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

[DisEnchantment]

Speculate at will

 

[gdansk]

For the same price I would choose dual vcache over a slight speed bump on the single CCD. But as I've said before I'm sure AMD could sell a dual vcache 9970X3D for +$150-200 over a 9950X3D.

It would be an inferior gamer part than even the 9800X3D. And inferior 1T to 9950X. And an inferior MT to 9950X. But cost twice as much. So what's it for? And who is it for?

 

[Win2012R2]

Overclocking does not correlate with what AMD will ship.

Would 3D chiplet in 16 core version be slower than in 8 cores? I can't see how they'd allow it to happen, and if so, I'd rather get both chiplets with 3D even if they both run at max 5.4 Ghz, that's fine by me.

It would be an inferior gamer part

Like any company AMD wants to upsell to have higher ASP - one way to do it is to put better 3D chiplet into more expensive SKU. There will be millions of gamers buying 2.5-3k 5090 and they'd want to pair it with top notch CPU, the best there is - adding $300 on top of $700 is a no brainer for them.

If 2nd non-3D chiplet got 5% faster clocks (5.7 vs 5.4) then overall that's like 2% diff for whole chip, totally nothing for risks of bad thread management, which was the case in 7950x3d and made me turn off non-3D chiplet and thus lose HALF of MT, so no, I'd take both 3D chiplets even if they are 5.4 Ghz max each.

 

[gdansk]

then overall that's like 2% diff for whole chip

It reduces the "uplift" in 1T from 1.14x (allegedly, more like 1.1x) to 1.07x. Why not buy a much cheaper 7950X or 7950X3D or 9950X or 9800X3D at that point.

It would be denounced, rightfully, as a pointless money grab part. AMD shouldn't hurt their reputation in the only place they have a positive reception.

 

[Win2012R2]

AMD shouldn't hurt their reputation

It reduces the "uplift" in 1T from 1.14x (allegedly, more like 1.1x) to 1.07x

People who must have max frequency can get 9950X then, or 7950X cheaper if they so desire.

The target market for 3D chips was trained that frequency will be lower, which still works out for specific workloads they care about (games), so I don't see a problem - release it as 2nd SKU, yes it will be more expensive because of two 3D cache, that is rational and won't affect AMDs rep in any bad way, on the contrary - they will have offered CHOICE.

 

[gdansk]

People who must have max frequency can get 9950X then, or 7950X cheaper if they so desire.

The target market for 3D chips was trained that frequency will be lower, which still works out for specific workloads they care about (games), so I don't see a problem - release it as 2nd SKU, yes it will be more expensive because of two 3D cache, that is rational and won't affect AMDs rep in any bad way, on the contrary - they will have offered CHOICE.

It is worse than the 9950X3D. 9950X3D will game just as well and offer better 1T performance. There's no point for a dual X3D part. It costs more and has worse performance. That's why it won't exist. People keep asking for a pointless product. Consumers are perhaps actually genuinely against rational purchasing decisions but I doubt it.

What would actually be an improvement while using the same amount of silicon would be taking both of those 3D caches and stacking them on a single die. But TSMC/AMD aren't doing that.

 

[Win2012R2]

9950X3D will game just as well

Well my 7950X3D did not game well until I disable non-3D chiplet, you mention pointless product, but it's a lot more pointless to have to disable half of it because AMD's and Windows in general scheduling sucks.

 

[gdansk]

Well my 7950X3D did not game well until I disable non-3D chiplet, you mention pointless product, but it's a lot more pointless to have to disable half of it because AMD's and Windows in general scheduling sucks.

You can solve it in software. Core pinning is easy (even in backwards/retarded operating systems like Windows). I don't expect reviewers to get it right but I would expect someone posting on a hardware forum to know it or to learn it quickly.

 

[Win2012R2]

You can solve it in software.

Yes I can, and I do in my software which has no such problems, but when I game I want peace of mind and not having to fix stuff in expensive hardware that I bought - it should just work!

