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

[DisEnchantment]

Speculate at will

 

[aigomorla]

REAL world record

Is asking for proof on world records not allowed?
Do we have a "Realistic World Record" or a "Weighted World Record" or a "politically correct world record"?

Id like to see a World Record.

In what universe does using LN2 constitute a REAL world record ?

When it attains what the definition describes.

World record - Wikipedia

en.wikipedia.org

Why do you guys need to be so stuck up to keep calling it foul when it did exactly what it did?
This is what aggravates me.
Just say YES, and we all can move along.

 

[soresu]

Just wait a bit until Storm Peak (or w/ever SP5/6 Threadrippers are called) and then you'll be able to LN2 pot the thing as well

True, but you don't get dual socket Threadrippers so it might be difficult to match the MT performance of a 2S Genoa system.

 

[Thunder 57]

Decode is the most irrelevant part ever for fat opcache cores and you know it.
Stop being so obnoxiously silly.

Once again, you don't have to be so obtuse.
Well, people working at DCAI are all like you but I digress.

Well tell us something dude. You keep saying Apple like. We all know Zen 5 will be wider. How much? 5 wide? 6? 8? Give us a hint of something and people might listen. Calling people obnoxious and obtuse when challenged doesn't give you credibility.

 

[aigomorla]

True, but you don't get dual socket Threadrippers so it might be difficult to match the MT performance of a 2S Genoa system.

its been done..

Dual Ryzen Threadripper Pro 3995WX CPU Benchmark Appears

Dual socket Threadripper motherboards could be coming soon

www.tomshardware.com

But who knows we may get a dual storm peak, but AMD has been really tight about trying to keep HEDT away from EYPC.

 

[Abwx]

Sigh... can you please tell me what a World Record is and means?
To my understanding, it meant top record in the world regardless of what means you went to achieve it does it not?
And i am a intel supporter, im also a AMD supporter, and a Nvidia supporter.
I don't take sides, and always keep that line of neutrality at all times.

96 cores against 60 cores and with comparable IPC, i think that the conclusion doesnt even need more than a fraction of a second to come to the 100% sure conclusion.

FI Mark posted a CB run of about 83 000 pts at stock 360W with a 96C CPU, can you tell us how much a STOCK 60C 8490H SPR does.?.

Here it is, 85 000 pts for 2 x 8490H, so how much for a single CPU..?..

Intel Xeon ‘Sapphire Rapids’ vs AMD Ryzen Threadripper Pro

Intel launches its long awaited ‘Sapphire Rapids’ workstation processors, but do they have enough to surpass AMD’s Ryzen Threadripper Pro?

develop3d.com

 

[adroc_thurston]

True, but you don't get dual socket Threadrippers so it might be difficult to match the MT performance of a 2S Genoa system.

Eh various XOC leaderboards usually have 1p divisions so it's all gucci still.

Well tell us something dude

It's literally this, that's how you build an ultrawide core.
Apple just arrived there first (and got stuck there, for obvious reasons).

 

[aigomorla]

96 cores against 60 cores and with comparable IPC, i think that the conclusion doesnt even need more than a fraction of a second to come to the 100% sure conclusion.

But is that a world record?
Does it claim world record?
Has it been recorded as the world record?

Cinebench - R23 Multi Core with BenchMate overclocking records @ HWBOT

Overclocking records

hwbot.org

Does it satisfy the definition of World Record?
Your claiming "oh data shows..."

Data shows i'll never win the lottery either... but will i stop trying?

Can we just say OK... they don't have the world record, and move on.
This is honestly all i am asking for.

Your insulting the many hours these guys spend trying to get that crown.
I really do not appreciate that, especially when i used to be on that bench table.
This is why i am holding my ground firm on this.
If someone came to you and called your hobby junk / fake / so unrealistic and unproductive... how would you honestly reply?

And this whole rabbid intel supporter...
I really think SPR is a pile of dung... especially with a refresh coming out end of this year.
Im also holding out for Storm Peak, before i finalize my HEDT.
But will i give credit to where its due? Yes.
AMD lost its WR, which was held by 9474F, to SPR.
Can we stop pretending they still hold a WR, and just say its a better performer in general without the words WORLD RECORD.

