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

Page 287 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

coercitiv

Diamond Member
Jan 24, 2014
6,257
12,192
136
It's similar to CPU scaling benchmarks for video games, which are usually performed at low resolutions to ensure there isn't a GPU bottleneck. The practical 'average consumer' argument would be "well at 1440p/4k CPU selection doesn't matter because all the CPUs will be within a few frames of each other". What's the point in even benchmarking then?
The practical 'average consumer' argument in this case would be 'I need the CPU that delivers at least 60-165FPS+ depending on needs' after the GPU bottleneck has been removed. The fact that you're pretending not to know this only detracts from the quality of the conversation.

That being said, it's interesting that you bring up video games in this discussion, considering many games have a hard time scaling strongly past 6-8 cores. Do you reckon game reviewers should intentionally favor games with high MT scaling in their roundups? We all know future titles will scale better with more cores, so it would appear benchmark quality would improve if we removed games with low thread scaling. I'm essentially asking the same thing you did for GB6, only for gaming benchmarks, one of the more common workloads for the average consumer.
 

Mopetar

Diamond Member
Jan 31, 2011
7,936
6,233
136
In that case why don’t AMD simply increase the number of wafers they order from TSMC? If you say AMD can use those extra wafers for more laptop chips, and that the market would absorb those chips, that is.

Wafers aren't like loaves of bread where you can just drop by the store and pick up another when you need it. AMD would have purchased the wafer allotments they are using now well before the present the time, and adding additional production would require first that some other company not wanting to use the wafer allotments it purchased and AMD being willing to pay more than anyone else who might also want them.

Even if they could get more wafers, if AMD could make more profit from using them for increasing their share in any other market segment, they'd do that first.

Chasing market share for its own sake is foolish and shortsighted. Apple could probably sell twice as many Macs by cutting the prices substantially, but then they wouldn't be nearly as profitable.
 

FlameTail

Platinum Member
Dec 15, 2021
2,356
1,276
106
That being said, it's interesting that you bring up video games in this discussion, considering many games have a hard time scaling strongly past 6-8 cores. Do you reckon game reviewers should intentionally favor games with high MT scaling in their roundups? We all know future titles will scale better with more cores [...]
Are we talking physical or logical cores?

When you say games don't scale well beyond 6-8 cores, is the implied meaning that they don't scale well beyond 12-16 threads?

Lion Cove will drop Hyperthreading in Arrow Lake, and consumer ARM CPUs don't have SMT anyway.

Like, let's take the Snapdragon X Elite and Ryzen 7600U.

X Elite: 12 cores/12 threads
7600U: 6 cores/12 threads

Statement A: Games don't scale well beyond 6 physical cores

Statement B: Games don't scale well beyond 12 threads

If Statement A is true, then the 7600U should be faster in games.

If Statement B is true, then X Elite should be faster in games (since physical cores >> logical cores, in terms of performance). *This is of course is purely theoretical, so it excludes game compatibility issues with X Elite.
 

FlameTail

Platinum Member
Dec 15, 2021
2,356
1,276
106
Wafers aren't like loaves of bread where you can just drop by the store and pick up another when you need it. AMD would have purchased the wafer allotments they are using now well before the present the time, and adding additional production would require first that some other company not wanting to use the wafer allotments it purchased and AMD being willing to pay more than anyone else who might also want them.

Even if they could get more wafers, if AMD could make more profit from using them for increasing their share in any other market segment, they'd do that first.

Chasing market share for its own sake is foolish and shortsighted. Apple could probably sell twice as many Macs by cutting the prices substantially, but then they wouldn't be nearly as profitable.
TSMC N5/N4 nodes are not running at full utilisation currently (?). So AMD could buy more capacity yes?
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,804
3,260
136
My god man, quit abusing the tool and put that $@#% into a database.
Poor excel, it should not have to stand for such abuse!
This from an excel user =)
The src is a db that has a front end web app , we have scenarios where people don't have inet access but the data is important/useful. So yes in this case excel is an offline database....lol
 
  • Like
Reactions: lightmanek

gdansk

Platinum Member
Feb 8, 2011
2,212
2,836
136
The src is a db that has a front end web app , we have scenarios where people don't have inet access but the data is important/useful. So yes in this case excel is an offline database....lol
there's a few neato databases like sqlite and duckdb for which presumably the same front end could be used offline though they'd have to run a web server locally (stranger things have been done).

