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

[DisEnchantment]

Speculate at will

 

[adroc_thurston]

Not to the level it was in 2020 (overall).

Will take a while for PC units to return there.
Not a relevant metric for DIY tho.

 

[Fjodor2001]

And you even cherry-picked a much cheaper 8core just to It would look better in comparison

Yes, because you started by intentionally cherry-picking the most expensive Zen1 8 core CPU (1800X) to make it look worse in comparison, so I thought I'd return the favour in the opposite direction.

 

[Fjodor2001]

He didn't mean It as you think. I will try to explain It to you.

How can a single core offer for example 30% higher performance but the rest of the cores stay the same compared to Zen4?
The only possibility is If the turboboost for a single core increased from 5.7GHz to 7.4GHz.
This is already very unrealistic, because that clock is just too high.
The other illogical thing would be to expect one core to clock that high, but using two or more would result in the same clocks as Zen4 has.

BTW @adroc_thurston already said on this forum that Zen5 won't increase clocks compared to Zen4.
Then how can a single core become faster? Only by higher IPC.

If IPC of one core increased by 30% then what do you think will happen to the rest of the cores and the performance of the whole chip?
Of course the MT performance will increase by about the same amount unless power consumption will hit the power limit resulting in lower performance because of lowered clockspeed.

Core count is also important for MT, but It's limited by the number of threads used by soft.
I will give you a simple example, there are 2 CPUs.
CPU A: 16C32T 5.6GHz turbo, 5GHz all-core clock, + 50% IPC compared to Zen4
CPU B: 24C48T 5.6GHz turbo, 5GHz all-core clock, the same IPC as Zen4
What is the performance?
At 48 Threads they would perform the same, because CPU A can execute only 32 threads at once, but at fewer threads CPU A will be faster and at 1-16T the difference will be 50% faster.

P.S. I don't know If what @adroc_thurston said is true or not. We will see after Zen5 is out.

I'm fully aware of that basic stuff and agree with it, so no need trying to convince me.

In essence, better to have 1 core performing X*Y than X cores each performing Y, all else equal. Problem is it's not technically possible to create such a 1 core CPU!

We've already hit the frequency wall more or less, and IPC can only be increased so much. However the node shrinks still provide more transistors per die area, which can be used to add more cores. Also, any IPC increase will apply to all cores anyway (assuming identical core types are used)!

So 2x amount of cores will bump MT performance 100% (minus some performance loss e.g. due to lower frequency since less Watts available per core). Bumping MT performance by increasing IPC by 100% would of course be even better, but it's simply not possible to achieve in a short time frame. Example: 20% IPC increase per CPU generation, for 4 CPU generations, would result in 1.2^4=2.07 (i.e. ~100% higher performance). If we assume 2 years per CPU generation, that's 8 years! And again, nothing prevents you from increasing core count too anyway. It's not either or.

Now with regards to how many threads can be fully loaded at the same time, that will depend on the workload. But if you're already using 32T like you mentioned, chances are quite high that your workload will also be able to make use of 48T. And for those that will not need 48T or even 32/16T, they can always buy lower end CPU variants with less cores.

 

[A///]

hasan's rag has published some semi confirmation that z5 uses the same ciod by leakers.

 

[Saylick]

hasan's rag has published some semi confirmation that z5 uses the same ciod by leakers.

That's been kinda confirmed via this thread here, I believe.

 

[TESKATLIPOKA]

Yes, because you started by intentionally cherry-picking the most expensive Zen1 8 core CPU (1800X) to make it look worse in comparison, so I thought I'd return the favour in the opposite direction.

I wrote this:

8C16T Ryzen 7 1800X: $499 (100%)
16C32T Ryzen 9 3850X: $749 (150%)
16C32T Ryzen 9 5950X: $799 (160%)
16C32T Ryzen 9 7950X(X3D): $699 (140%)

They are the fastest desktop offering from each generation.
Where is some intentional cherry-picking in that?

 

[Fjodor2001]

I wrote this:

They are the fastest desktop offering from each generation.
Where is some intentional cherry-picking in that?

You "forgot" to mention that it was possible to get 8 cores for $329 with the AMD Zen1 1700 in 2017. No need to buy the most expensive Zen1 variant being 1800X at $499 which also had 8 cores.

