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

[DisEnchantment]

Speculate at will

 

[Goop_reformed]

Homie, why you using chips out right now to say you don't doubt potential numbers for... future chips? That's wild lmfao

Past behavior is the best predictor of future behavior. With extremely limited budget, amd managed aa awesome comeback with 4 consecutive generations of zen. That's why I was saying several page back if zen 5 isn't at least 25% above zen 4 that'd considered a disappointment in retrospect.

 

[Markfw]

Past behavior is the best predictor of future behavior. With very extremely limited budget, amd managed a awesome comeback with 4 consecutive generations of zen. That's why I was saying several page back if zen 5 isn't at least 25% above zen 4 that'd considered a disappointment in retrospect.

This is my logic....

 

[Goop_reformed]

This is my logic....

But would you agree anything less than 25% is a disappointment though

 

[Geddagod]

Past behavior is the best predictor of future behavior.

Sure, but you look for patterns, not looking at one generation of chips.
That would be like me saying "hey look at how good my 13900k is (I don't actually have one lol), the 14900k is going to be so good.

With very extremely limited budget, amd managed a awesome comeback with 4 consecutive generations of zen.

Wonder how much of that is Intel literally just delaying products to narnia...

That's why I was saying several page back if zen 5 isn't at least 25% above zen 4 that'd considered a disappointment in retrospect.

IPC? Deff no. Even if it's just overall performance, still no. Zen 3 was a similar uplift in MT IIRC, or even less.
It also depends on what the scope of Zen 5 was supposed to be ig, but they don't need a giant architectural uplift (as in Zen 1 level) to be more than competitive vs Intel.
This is what happens when people hype up a product much too hard before launch. If it meets those sky high leaks, well it's fine, but if it doesn't, then people start thinking it's "bad" when in a vacuum it would be pretty good.

This is my logic....

Sure...

 

[Thunder 57]

Past behavior is the best predictor of future behavior. With extremely limited budget, amd managed aa awesome comeback with 4 consecutive generations of zen. That's why I was saying several page back if zen 5 isn't at least 25% above zen 4 that'd considered a disappointment in retrospect.

How did Intel do with a massive budget? Sometimes being smaller and more agile is a benefit.

 

[adroc_thurston]

It also depends on what the scope of Zen 5 was supposed to be ig, but they don't need a giant architectural uplift (as in Zen 1 level) to be more than competitive vs Intel.

That's not how any of this works.
You usually project comp extremely aggressively which is what AMD did.

How did Intel do with a massive budget? Sometimes being smaller and more agile is a benefit.

Intel had a very unique case of cultural decay (and manpower departure to a lesser extent).

 

[Thunder 57]

That's not how any of this works.
You usually project comp extremely aggressively which is what AMD did.

Intel had a very unique case of cultural decay (and manpower departure to a lesser extent).

I don't disagree. I won't go into why I think some of that is happening as this is not P&N. I was just responding saying a nice budget doesn't necessarily produce a good product.

 

[Geddagod]

That's not how any of this works./

Wait, you telling me if AMD projects a 20% IPC bump (like a zen 3 level increase) they are going to be disappointed if they only get 20% and not 40%?

You usually project comp extremely aggressively which is what AMD did.

To a certain extent, sure, but I'm guessing Intel wasn't projecting 20% ipc bump for RWC. There's a difference between "aggressive" and "delusional"

 

[adroc_thurston]

I was just responding saying a nice budget doesn't necessarily produce a good product.

It usually correlates well unless the company in question has major structural issue.

Wait, you telling me if AMD projects a 20% IPC bump (like a zen 3 level increase) they are going to be disappointed if they only get 20% and not 40%?

Projections as in competitive projections.
You target relative to what competition has.

 

[Geddagod]

You target relative to what competition has.

You also release targets based on your own previous products
Hence that 2x EMR perf projection from that Intel leak for GNR.

But if you're talking about if companies aim to target perf based on what their competitor is also releasing (early in design), well sure, but as I said earlier, AMD doesn't need a massive architectural IPC uplift in order to be competitive with what Intel has coming up. Even a standard zen 3 level IPC increase is fine. And from a gen-on-gen perspective, that won't be disappointing either, unless AMD was aiming for much higher increases and failed to hit them.

 

[Joe NYC]

?
Mainstream -U chips have been the NUC fuel since their inception lol.

Some of the NUCs are using the HS models and there is also an HX for more power. The U models would be for super low power...

 

[Joe NYC]

I don't believe Strix Halo will be paired with TOP line dGPUs unlike Dragon Range, so why would they pay for extra cache?

I think the reference to Dragon Range was that some OEMs are willing to release a product with a pricey CPU.

While, I think, there will still be a nominal Dragon Range successor (put together at minimum cost to AMD), the main effort, and where more money went is Strix Halo, which will likely have the same CPU power (roughly) plus strong GPU.