anyway I think it is a proven fact: there's no getting business away from misuing Excel except sometimes managers want to misuse Google Sheets too
 
  • Like
Reactions: lightmanek

itsmydamnation

Platinum Member
Feb 6, 2011
2,804
3,260
136
there's a few neato databases like sqlite and duckdb for which presumably the same front end could be used offline though they'd have to run a web server locally (stranger things have been done).

anyway I think it is a proven fact: there's no getting business away from misuing Excel except sometimes managers want to misuse Google Sheets too
you dont want to know how many million dollar mistakes my program ( not even company ) has had from bad excel.............
 

Fjodor2001

Diamond Member
Feb 6, 2010
3,842
305
126
Wafers aren't like loaves of bread where you can just drop by the store and pick up another when you need it.
I never said so either. Of course they must order wafers well in advance, juts like everyone else.
Even if they could get more wafers, if AMD could make more profit from using them for increasing their share in any other market segment, they'd do that first.
One does not rule out the other. They can order enough wafers to cover both the additional laptop chips, and whatever other chips you think they can make sufficient profit from.
Chasing market share for its own sake is foolish and shortsighted. Apple could probably sell twice as many Macs by cutting the prices substantially, but then they wouldn't be nearly as profitable.
This is the main point. So you mean AMD could only gain market share in the laptop segment by entering into a price war? I.e. not because their laptop CPUs are good anough to sell more of at their current pricing?

Then that changes everything, and the claim that AMD could sell more laptop chips if they just had more wafers is kinda moot. Everyone can sell more if they drop the price low enough. If they are prepared to take a loss that they can sell even more.

But currently, my takeaway from what you wrote is that the AMD vs Intel CPU market share in the laptop segment is fair at current pricing.
 

AMDK11

Senior member
Jul 15, 2019
290
197
116
As for Zen5, we have two leaks that say the IPC increase is:
1. 10-15%+
2. 30%.

Does this data have to contradict each other? I think not. Many assume that a 30% IPC increase will cover most workloads or that it is an average of the growth curve.
I'll use the Zen3 IPC growth curve as an example:
IPC-SPEC.png

Let's assume a hypothetical slide leak is about Zen 3 which describes a 15%+ IPC increase and another MLID leak says +36%.
Are both data incorrect?

AMD provided an estimated average from the IPC growth curve and MLID provided an example from a given workload type.

Regarding MLID, I'm not saying he's right, I just gave it as an example. As for Zen3, if someone even whispers to him the truth that the IPC gain is 46% or 109%, will it be a lie when he starts talking about it?
I think many would consider such an increase to be typical or average of the growth curve and translate it into the performance of CB or other software.
He admitted he had no knowledge, so what he said, if not absolute fabrication, was a matter of interpretation.

For Zen5, I think expecting an average IPC growth curve of 30% is fantasy. But there will definitely be loads, whether INT or FP, that will show an IPC increase of 30% or even more. I have no doubt about that. What is supposedly coming from AMD slides, i.e. Zen5 10-15%+ is closer to the actual average IPC growth curve and I think the target is +15-20%.
 
Last edited:

AMDK11

Senior member
Jul 15, 2019
290
197
116
I guess you mean it’s MLID that claims 10-15%. But which leak is it that claims 30%?
Yes, you are right that this slide showing Zen 5 and Zen 6 supposedly from AMD publishes the MLID.

As for the 30% higher IPC, maybe it's not the MLID, although I don't know why I assumed that was leaking from it. Either way, there are rumors floating around the internet about +30%.

All in all, it doesn't change the tone of my statement. Regardless of which point on the IPC growth curve we draw from the leaks, we may still be close to the truth, and the question is how to interpret the numbers someone provides (regardless of whether they were made up or not).

The Zen3 IPC growth curve ranges from +1% to +46% (one load +109%), and the average of this curve is +19%.

The Zen4 IPC growth curve ranges from +1% to +39%, and the average of this curve is +13%.
website


Someone (I mean leaks) saying that the increase in IPC is, among other things, +30% or +20-30% means nothing if we do not know whether this increase is within the average IPC growth curve. Without knowing what specific type of load they refer to, these are rather empty numbers.
 
