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

[DisEnchantment]

Speculate at will

 

[Goop_reformed]

View attachment 86136

With all the chatter of frequency regression with Zen 5, a 300-400 MHz frequency regression would make it the weakest improvement for a Zen generation ever even with a fat 25% IPC gain. They have to keep same clocks otherwise the would need an impossibly high 35%+ IPC gain just to match the Zen 4 perf gains.
Doubtful Zen 5 can repeat the ~27% ST perf gain of Zen 4 with a clock regression. Couple that with FCLK plateauing from supposed 'same IOD' rumors.

On the other hand, at lower clocks efficiency should be greatly improved.

Hello everyone.

I've been lurking around for ages but finally registered an account just to make a comment about zen 5.

If this chart is correct then I find it rather disappointing that even with 25% IPC increase, overtal ST performance is only equal the performance jump from zen 2 to zen 3. It definitely should be a tad higher than 30% to justify 6 years of development which is even longer than the original zen.

 

[Markfw]

Hello everyone.

I've been lurking around for ages but finally registered an account just to make a comment about zen 5.

If this chart is correct then I find it rather disappointing that even with 25% IPC increase, overtal ST performance is only equal the performance jump from zen 2 to zen 3. It definitely should be a tad higher than 30% to justify 6 years of development which is even longer than the original zen.

I can't believe you spent al lthis time lurking just to come in now and make degrading comments about Zen 5. Is not even out yet !

 

[Goop_reformed]

I can't believe you spent al lthis time lurking just to come in now and make degrading comments about Zen 5. Is not even out yet !

Only saying that 25% ipc improvement with 5% reduction in frequency is disappointing.

 

[H433x0n]

Hello everyone.

I've been lurking around for ages but finally registered an account just to make a comment about zen 5.

If this chart is correct then I find it rather disappointing that even with 25% IPC increase, overtal ST performance is only equal the performance jump from zen 2 to zen 3. It definitely should be a tad higher than 30% to justify 6 years of development which is even longer than the original zen.

I think everybody is losing perspective due to the normal AMD hype cycle. I don’t get why this always happens with AMD products.

If Zen 5 has 25% IPC increase that’s a big deal. In the past 25 years both Intel and AMD have only managed to perform that feat once. AMD achieved it with the original Zen and Intel managed it with Conroe.

 

[Goop_reformed]

I think everybody is losing perspective due to the normal AMD hype cycle. I don’t get why this always happens with AMD products.

If Zen 5 has 25% IPC increase that’s a big deal. In the past 25 years both Intel and AMD have only managed to perform that feat once. AMD achieved it with the original Zen and Intel managed it with Conroe.

25% ipc increase with 5% frequency reduction is only 20% ST uplift which lowest in zen history. That is worrying to be honest, at least for desktop anyway. DC with AI chiplets and mobiles on the other hand is a different story altogether but I don't use any epyc nor moobile parts.

Let's put this into perspective: Right now raptor lake is 10% ahead of zen 4 more or less. If zen 5 is 20% ahead of zen 4 then it's just gonna be only 10% ahead of raptor lake.

 

[SpudLobby]

Is 3NE going to be ready in time for that? I e seen posts in other threads suggesting that the 3NE node would only be ready by mid-year. I don't know if that was based on anything other than speculation or rumors though.

Yes.

Why are we having this conversation? It's widely rumored MediaTek and Qualcomm both are going to opt for N3E for cost sensitivity and yield/power concerns, we know N3B is a bit of a bust relative to expectations and that TSMC put a lot of effort into a successor of sorts that still provided the essentials (save for SRAM...) of N3 minus the downsides, which is N3E.

It will be in mass production in 2024 - in fact as Kepler showed it's already in mass production for some parts because strictly it's 2H 2023.

 

[Joe NYC]

Hello everyone.

I've been lurking around for ages but finally registered an account just to make a comment about zen 5.

If this chart is correct then I find it rather disappointing that even with 25% IPC increase, overtal ST performance is only equal the performance jump from zen 2 to zen 3. It definitely should be a tad higher than 30% to justify 6 years of development which is even longer than the original zen.

Welcome aboard.

I think some people who have access to info on Zen 5 have been hinting at IPC increase quite a bit higher than 25%. In 35% range.

 

[SpudLobby]

You're overhyping the shrink from N5 to N4 way too much, especially considering the 7950x boosts at like >5Ghz lol.
Fatter architecture causes increased power draw iso frequency. It gets more IPC sure, but that's not the point here. The potential ST frequency regression isn't likely due to a increase in power consumption, but because clocking wider cores super high is just hard, esp when you also have to balance it out with area.
The MT clock regression however, is likely due to potentially increased power consumption, as MT performance is much more limited by power draw.
We have seen this exact same pattern with Intel with Cypress Cove vs Skylake. They widened the core quite a bit, lost frequency iso power, though ST clocks here remained roughly the same. It's a valid expectation to imagine a zen 5 with something like 30% IPC, with a 5-10% increase in power consumption iso clock, and 200MHz shaved off in peak ST frequency.