 

[igor_kavinski]

Even a standard zen 3 level IPC increase is fine. And from a gen-on-gen perspective, that won't be disappointing either, unless AMD was aiming for much higher increases and failed to hit them.

Maybe the early silicon beat all their theoretical estimates and simulations. Is that really hard to believe?

 

[Geddagod]

Maybe the early silicon beat all their theoretical estimates and simulations. Is that really hard to believe?

I'm not even commenting on what Zen 5 leaks are claiming it will or won't bring.
All I'm saying is that a 25% increase in perf isn't disappointing. The only way it would be disappointing is if AMD aimed for higher, and they failed to meet it. I'm not saying that's what happened, I'm saying that's the only way a 25% jump in perf would be disappointing.
I don't know how much longer I can keep beating this drum lol

 

[Geddagod]

the main effort, and where more money went is Strix Halo, which will likely have the same CPU power (roughly) plus strong GPU.

Ye, I think most people care way more about Strix Halo than DRG... though idk about roughly the same CPU power, bcuz of TDPs...
I think it's a bit interesting Intel and AMD both are creating mobile skus that the other company doesn't seem to interested in competing in. ULP LNL for Intel, "mega APU" Strix from AMD.

 

[adroc_thurston]

You also release targets based on your own previous products

Product level yes.
IP level no.

Some of the NUCs are using the HS models

Those are just -U with higher power limits.

 

[Goop_reformed]

Sure, but you look for patterns, not looking at one generation of chips.
That would be like me saying "hey look at how good my 13900k is (I don't actually have one lol), the 14900k is going to be so good.

Wonder how much of that is Intel literally just delaying products to narnia...

IPC? Deff no. Even if it's just overall performance, still no. Zen 3 was a similar uplift in MT IIRC, or even less.
It also depends on what the scope of Zen 5 was supposed to be ig, but they don't need a giant architectural uplift (as in Zen 1 level) to be more than competitive vs Intel.
This is what happens when people hype up a product much too hard before launch. If it meets those sky high leaks, well it's fine, but if it doesn't, then people start thinking it's "bad" when in a vacuum it would be pretty good.

Sure...

That's false equivalence though. 12900k, 13900k, 14900k are essentially the same technologically. What you said about intel is also true, however amd's execution should not be downplayed. And people only hyping up zen 5 only because of the expectations from previous gens results. The rest is just smoke.

How did Intel do with a massive budget? Sometimes being smaller and more agile is a benefit.

Well intel has more than 10x employees at some point, therefore management was substantially more vital. And when intel hit a snag those management positions became useless.

 

[itsmydamnation]

. And people only hyping up zen 5 only because of the expectations from previous gens results. The rest is just smoke.



.

What's smoke about:
2 more Alus
More load store bandwidth
More L1D
More instruction bandwidth
Likely significantly larger rob and prf

Like we are talking more change then Intel did from skylake to ice lake to golden cove but apparently amd suck and can only get 15% IPC at reduced clocks......

If I was a betting man (I am) I'm willing to bet spec int 1t is closer to 32% then it is to 15% at equal clock.

 

[Thunder 57]

What's smoke about:
2 more Alus
More load store bandwidth
More L1D
More instruction bandwidth
Likely significantly larger rob and prf

Like we are talking more change then Intel did from skylake to ice lake to golden cove but apparently amd suck and can only get 15% IPC at reduced clocks......

If I was a betting man (I am) I'm willing to bet spec int 1t is closer to 32% then it is to 15% at equal clock.

Maybe we should have a predicition thread?

If I had to throw out a number I'd say lower 20's, maybe 22-23%. Still waiting for @inf64's prediction since he seems to nearly nail it every time according to his signiture.

 

[adroc_thurston]

If I had to throw out a number I'd say lower 20's, maybe 22-23%

Reasonable, yes.
I had the same expectations before the numbers(tm) dropped.

 

[Goop_reformed]

What's smoke about:
2 more Alus
More load store bandwidth
More L1D
More instruction bandwidth
Likely significantly larger rob and prf

Like we are talking more change then Intel did from skylake to ice lake to golden cove but apparently amd suck and can only get 15% IPC at reduced clocks......

If I was a betting man (I am) I'm willing to bet spec int 1t is closer to 32% then it is to 15% at equal clock.

Should have mentioned the smoke is about 3 gen of golden cove. Zen 5 is obviously not smoke at all.

 

[Geddagod]

That's false equivalence though.

No it's not lol

12900k, 13900k, 14900k are essentially the same technologically.

So what you're saying is you look at a pattern of multiple generations of products released recently, not just the latest one?
The false equivalence is pretending @Markfw's comment made literally any sense at all. There is literally no relationship between Zen 4 going Vroom and thinking that Zen 5 is going to be a massive jump over Zen 4 lol. You can try to extrapolate the "pattern of multiple generations" claim from that, and that's correct, but that's not at all what he said, and not at all what I responded to him with...