That said, I do agree it's fair if you're only intention is to calculate some price index for AMD top-of-the-line desktop CPUs though. But putting it into a wider context by also including the competition (Intel which only had 4C/8T at the time), I think the much lower priced 8 core AMD 1700 should also be mentioned. The AMD 1700 was really revolutionary in this regard, since at that time to get an 8 core Intel CPU you had to buy into the Intel HEDT platform at a much higher cost. I'd like to see a similar "core count revolution" happening again soon, regardless of whether it's from Zen5 (or later Zen version), or Intel for that matter.

 

[A///]

That's been kinda confirmed via this thread here, I believe.

yeah by adrock thurston, though people seemed to be questioning his posts lately

 

[TESKATLIPOKA]

You "forgot" to mention that it was possible to get 8 cores for $329 with the AMD Zen1 1700. No need to buy the most expensive Zen1 variant being 1800X at $499 which also had 8 cores.

In that comparison, I was giving proof that your expectation about Zen5 having 50% more cores for the same $699 price is unrealistic. That's why I provided top desktop parts from each generation, because this 24C Zen5 would also be the top product of that generation.

So please enlighten me why I should have mentioned this cheaper R7 1700? There was clearly no reason to mention It in my comparison.
Yet you accuse me of cherry-picking or intentionally forgetting something.

 

[Fjodor2001]

In that comparison, I was giving proof that your expectation about Zen5 having 50% more cores for the same $699 price is unrealistic. That's why I provided top desktop parts from each generation, because this Zen5 would also be the top product of that generation.

So please enlighten me why I should have mentioned this cheaper R7 1700? There was clearly no reason to mention It in my comparison.

Read my full post, not only part of it.

 

[igor_kavinski]

Only mobile can be a bit of an issue since Intel is very keen about flooding the market with cheapo RPL 282 parts.

Bad news is, I've tried using an RPL laptop with only two P-cores in Windows 11. Performance is pretty snappy though not sure if the user suffers when there are lots of background apps open with dozens of tabs. But based on 5 minutes of launching apps on such a laptop in a store, I would buy it in a heartbeat, especially if I didn't know it had only two P-cores.

 

[TESKATLIPOKA]

Read my full post, not only part of it.

Here is the whole post.

AMD revenue drops 9% as PC chip sales decline sharply

AMD's report comes as the PC industry is in a deep slump, with shipments dropping 30% in the first quarter, according to IDC.

www.cnbc.com

"AMD revenue drops 9% as PC chip sales decline sharply"

"AMD’s report comes as the PC industry is in a deep slump, with shipments dropping 30% in the first quarter, according to IDC."

Huh? 7950X was $699 at launch. As I said, a potential 24C Zen5 (whether there will be such a CPU or not) should be priced the same, i.e. $699 and not $999. And 16C Zen5 lower.

AMD had stayed on max 16C for standard desktop for far too long. They're going to be run over by Intel soon in MT perf if they do not increase core count. They need to bump core count just like they did with Zen (8C) and Zen2 (16C). AMD has stagnated since then.


Of course Zen5 have better perf than Zen4. Why release it otherwise?

But people are not prepared to pay the amount of money for it that you think. They'll buy Zen3/Zen4 or Intel instead. Also, Zen5 will soon after release be competing with Intel Arrow Lake (8P + 32E), which potentially will stomp on Zen5 in MT workloads.

The bold one wasn't quoted at that time, so I added It this time.

I reread It and? What do you want to say exactly?
That AMD should add more cores in Zen5 for the same price because a hypothetical Arrow Lake (8P + 32E) will be faster when using 32 vs 48 threads? Or that they are using 16 cores for too long?

edit:
I don't see a big problem with Zen5 staying at only 16 cores, not that many people need more than that. If someone needs more, then just buy a server(workstation) one from AMD or Intel, which has more cores.
The problem for me would be If desktop Zen5 was overpriced compared to what It brings to the table. For example $999 would be very expensive for a 16C32T even If performance would be 50% better.

 

[Fjodor2001]

Double post. Read next one below instead. Sorry!

 

[Fjodor2001]

Here is the whole post.

The bold one wasn't quoted at that time, so I added It this time.

I reread It and? What do you want to say exactly?
That AMD should add more cores in Zen5 for the same price because a hypothetical Arrow Lake (8P + 32E) will be faster when using 32 vs 48 threads? Or that they are using 16 cores for too long?

edit:
I don't see a big problem with Zen5 staying at only 16 cores, not that many people need more than that. If someone needs more, then just buy a server(workstation) one from AMD or Intel, which has more cores.
The problem for me would be If desktop Zen5 was overpriced compared to what It brings to the table. For example $999 would be very expensive for a 16C32T even If performance would be 50% better.