Last edited:

itsmydamnation

Platinum Member
Feb 6, 2011
2,804
3,260
136
Yes, you are right that this slide showing Zen 5 and Zen 6 supposedly from AMD publishes the MLID.

As for the 30% higher IPC, maybe it's not the MLID, although I don't know why I assumed that was leaking from it. Either way, there are rumors floating around the internet about +30%.

All in all, it doesn't change the tone of my statement. Regardless of which point on the IPC growth curve we draw from the leaks, we may still be close to the truth, and the question is how to interpret the numbers someone provides (regardless of whether they were made up or not).

The Zen3 IPC growth curve ranges from +1% to +46% (one load +109%), and the average of this curve is +19%.

Someone (I mean leaks) saying that the increase in IPC is, among other things, +30% or +20-30% means nothing if we do not know whether this increase is within the average IPC growth curve. Without knowing what specific type of load they refer to, these are rather empty numbers.

isnt the MILD data SIR for a platform that has significantly more core then the predecessor but not significantly more memory bandwidth?
 

AMDK11

Senior member
Jul 15, 2019
290
197
116
isnt the MILD data SIR for a platform that has significantly more core then the predecessor but not significantly more memory bandwidth?
RAM bandwidth is a limit for 16 cores, but not for single-threaded workloads. Besides, there are always some bottlenecks even in the core microarchitecture itself, which will also be in Zen5, and some of them will be removed in Zen6 and subsequent generations. And so on without end.
 
Last edited:

AMDK11

Senior member
Jul 15, 2019
290
197
116
I really hope this is not what adroc means when he says he would sound schizo if he told us how fast Zen 5 really is.
I guess that's what he meant, unfortunately it's true, we don't know the average Zen5 IPC growth curve.

I am 99% sure that the curve will also start at +1-3% and end somewhere at +40-50% with possibly a minor exception.

I won't be at all surprised if the average IPC growth curve is around 15%. It is possible, although it may be disappointing for many.

An average IPC increase of 25-30% would be an exceptional and very welcome breakthrough. But there's no need to delude yourself.
 
Last edited:

Mopetar

Diamond Member
Jan 31, 2011
7,936
6,233
136
TSMC N5/N4 nodes are not running at full utilisation currently (?). So AMD could buy more capacity yes?

Assuming they have the capacity for packaging existing chips, then yes. I don't think we'd see more mobile chips though since it's not like AMD can shift excess into the DIY market. Unless Dell, Lenovo, etc. want to contract for more of those chips AMD doesn't have much reason to produce more of them. Especially when other products have better margins.
 

RnR_au

Golden Member
Jun 6, 2021
1,737
4,255
106
isnt the MILD data SIR for a platform that has significantly more core then the predecessor but not significantly more memory bandwidth?
Thought it was @adroc_thurston mentioning +32% for SIR? Or am I misremembering?

(I mean come on... 287 pages for a cpu that hasn't even been released yet... this thread is hard to keep up with... :p )
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,804
3,260
136
Thought it was @adroc_thurston mentioning +32% for SIR? Or am I misremembering?

(I mean come on... 287 pages for a cpu that hasn't even been released yet... this thread is hard to keep up with... :p )
He said 32 for spec int 1t, but then we yolo'd to 41% , and that's the current position of the hype train that I'm all aboard for.....
 

H433x0n

Senior member
Mar 15, 2023
926
1,013
96
I’ve had more than 1 person tell me that CB R23 1T is >=2800 which is a >=40% increase. If I had to take a geomean of Zen 5 leakers it’d probably be a 40% 1T perf increase overall.

I don’t personally believe that but it’s at least consistent. The only person saying it’s not hype™ is MLID who’s track record is spotty. Although he seemed to get the details about the Zen 5 delay and 800 series chipset right .. so ymmv.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,804
3,260
136
The AMD hype train giveth... and taketh away. :cry:
im still waiting for the low ballers to explain how from zen 1 to 4, core width didn't grow and OOOE window increased a small amount ( especially relative to others ) while doing ~ 50% more ipc and yet going to 6 wide plus a whole new fronted we know nothing about except its big shinny and new and we are getting -5 to 15% with clock regression.....

its not like there arent cores that size with that much more IPC on the market.