Plus I want both chiplets to have 3D for deterministic performance (and I don't like the idea of non-3d-chilet creating traffic to 3D chiplet which could have been satisfied there in the first place), and I am prepared to pay for it, where as you are giving lots of excuses for AMD not to make extra money - this isn't designing new top end GPU that won't sell - they already got chiplets and they already produce 16 core version, nothing stops them from cheaply doing limited runs of $999 halo SKU, heck - call it EPYC 5005 if you must.

 

[biostud]

You can solve it in software. Core pinning is easy (even in backwards/retarded operating systems like Windows). I don't expect reviewers to get it right but I would expect someone posting on a hardware forum to know it or to learn it quickly.

The point is, if there is a market for it, and they can sell and earn money doing so, why wouldn't they do it? I mean what is the point of the 14900KS, if not catering for the same buyers who would gladly buy a 9970X3D?

 

[Hail The Brain Slug]

Yes I can, and I do in my software which has no such problems, but when I game I want peace of mind and not having to fix stuff in expensive hardware that I bought - it should just work!

Plus I want both chiplets to have 3D for deterministic performance (and I don't like the idea of non-3d-chilet creating traffic to 3D chiplet which could have been satisfied there in the first place), and I am prepared to pay for it, where as you are giving lots of excuses for AMD not to make extra money - this isn't designing new top end GPU that won't sell - they already got chiplets and they already produce 16 core version, nothing stops them from cheaply doing limited runs of $999 halo SKU, heck - call it EPYC 5005 if you must.

Given how there have been rumors and leaks of both configurations, I wouldn't be surprised if they release both. They can limit the production of the dual V$ part just to satisfy the small but rabid market for it, winning mindshare.

 

[Win2012R2]

The point is, if there is a market for it

I can see only one (other than incompetence) reason - they don't want competition in server market, but it's only 16 cores here and they skip 3D in Turin anyway, just brand it as EPYC 5005 and sell - don't double dip though, offer it at the same time.

 

[biostud]

Yes I can, and I do in my software which has no such problems, but when I game I want peace of mind and not having to fix stuff in expensive hardware that I bought - it should just work!

Plus I want both chiplets to have 3D for deterministic performance (and I don't like the idea of non-3d-chilet creating traffic to 3D chiplet which could have been satisfied there in the first place), and I am prepared to pay for it, where as you are giving lots of excuses for AMD not to make extra money - this isn't designing new top end GPU that won't sell - they already got chiplets and they already produce 16 core version, nothing stops them from cheaply doing limited runs of $999 halo SKU, heck - call it EPYC 5005 if you must.

Exactly, I would think that those people who run game servers would like dual cache CPUs as well, as you say people who want EPYC large cache models in their desktop.

 

[gdansk]

I want both chiplets to have 3D for deterministic performance

Tiny niche. There is no market.

And dual vcache will be worse in gaming because some games prefer clock rate (and less cache latency) to more cache. With mixed CCD you can get the best of both, though you have to know which CCD games you play prefer. Mixed is genuinely the better & cheaper configuration. Its only weakness is Windows but that can be worked around.

 

[Win2012R2]

Mixed is genuinely the better configuration

Maybe it was, but not now that frequency delta is very small - 5.4 vs 5.65 is not the same as 4.85 vs 5.65, the downside of uneven chiplets is worse than any possible tiny upside.

 

[Hail The Brain Slug]

Maybe it was, but not now that frequency delta is very small - 5.4 vs 5.65 is not the same as 4.85 vs 5.65, the downside of uneven chiplets is worse than any possible tiny upside.

7950X3D V$ FMax was 5.25

 

[Win2012R2]

7950X3D V$ FMax was 5.25

Single core, when I load all of them I am getting 4.85 ghz, either way 5.4-5.5 is a lot closer to 5.65 on another chiplet, is 2-3% extra perf worth uneven chiplets? Not for me

 

[Hail The Brain Slug]

Single core, when I load all of them I am getting 4.85 ghz, either way 5.4-5.5 is a lot closer to 5.65 on another chiplet, is 2-3% extra perf worth uneven chiplets? Not for me

Fair enough, that's part of why 7000 X3D was so great at efficiency, they ran the V/F curve super lean and it hurt all core clocks a bit.