I meant the post which you last responded to, i.e. this one.

When you responded, you only quoted the first paragraph in it, and judging by your response it seemed like you hadn't even read the second paragraph either.

 

[Saylick]

Bad news is, I've tried using an RPL laptop with only two P-cores in Windows 11. Performance is pretty snappy though not sure if the user suffers when there are lots of background apps open with dozens of tabs. But based on 5 minutes of launching apps on such a laptop in a store, I would buy it in a heartbeat, especially if I didn't know it had only two P-cores.

Yeah, my work laptop is 2+8 RPL and it's surprisingly snappy.

 

[Fjodor2001]

I don't see a big problem with Zen5 staying at only 16 cores, not that many people need more than that. If someone needs more, then just buy a server(workstation) one from AMD or Intel, which has more cores.

Well, if you want to elaborate along those lines, how many people need more performance than 7950X provides anyway, regardless if whether it's from IPC or core count increase?

I think you can argue that those users that actually need more performance than that are often users with very MT heavy workloads, so they do not care if the higher MT performance comes from IPC or core count increase. And in that case, if we're saying there is not demand for higher MT performance via core count increase, why should there be demand for higher MT performance via IPC increase?

The problem for me would be If desktop Zen5 was overpriced compared to what It brings to the table. For example $999 would be very expensive for a 16C32T even If performance would be 50% better.

Agreed. I wonder how many users will be prepared to pay $999 for a speculated 16C/32T 8950X with ~30% higher IPC than 7950X, if they can get 7950X for about half the price.

 

[adroc_thurston]

Bad news is, I've tried using an RPL laptop with only two P-cores in Windows 11. Performance is pretty snappy though not sure if the user suffers when there are lots of background apps open with dozens of tabs. But based on 5 minutes of launching apps on such a laptop in a store, I would buy it in a heartbeat, especially if I didn't know it had only two P-cores.

Yeah they're goodenough parts and Intel sells 'em for cheap.

 

[Thibsie]

Yeah, my work laptop is 2+8 RPL and it's surprisingly snappy.

It is. Unless you go into workloads like virtualization. Then you pretty quickly regret it...

 

[adroc_thurston]

It is. Unless you go into workloads like virtualization. Then you pretty quickly regret it...

Well, no one really runs VMs on 282 ultrabooks.

 

[H433x0n]

Bad news is, I've tried using an RPL laptop with only two P-cores in Windows 11. Performance is pretty snappy though not sure if the user suffers when there are lots of background apps open with dozens of tabs. But based on 5 minutes of launching apps on such a laptop in a store, I would buy it in a heartbeat, especially if I didn't know it had only two P-cores.

How is that bad news? It's good news overall since this seems to be where the industry is heading (both Intel & eventually AMD) so having a relatively seamless experience is bullish for future processors and Windows' OS.

 

[igor_kavinski]

How is that bad news?

We went from 4 muscle cores to 2 muscle cores plus 8 pathetic cores. It's a regression. Games would suffer the most on just two real cores.

 

[adroc_thurston]

Games would suffer the most on just two real cores.

Not a gaming chip, whatever

 

[igor_kavinski]

Not a gaming chip, whatever

Casual gaming would be worse off with that chip. Admit it. It's not good. 4P+4E or 4P+6E would be more balanced and useful for everyone. Intel got away with selling an inferior/smaller chip for the same price.

 

[adroc_thurston]

Casual gaming would be worse off with that chip.

You don't game on 282.

4P+4E

Guess what exactly you're getting in LNL-M!

 

[mmaenpaa]

Bad news is, I've tried using an RPL laptop with only two P-cores in Windows 11. Performance is pretty snappy though not sure if the user suffers when there are lots of background apps open with dozens of tabs. But based on 5 minutes of launching apps on such a laptop in a store, I would buy it in a heartbeat, especially if I didn't know it had only two P-cores.

I have installed a few 2P+8E core laptops (i5-1235U). Various SW installation feeling was anything but snappy compared to Ryzen 6/8 core offerings). i7-13700H laptop was snappy, but it is of course another category (6P+8E cores).
Based on my own subjective experience of course.