 

[Markfw]

But is that a world record?
Does it claim world record?
Has it been recorded as the world record?

Cinebench - R23 Multi Core with BenchMate overclocking records @ HWBOT

Overclocking records

hwbot.org

Does it satisfy the definition of World Record?
Your claiming "oh data shows..."

Data shows i'll never win the lottery either... but will i stop trying?

Can we just say OK... they don't have the world record, and move on.
This is honestly all i am asking for.

Your insulting the many hours these guys spend trying to get that crown.
I really do not appreciate that, especially when i used to be on that bench table.
This is why i am holding my ground firm on this.

Soo. I posted benchmarks and links to benchmarks that show that at STOCK it wins the world record and using LN2 there is no statististics. Thats not a fair comparison and you know it. And in the real world, it will never use LN2, its just a show that means nothing to 99% of everybody, but you defend it since you think Intel is king. THEY ARE NOT in the server world, get over it !

 

[Abwx]

But is that a world record?
Does it claim world record?
Has it been recorded as the world record?

Cinebench - R23 Multi Core with BenchMate overclocking records @ HWBOT

Overclocking records

hwbot.org

Does it satisfy the definition of World Record?
Your claiming "oh data shows..."

Data shows i'll never win the lottery either... but will i stop trying?

Can we just say OK... they don't have the world record, and move on.
This is honestly all i am asking for.

Your insulting the many hours these guys spend trying to get that crown.
I really do not appreciate that, especially when i used to be on that bench table.
This is why i am holding my ground firm on this.

You know perfectly well that 2 x 96C Epyc should do around 160 000 pts at 720W CPU power, FTR a single 8490H use 530W on Cinebench, actually we dont even need a 2 x 96C screenshot, Mark provided the basis to iterate the expected score.

If you cant reach this basic conclusion then you have nothing to do in a technical forum, find a hobby that suit you better than CPU tech, sorry, but it cant be said otherwise when facing such a denial.

 

[aigomorla]

You know perfectly well that 2 x 96C Epyc should do around 160 000 pts at 720W CPU power

really.. then do it and post it on HWBOT.
Im really sure if it was possible and could be done, it would of already been posted.

And to my knowledge the 9474F is a 96core system.

Topbcy`s Cinebench - R23 Multi Core with BenchMate score: 123593 cb with a Epyc 9474F

The 2x Epyc 9474F @ 4000.9MHzscores getScoreFormatted in the Cinebench - R23 Multi Core with BenchMate benchmark. Topbcyranks #9 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

So please show me this 160,000 point you are so claiming.
Show me this World Record, and you wont hear a single comment more about it from me.

sorry, but it cant be said otherwise when facing such a denial.

Im the only one whose showing real proof on the definition of World Record here.
Has anyone of you actually pulled a real world record outside what i linked?

Please show me a REAL world record

Soo. I posted benchmarks and links to benchmarks that show that at STOCK it wins the world record and using LN2 there is no statististics. Thats not a fair comparison and you know it

Mark im asking for PROOF.
You guys are saying oh its possible.

Again, its possible my Smart Watch will beat both AMD and Intel at benchmarking my heart rate and step count.
But is that a world record?

Why are you guys so adamant at defending something which you guys can't even provide valid proofs for.
Show me Proof.

Because again you guys are spitting on my ex-hobby calling it fake and not real world, and again, i really am not down with that.

If you cant reach this basic conclusion then you have nothing to do in a technical forum, find a hobby that suit you better than CPU tech.

OH really? this is your reply to someone when you can't provide proof to back up your own statement?
This is getting really funny the more deeper i go down the rabbit hole.

Someone please pass me the pop corn.

 

[gdansk]

This is the Zen 5 thread...

 

[Abwx]

really.. then do it and post it on HWBOT.
Im really sure if it was possible and could be done, it would of already been posted.

And to my knowledge the 9474F is a 96core system.