9000 X3D gets a nice, juiced V/F curve to waste as much power as it wants, so all core clocks on the 9950X3D should be excellent.

 

[gdansk]

Maybe it was, but not now that frequency delta is very small - 5.4 vs 5.65 is not the same as 4.85 vs 5.65, the downside of uneven chiplets is worse than any possible tiny upside.

It's 5750MHz still. It remains the genuinely better and cheaper configuration. The complaint is that some people are too lazy to run software so they'll pay more for worse.

It's possible customers do want to pay more for a worse config but I don't think AMD will heed their supplications when it increases BoM and consumes two possible 9800X3D and 9950X3D for only a few hundred dollars more.

 

[Win2012R2]

some people are too lazy to run software so they'll pay more for worse.

Yes, exactly - people will pay for convenience, that happens a lot in many industries, the question is why should not AMD oblige those people and make a quick buck, maybe invest this easy lazy money into Radeon. Can AMD afford to be so lazy not to take my money?

 

[gdansk]

Yes, exactly - people will pay for convenience, that happens a lot in many industries, the question is why should not AMD oblige those people and make a quick buck, maybe invest this easy lazy money into Radeon. Can AMD afford to be so lazy not to take my money?

The supply constraint is 3D cache chiplets. And they can't keep the 9800X3D in stock.

The niche is so small it isn't worth pursuing until very, very late in the product cycle.

 

[Win2012R2]

The supply constraint is 3D cache chiplets. And they can't keep the 9800X3D in stock

Sell dual 3D ones for $999 then while the market can take it, cream off it whilst keeping 9800x3d out of stock just like Nvidia did with 3080 where as 3090 was buyable.

And make more bloody 3D cache - it's old 6-7nm tech, clearly next gen it might have to become almost standard feature as Intel will catch up, at least this should be assumed, oh - and fire those whose job was to ensure increased supply of those chiplets - it's not like it wasn't predictable that these CPUs will be in high demand.

 

[Joe NYC]

It is not. X3D chiplets are only marginally higher clock rate than last generation, they're still behind.

We will see what the the limit of the x3d chiplet is in 9950x3d. It may be higher than 9800x3d

 

[Win2012R2]

It may be higher than 9800x3d

It has to be if they want to upsell people and get rid of non-3d consumer chiplets that otherwise won't sell as well.

The only logical reason why they have not launched 16 core version before Xmas is to bin better.

 

[gdansk]

It isn't happening in 2025 and 99.5% people asking for it are wrong to even want that configuration.

The rumor made some sense when it was thought 3D cache CCDs would clock the same but they are 300-400MHz behind so it is an inferior product that costs more to make. Until they're drowning in 3D cache CCDs there is no reason to make it.

 

[Joe NYC]

It isn't happening in 2025 and 99.5% people asking for it are wrong to even want that configuration.

I don't think the shortages are going to persist too far into 2025. Micro Center, for example, (which probably had some special deal with AMD) is now right now fully stocked in all the stores I checked.

I think the reason people want the 2 V-Cache configuration is because it would be a non-brainer for gaming, would likely beat 9800x3d in pretty much all cases (if the clock speed was just a little higher).

Unlike 7950x3d which often loses to 7800x3d in gaming, in number of games, despite higher clock speed on the V-Cache chiplet vs. 7800x3d.

The rumor made some sense when it was thought 3D cache CCDs would clock the same but they are 300-400MHz behind so it is an inferior product that costs more to make. Until they're drowning in 3D cache CCDs there is no reason to make it.

Narrowing the clock speed deficit from 500 MHz (7800x3d vs 7700x) to 300 MHz (9800x3d vs 9700x) was enough to make 9800x3d approx. equal in to 9700 in non-gaming benchmarks. Which would predict similar behavior in 2 CCD, 2 V-Cache configuration.