That's correct: RE:

Wider and larger architecture iso-clock, iso-process draws more power ceteris paribus AKA at least with similar structures elsewhere in total or gating, blah blah and depends on if you count total board power or what's going on with the memory hierarchy.

Yeah agree RE: ST clocks. meeting timing with a big wide core and the density AMD will use probably shaved some off.

At any rate, I'm pretty positive Zen 5 will still offer an iso-performance power gains or energy efficiency gains over Z4 from lowering the voltage and using a wider, slower and smarter structure. N4P will help too, but only a teeny bit.

 

[Goop_reformed]

Welcome aboard.

I think some people who have access to info on Zen 5 have been hinting at IPC increase quite a bit higher than 25%. In 35% range.

To be fair zen 5 samples are out for months as of now which is very unusual.

 

[Markfw]

25% ipc increase with 5% frequency reduction is only 20% ST uplift which lowest in zen history. That is worrying to be honest, at least for desktop anyway. DC with AI chiplets and mobiles on the other hand is a different story altogether but I don't use any epyc nor moobile parts.

Let's put this into perspective: Right now raptor lake is 10% ahead of zen 4 more or less. If zen 5 is 20% ahead of zen 4 then it's just gonna be only 10% ahead of raptor lake.

Again, let me say, Zen 5 is not out yet ! Second, I would question you saying that Raptor lake is ahead of Zen 4. In what benchmark ? and at what power usage ?

 

[SpudLobby]

Hello everyone.

I've been lurking around for ages but finally registered an account just to make a comment about zen 5.

If this chart is correct then I find it rather disappointing that even with 25% IPC increase, overtal ST performance is only equal the performance jump from zen 2 to zen 3. It definitely should be a tad higher than 30% to justify 6 years of development which is even longer than the original zen.

This is over the top. I'm revolted into defending Spec and his ilk to a T on this even though it's plausibly somewhat their fault what with the excitability of junior high students on rumors and leaks which change conveniently and disappear with the enthusiasm of Joseph_Stalin's_comrade.jpeg as if nothing but the most recent rumor was ever canon. But at any rate, be realistic. Do not do this game of "AMDOA" for a very solid upgrade just 2 years after Zen 4.

25% - these are fractions of previous values which do grow in absolute. For how far Zen has come and where Intel, Apple, and the Cortex cores are today this is great, A++.

It will put them right in +-5% with a Cortex X4 in some benchmarks albeit with higher clocks and shipping in real products you can buy for a PC (or server at that soon), same for Apple's core. The gaps will have been closed on perf/GHz with this. I doubt it will really have the low power operation of the mobile ones, but it'll be good enough and an improvement upon Zen 4 I think. There will not even be any server competition yet.

Strix Point, the main monolithic part for Zen 5, also has RDNA3.5 or whatever it is, so there's that too.

 

[Goop_reformed]

Again, let me say, Zen 5 is not out yet ! Second, I would question you saying that Raptor lake is ahead of Zen 4. In what benchmark ? and at what power usage ?

Trying to find specs bench and a direct head to head between both uarch. https://www.tomshardware.com/news/amd-ryzen-9-7950x-vs-intel-core-i9-13900k should be good in the mean time.

 

[Hitman928]

Trying to find specs bench and a direct head to head between both uarch. https://www.tomshardware.com/news/amd-ryzen-9-7950x-vs-intel-core-i9-13900k should be good in the mean time.

Here you go,

Intel Core i9-13900K and i5-13600K Review: Raptor Lake Brings More Bite

www.anandtech.com

The key points to digest here is that Intel has done well to bridge the gap in single-threaded performance to Ryzen 7000 in most of the tests, and overall, it's a consistent trade-off between which test favors which mixture of architecture, frequency, and most importantly of all, IPC performance.

Right now, however SPEC paints a picture where it's pretty much neck and neck between Raptor Cove and Zen 4.

 

[gruffi]

Only saying that 25% ipc improvement with 5% reduction in frequency is disappointing.

Not really. And better than Bulldozer. Which had like 15% IPC reduction and 20% clock speed improvement. Which means it wasn't faster than its predecessor at all in 1T. And for how long was Bulldozer in the development?

Let's put this into perspective: Right now raptor lake is 10% ahead of zen 4 more or less. If zen 5 is 20% ahead of zen 4 then it's just gonna be only 10% ahead of raptor lake.

