not a good achievement, that's lower than zen 2 to zen 3, and zen 5 is a brand new core too.

 

[adroc_thurston]

But it will fall flat due to the hype train.

No, not the problem.

AMD introduces Zen5 "Strix Point" APUs and new Ryzen AI 300 branding - VideoCardz.com

AMD Strix Point: Zen5 with RDNA3.5 now official AMD is changing the branding for its new AI processors. AMD is revamping its naming scheme just 20 months after introducing a new one, this time for the upcoming mobile SKUs dubbed “Strix Point,” featuring the Zen5 architecture. These Strix Point...

videocardz.com

View attachment 100268

View attachment 100269

That's more like it, swats Dragon Range into nonexistence.

I wonder if they will release 3nm versions next year with higher performance. The slides from a while back say 4nm and 3nm for Zen 5. I thought 5c would be 3nm, but they built both on 4nm.

Node isn't the issue.

 

[H433x0n]

16% IPC is a good achievement. But it will fall flat due to the hype train.

I do commend AMD for having consistent execution and delivering double digit IPC gains every generation. Will be waiting for the Kraken goodness.

Look at how they got to that +16% and what’s included in the averages.

WebXprt and Speedometer benchmarks appear to still be worse than RPL or at parity.

 

[SpudLobby]

This is one of the things I actually look forward to AI being useful for - "hey Siri, watch this Youtube video for me and provide a written transcript that elides all the ads, [once Google finally figures out how to make me actually see any] talk about sponsors, and self-promotional crap"

I generally refuse to watch Youtube videos that are just talking - and I've found it is safe to assume that if you're given a link to Youtube video as a source of "news" 99% of the time it is pushing conspiracy theories and/or highly slanted partisan bs. If Google ever succeeds in making me see ads on Youtube with either Linux/Firefox/uBlock or iPhone/Safari/FirefoxFocus I'll give it up completely unless/until I can get Siri to watch it for me. Then Google will be out to figure out a way to prevent AIs from watching Youtube - good luck find a CAPTCHA for Youtube that AIs can't do but doesn't annoy humans to the point of near insanity lol!

Doug: yep. Gemini 1.5 Pro is already quite useful for this.

Zen 5 and X5 - both ended up about the same.

The X925 also has about a significant IPC lead since they already had about a 25% lead (higher in GB6 tho).

And is prob. Better for low power. So. Different tradeoffs

 

[trivik12]

At this point AI hype matters more. That is why we hve APU like AMD Ryzen AI 9 HX370 getting release :-(

 

[gdansk]

The X925 also has about a significant IPC lead since they already had about a 25% lead (higher in GB6 tho).

X5 also will clock higher than its predecessor.
But I meant the same IPC improvement relative to its predecessor and the high expectations some had.

 

[Saylick]

16% is really good. its not as if Zen 4 cucked or anything.

It's good from an IPC standpoint, but when you compare it to past generations and the overall performance uplift was (clocks * IPC), 16% higher IPC with a slight clock decrease for the lower SKUs is kinda meh.

Consider that Zen 2 brought 15% IPC and a few hundred MHz clock uplift, that gives you like 20% overall improvement to PPC. Zen 3 gave us 19% IPC with also another few hundred MHz clock speed increase. Zen 4 gave us a smaller IPC uplift, but the clocks were substantially higher. But now Zen 5 comes along with arguably the largest architectural overhaul and it only arrives with 16% IPC uplift and no real clock speed improvement? What gives?

 

[Tuna-Fish]

naa, that's only for client parts.
sIOD does very well with mem, you've seen one of the mandarin-speakers on Twitter post his Z4 TR results with 2 channels active.

No, I specifically mean that if you have a 48c Epyc with 6 channels populated, it should in theory have memory perf per core of nearly identical to a 96c Epyc with all channels populated. In practice this is not true and the 96c EPYC slightly underperforms. If they fixed that, they could see nontrivial improvement in 96c Turin vs 96c Genoa.

 

[poke01]

16% IPC is a good achievement. But it will fall flat due to the hype train.

I do commend AMD for having consistent execution and delivering double digit IPC gains every generation. Will be waiting for the Kraken goodness.

AMD's generation is every 2 years. Thats 8% IPC every year. This was meant to be a redesigned core, most had higher expectations

 

[adroc_thurston]

If they fixed that, they could see nontrivial improvement in 96c Turin vs 96c Genoa.

Naa, the internal fabric isn't any wider.
Again, this doesn't really math out. Time to annoy people.

AMD's generation is every 2 years. Thats 8% IPC every year. This was meant to be a redesigned core, most had higher expectations

Not the issue. It just doesn't mash with server perf AMD communicated.

 

[Abwx]

Look at how they got to that +16% and what’s included in the averages.

WebXprt and Speedometer benchmarks appear to still be worse than RPL or at parity.

WebXprt is an Intel benchmark, as relevant as your prediction of a 170W TDP 8 cores...

Anyway i was spot on for the improvement in Blender.

Page 443 - Discussion - Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)

Page 443 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Saylick]

It begins!

Take a swig whenever the word AI is spoken...

 

[RnR_au]

Is it a beer or a shot whenever we hear 'AI' ? I wanna be sure of the rules...

Edit: ... ooh... I'm out of beer already...

 

[Tuna-Fish]

It begins!

Take a swig whenever the word AI is spoken...

I am doing this.

I am limiting myself to one 500ml glass, for health reasons. I will update you guys when I run out.