Topbcy`s Cinebench - R23 Multi Core with BenchMate score: 123593 cb with a Epyc 9474F

The 2x Epyc 9474F @ 4000.9MHzscores getScoreFormatted in the Cinebench - R23 Multi Core with BenchMate benchmark. Topbcyranks #9 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

View attachment 82943

So please show me this 160,000 point you are so claiming.
Show me this World Record, and you wont hear a single comment more about it from me.



Im the only one whose showing real proof on the definition of World Record here.
Has anyone of you actually pulled a real world record outside what i linked?

Please show me a REAL world record.



Mark im asking for PROOF.
You guys are saying oh its possible.

Again, its possible my Smart Watch will beat both AMD and Intel at benchmarking my heart rate and step count.
But is that a world record?

Why are you guys so adamant at defending something which you guys can't even provide valid proofs for.
Show me Proof.

Because again you guys are spitting on my ex-hobby calling it fake and not real world, and again, i really am not down with that.

LN2 results are totaly irrelevant, this is just fancyness, yourself you never used a CPU under LN2 for regular usage, even when overclocked.

You are just a childisch character that got pissed because his beloved brand has been brought to irrelevancy and that your Intel preference, wich sweat in many of your posts, will force you to replace your current hardware with a junky CPU in respect of what is available from the competition, period.

 

[bakyt115]

You are moving the goal posts here, we know what is the IPC improvements as well as frequency, i showed you that perfs expectations from the craziest speculator, MLID, were about fullfilled and that those from RGT were very accurate, i posted numbers of their expectations and the final numbers.

the problem with them is that they don't say it's their EXPECTATIONs or SPECULATIONS. they literally saying that they have "sources".
this is misleading and unfair.

this is RTGS claim about phoenix.

 

[Abwx]

the problem with them is that they don't say it's their EXPECTATIONs or SPECULATIONS. they literally saying that they have "sources".
this is misleading and unfair.

this is RTGS claim about phoenix.

And what about if AMD actually tested some hybrid solutions, either to check feasability or for future implementations.?.

This project could well have been tested with a regular Phoenix die with disabled iGP, a lowish RDNA discrete GPU like a 6600/7600 as external GPU and some faulty V-cache that has only 32MB being functionnal, that RGT got some info about it without knowing all holding and borderings and did bold speculations about it is nothing to be suprised of.

 

[bakyt115]

And what about if AMD actually tested some hybrid solutions,

what if they are not?

 

[esquared]

Instead of handing out infractions for bad behavior and name calling I am locking this thread for a day or two.
To give everyone a chance to cool off.

Also, this is an AMD titled thread. Discussion is limited to:

Zen 5 Discussion (EPYC Turin and Strix Point/Granite Ridge - Ryzen 8000)​

Any other discussion besides Zen 5 needs to find an appropriately titled thread. Or if you folks are so inclined,
start your own thread.
If you're not discussing Zen5, and you know who you are, stay out of the thread.


esquared
Anandtech Forum Director

 

[esquared]

The thread is back open.

Be forewarned that any further trolling in this thread will result in infractions and vacations.


esquared
Anandtech Forum Director

 

[yuri69]

I think the point is that people were generally disappointed when Zen 4 was first given some performance estimates by Lisa Su. The IPC was given as what 8%? Why would that disappoint people so much? Because it is below industry trends, sure, but they compensated for that with clock rate. People had expectations of more IPC from rumors and in some cases expecting there were would be more cores.

Not saying Zen 4 is a dud just that expectations were once again higher than reality.

The Zen 4 disappointment has multiple sources.

The industry trends and AMD's trend from Excavator to Zen 3 are the main factor. Many remember charts as: IPC gains above competition's

This expectation was fuelled by leakerz who produced those completely fabricated 20+% IPC claims.

Designing the core for uber speed instead of deeply reworking the core led to a diabalance - the power consumption/heat suddenly went up. No more "zen" feeling.

Even more bottlenecked by IF, odd/useless L2 increase, miniscule tweaks apart from AVX512 - the unusually long time-to-release promised more.

 

[adroc_thurston]

Designing the core for uber speed

They didn't, it's just physdes flexing.

instead of deeply reworking the core led

How many times do I have to tell everyone that AMD does a proper tock every other gen (they're odd-numbered) and every even is iterative.