Raptor Lake is not the benchmark. Zen is the benchmark. Intel needs a crapload of power for its performance. It's simply brute force, also with higher clock speeds, nothing else. That's not how smart and sustainable development works. Intel is at a dead end with their current uarch. And they know it. Let's see if Intel can manage to increase the IPC while achieving similar clock speeds with Arrow Lake and successors.
For now AMD achieves about the same IPC with a narrower front end and a narrower integer backend. Which is a quite impressive work of engineering. That's why Zen 5 is so interesting in my opinion. The Zen uarch is going a lot wider too now. Like Apple's M1 or Intel's Alder Lake before.


Personally I expect 20-25% IPC improvement and no clock speed degradation. Which would be quite a big deal. Considering it's probably the same process node as Zen 4 (TSMC N4 is just a tweaked N5 process). +20-25% IPC would be a quite significant generational leap. If Zen 5 turns out to be better then that it would be a pleasant surprise to me.

 

[PJVol]

I’d be shocked if ST frequency only dipped by 200mhz. I don’t know if a core has ever gotten 30-40% wider and only had to give up 3-4% frequency in exchange while on a similar node. Such a feat is possible but it’d require way more silicon real estate and a much larger power budget and that doesn’t seem to be AMD’s style.

What ST frequency has to do with the power budget?

Yeah agree RE: ST clocks. meeting timing with a big wide core and the density AMD will use probably shaved some off.

Indeed, timing can be a problem when going wide, partially to be solved by pipelining, but I don't know is there any STA data leaked/available?

 

[deasd]

Hello everyone.

I've been lurking around for ages but finally registered an account just to make a comment about zen 5.

If this chart is correct then I find it rather disappointing that even with 25% IPC increase, overtal ST performance is only equal the performance jump from zen 2 to zen 3. It definitely should be a tad higher than 30% to justify 6 years of development which is even longer than the original zen.

If you have lurked for ages, you should already knew Redgamingtech is not reliable.

 

[Timmah!]

I for one will be disappointed by that clock regression. Those 5,7GHz speeds were the main selling point of Zen4. Even if its overall faster anyway, you´ll just gonna feel its another "2 steps forward, 1 step backward" situation and at some point in the future the clocks will be back at Zen4 levels or beyond that.

It has silver lining though, the less impressive it will be, the less obliged will i feel to upgrade to it :-D

 

[Timmah!]

Did not know he was temporarily banned, but it's cases like this I'm kinda glad he is so other opinions can be discussed without being drowned.

Nah, unban him. Granted, he was annoying sometimes, but he seemed to have some insider info. Ban him only when Zen 5 gets released and if he turns out to be a fraud :-D

 

[inf64]

If AMD aimed for similar IPC jump they got from Zen 1 -> Zen 3, then Zen 3 -> Zen 5 should be ~42% IPC increase , and hence ~25% versus Zen 4. If they aimed to get something like Bulldozer(EX) -> Zen 1, then Zen 5 should bring 35% higher IPC versus Zen 4.

 

[Glo.]

I would like to remind us all, that AMD hype trains like to derail themselves after the releases of new AMD hardware .

Lets keep our expectations in check, and don't overthink stuff that is thrown on the internet.

 

[Markfw]

Trying to find specs bench and a direct head to head between both uarch. https://www.tomshardware.com/news/amd-ryzen-9-7950x-vs-intel-core-i9-13900k should be good in the mean time.

Gaming, does NOT include the 7950x3d, which is the current winner. The rest except power and price, tie. Power, AMD is more efficient . Price. That was a year ago. I just got a 7950x for $550 and a motherboard (670) for $150, Intel MIGHT have been more expensive a year ago, not anymore. Where is this 10% then ? In your own outdated link its not there. Current is is certainly not there.

 

[CakeMonster]

The assumption is that Z5 stays on a N5 derived node (N4*) and that Z6 will use N3*?

 

[Mopetar]

I would like to remind us all, that AMD hype trains like to derail themselves after the releases of new AMD hardware .

But that just means it's time to start birding the next train! If you have to get off anyways, does it matter whether it pulls into the station or comes sliding in sideways?

 

[Joe NYC]

It’s discussions like these that I wish @adroc_thurston wasn’t on a temp ban.

Seriously?

Banning the most informative posters really should stop...

I don't think the people here on Anandtech are snowflakes, who will melt after a more heated discussion with @adroc_thurston

 

[Goop_reformed]

Again, let me say, Zen 5 is not out yet ! Second, I would question you saying that Raptor lake is ahead of Zen 4. In what benchmark ? and at what power usage ?

I wrote a very long post to reply but somehow was lost over a f5 Rip. Anyway, if zen 5 only manages a 20% uplift in ST then what about zen 6? I think the actually uplift should be at least 30% otherwise zen 6 would be in trouble with the royal cores.

Again I'm not bashing amd in anyway, in fact I have vested interests in amd upward performance. They paid my tuition after all.