And people only hyping up zen 5 only because of the expectations from previous gens results.

Cap
The majority consensus (on this forum at least) was that this was likely going to be a Zen 3 level jump, not a Zen 1 level one. It wasn't until @adroc_thurston joined that it shifted, and even then there's still healthy skepticism of that higher level IPC claims
But even if that were true, it's from looking at previous generationS not the one current generation

Like we are talking more change then Intel did from skylake to ice lake to golden cove but apparently amd suck and can only get 15% IPC at reduced clocks......

Bad argument. GLC only got a 15% jump in server lmao. And SNC got 18% IIRC. The difference between 18 to 15% is that much lol. Plus, if you believe in the 15% rumor, I doubt you are as likely to believe in the frequency regression rumor as well, at least not a significant frequency regression. And SNC was a frequency regression over SKL, even RKL's 'SNC' version just tied SKL. Plus, the all core frequency regression was seen in both SNC ICL and CYPRC RKL. Also, GLC was technically a Fmax frequency regression over WLC, and that used a better node than WLC to boot (and the node difference there is prob bigger than the one between N4 and N5 for Zen 5 vs Zen 4).

 

[adroc_thurston]

And SNC was a frequency regression over SKL

Ehhhhhh not in server.
Client yes, but 10+ wasn't a good node on a good day.

 

[Goop_reformed]

No it's not lol

So what you're saying is you look at a pattern of multiple generations of products released recently, not just the latest one?
The false equivalence is pretending @Markfw's comment made literally any sense at all. There is literally no relationship between Zen 4 going Vroom and thinking that Zen 5 is going to be a massive jump over Zen 4 lol. You can try to extrapolate the "pattern of multiple generations" claim from that, and that's correct, but that's not at all what he said, and not at all what I responded to him with...

Cap
The majority consensus (on this forum at least) was that this was likely going to be a Zen 3 level jump, not a Zen 1 level one. It wasn't until @adroc_thurston joined that it shifted, and even then there's still healthy skepticism of that higher level IPC claims
But even if that were true, it's from looking at previous generationS not the one current generation

Bad argument. GLC only got a 15% jump in server lmao. And SNC got 18% IIRC. The difference between 18 to 15% is that much lol. Plus, if you believe in the 15% rumor, I doubt you are as likely to believe in the frequency regression rumor as well, at least not a significant frequency regression. And SNC was a frequency regression over SKL, even RKL's 'SNC' version just tied SKL. Plus, the all core frequency regression was seen in both SNC ICL and CYPRC RKL. Also, GLC was technically a Fmax frequency regression over WLC, and that used a better node than WLC to boot (and the node difference there is prob bigger than the one between N4 and N5 for Zen 5 vs Zen 4).

Zen 1 was released in 2017, and there has been 5 gen of zen since, and every performance uplift has been increasing since the last. Compared that to intel and you'd see my point. Even if you ignore all that, Mike's comment about zen 5 should at least be something to be addressed. There were also rumors about the excitement behind the scene about zen 5, this has been addressed multiple times already.

Also, 12900k 13900k 14900k are the same damn thing and it's factually wrong to claim otherwise. The false equivalence in this case is about the rate of which intel and amd churning out performance upgrades with each gen is not the same.

 

[Markfw]

No it's not lol

So what you're saying is you look at a pattern of multiple generations of products released recently, not just the latest one?
The false equivalence is pretending @Markfw's comment made literally any sense at all. There is literally no relationship between Zen 4 going Vroom and thinking that Zen 5 is going to be a massive jump over Zen 4 lol. You can try to extrapolate the "pattern of multiple generations" claim from that, and that's correct, but that's not at all what he said, and not at all what I responded to him with...

Cap
The majority consensus (on this forum at least) was that this was likely going to be a Zen 3 level jump, not a Zen 1 level one. It wasn't until @adroc_thurston joined that it shifted, and even then there's still healthy skepticism of that higher level IPC claims
But even if that were true, it's from looking at previous generationS not the one current generation

Bad argument. GLC only got a 15% jump in server lmao. And SNC got 18% IIRC. The difference between 18 to 15% is that much lol. Plus, if you believe in the 15% rumor, I doubt you are as likely to believe in the frequency regression rumor as well, at least not a significant frequency regression. And SNC was a frequency regression over SKL, even RKL's 'SNC' version just tied SKL. Plus, the all core frequency regression was seen in both SNC ICL and CYPRC RKL. Also, GLC was technically a Fmax frequency regression over WLC, and that used a better node than WLC to boot (and the node difference there is prob bigger than the one between N4 and N5 for Zen 5 vs Zen 4).

So, people are not allowed an opinion unless they can prove it ? Go back to your Intel thread. I could be wrong about Zen 5, but my gut tells me its going to be good.

THATS MY OPINION !