 

[Kaffeekenan]

I guess the new Infinity Fabric will play a major role in Zen5's IPC-increase. Does anyone have any clue, what we can expect from the new fabric? I have not found any information online so far. Do we maybe have some patents or anything that may point to the possible enhancements?

 

[SteinFG]

I guess the new Infinity Fabric will play a major role in Zen5's IPC-increase. Does anyone have any clue, what we can expect from the new fabric? I have not found any information online so far. Do we maybe have some patents or anything that may point to the possible enhancements?

I think Core to Core comms inside a CCX is not fabric's domain. might be wrong though.
Fabric is for CCX to CCX to memory comms.
IO die stays the same most likely, so no new improvment to fabric expected.
Zen4 -> Zen5 will be like Zen2 -> Zen3:
Same class node, new CCX, big improvment because of that.

 

[Kepler_L2]

I guess the new Infinity Fabric will play a major role in Zen5's IPC-increase. Does anyone have any clue, what we can expect from the new fabric? I have not found any information online so far. Do we maybe have some patents or anything that may point to the possible enhancements?

The big Infinity Fabric update is Zen6 AFAIK.

 

[Abwx]

Some meager news about Zen 5 in WCCF Tech summarized article of news highjacked, as usual , at TPU :

What we are looking for with Zen 5 is to bring it into the desktop space as quickly as we possibly can. As excited as we are about Zen 4, I think what you've seen from AMD is that every step in that processor core architecture gives us just such an amazing uplift in terms of capability, in terms of IPC, in terms of performance across every possible workload that you see in desktop and mobile applications. We're working very, very hard to get it into the market as fast as possible, I know that Mark has talked in the past about how Zen 5 is on-track, it's in-design, it's taped-out etc so we are working very hard on it and I think you're going to be very excited when you see that product come to market.

AMD Says It Won't Follow Intel's P-Core & E-Core Hybrid Approach, Talks Zen 5, x86S & More

AMD has said they won't be following Intel's approach to Hybrid cores and architectures and also talked about what to expect from Zen 5 CPUs.

wccftech.com

And TPU s full inteview of AMD s Mc Afee encompassing several subjects :

TPU Interviews AMD Vice President: Ryzen AI, X3D, Zen 4 Future Strategy and More

In this exclusive interview, we talk to the AMD Vice President who's responsible for AMD processors, both on desktop and mobile. Of course we had to ask about AI, but we also learned more about AM5 APUs, core counts, chiplets, Hybrid Architectures, Zen 5 and Intel's x86s proposal.

www.techpowerup.com

 

[Kaffeekenan]

really.. then do it and post it on HWBOT.
Im really sure if it was possible and could be done, it would of already been posted.

And to my knowledge the 9474F is a 96core system.

Topbcy`s Cinebench - R23 Multi Core with BenchMate score: 123593 cb with a Epyc 9474F

The 2x Epyc 9474F @ 4000.9MHzscores getScoreFormatted in the Cinebench - R23 Multi Core with BenchMate benchmark. Topbcyranks #9 worldwide and #1 in the hardware class. Find out more at HWBOT.

hwbot.org

View attachment 82943

So please show me this 160,000 point you are so claiming.
Show me this World Record, and you wont hear a single comment more about it from me.



Im the only one whose showing real proof on the definition of World Record here.
Has anyone of you actually pulled a real world record outside what i linked?

Please show me a REAL world record




Mark im asking for PROOF.
You guys are saying oh its possible.

Again, its possible my Smart Watch will beat both AMD and Intel at benchmarking my heart rate and step count.
But is that a world record?

Why are you guys so adamant at defending something which you guys can't even provide valid proofs for.
Show me Proof.

Because again you guys are spitting on my ex-hobby calling it fake and not real world, and again, i really am not down with that.



OH really? this is your reply to someone when you can't provide proof to back up your own statement?
This is getting really funny the more deeper i go down the rabbit hole.

Someone please pass me the pop corn.

You DO realize that the 123xxx points is for a 96 core system, right? So yes Genoa DOES kill SPR in Cinebench.

 

[DisEnchantment]

Looks like one of the Strix variants (1151) have full VGPRs like N31.