AMD Zen 2 (Ryzen 3000) - *** NO COMPETITOR HINTING ***

[Stephanie Peterson]

Intel and Amds current IPC is not much different, depending on work load of course.
I expect 4.7/4.8 on the higher binned single core (or limited core) boost and maybe 4.4 all core on the bigger parts - which would be going some out the gate.

 

[~>Dg<~]

i think it was on avg 15 percent behind and maybe why amd have aimed to get that close with these new cpus. as that would put them level or infront of intel chips. if they are level then its down to mhz which is where i cant see amd matching intel. which is going to be very interesting. if they have it will probably be a nice new 12 core for one of my pcs.

 

[LePhuronn]

At the risk of starting this argument again, I thought AMD's IPC in Ryzen 2000 series was similar to Intel's as the ~15% deficit was due to ~15% lower clock speed.

But it's held back by lower RAM speeds affecting the Infinity Fabric and internal tomfoolery not feeding data fast enough. I think it's been mentioned elsewhere that if Zen+ had a better "back end" then it would've destroyed Coffee Lake.

 

[Troezar]

But it's held back by lower RAM speeds affecting the Infinity Fabric and internal tomfoolery not feeding data fast enough. I think it's been mentioned elsewhere that if Zen+ had a better "back end" then it would've destroyed Coffee Lake.

That is very interesting. Gives me more confidence that Ryzen 3000 will deliver.

As an aside I fired up my old Athlon 4200+ system at the weekend after 8 years of being turned off. Surprisingly snappy and the MSI motherboard I modified is still soldiering on. 2GB of RAM and 512GB vRAM on the X1950 pro seems terrible now!

 

[chrcoluk]

I would say on desktop and especially on unpatched machines intel is still ahead on desktop but the advantage is below 10%.

For servers which are patched for meltdown, spectre etc. epyc is already ahead on IPC. As intel got absolutely destroyed by that stuff, especially on virtualization.

Whilst I expect improvements I think 4.5 base clock is one hell of a jump and perhaps unrealistic expectations.

Also wasnt the base models supposed to be going t 6 cores according to posts on here? if 6 cores is the base model then why is there 4 core samples been sent out?

 

[CuriousTomCat]

Also wasnt the base models supposed to be going t 6 cores according to posts on here? if 6 cores is the base model then why is there 4 core samples been sent out?

Maybe they're trying to keep the final specs a secret from the trators who leak information. Or maybe each sample that went out had a slightly different spec so they know which source the leak came from.

 

[DragonQ]

i think it was on avg 15 percent behind and maybe why amd have aimed to get that close with these new cpus. as that would put them level or infront of intel chips. if they are level then its down to mhz which is where i cant see amd matching intel. which is going to be very interesting. if they have it will probably be a nice new 12 core for one of my pcs.

No, Zen+'s IPC is nowhere near 15% behind Coffee Lake's on average. Where are you reading that??

 

[d_brennen]

No, Zen+'s IPC is nowhere near 15% behind Coffee Lake's on average. Where are you reading that??

He's talking gaming (as usual) where the 2600/2700 are 5-20% behind Intel. With the clock deficit removed its going to be <10%, probably lower on average too. With clock parity and improvements in Ryzen 3xxx, I think Intel's gaming "dominance" (of a few percent here and there for x amount more money...) is in jeopardy.

 

[~>Dg<~]

yeah its either amd in front or very close..then its clockspeed. then price . also heat.

 

[DragonQ]

He's talking gaming (as usual) where the 2600/2700 are 5-20% behind Intel. With the clock deficit removed its going to be <10%, probably lower on average too. With clock parity and improvements in Ryzen 3xxx, I think Intel's gaming "dominance" (of a few percent here and there for x amount more money...) is in jeopardy.

Yes in gaming only you could argue maybe 10-12% on average, as always it depends what games you test though. Newer ones tend to favour Intel less. In most other common consumer applications it's +/-5% so effectively even. This article is rather nice.

 

[Murphy]

At the risk of starting this argument again, I thought AMD's IPC in Ryzen 2000 series was similar to Intel's as the ~15% deficit was due to ~15% lower clock speed.

Depends what the instruction per cycle you're basing your measurement off of, if we're talking about the traditional FP then yea, even if we're talking about all the other types of instructions it would probably still be a yes but it's not so much how many instruction per cycle, at least not when talking about two designs that are fundamentally quiet similar, it's more about what happens to the data going in and out of an individual core, how the caches are arranged, how cores communicate with each other, caches, and the outside world.

But it's held back by lower RAM speeds affecting the Infinity Fabric and internal tomfoolery not feeding data fast enough. I think it's been mentioned elsewhere that if Zen+ had a better "back end" then it would've destroyed Coffee Lake.

It's not so much the bandwidth, it's the latency, it was 'held back' because the clock speed of the data fabric part of IF was tied to memory controller so faster RAM essentially overclocked the data fabric and reduced its latency.

 

[Hagar]

A little over a month to go and we will know how the 3000 Ryzen copes, the leaks on PCIe4 are quite exciting and that could be a major differentiator between Intel and AMD in the short term < 2 years.

[Offtopic] I just watched 2001 A Space Odessy again on TV and was fascinated to see how the crew were watching tablet devices whilst eating lunch - it was written in 1968 - a lot of fun just watching Arthur C Clarkes predictions.
On a humorous note HAL the onboard computer which went bonkers was a 9000 series. Intel are there, AMD have a way to go (except the 9590 which was a bit bonkers.
[/Offtopic]

 

[LePhuronn]

HAL 9000...

High Altitude Lake?
Hydrogen Ablaze Lake?
Highly Agitated Lake?

 

[humbug]

But it's held back by lower RAM speeds affecting the Infinity Fabric and internal tomfoolery not feeding data fast enough. I think it's been mentioned elsewhere that if Zen+ had a better "back end" then it would've destroyed Coffee Lake.

Correct.

Zen: 2048 uOps
Coffeelake: 1536 uOps

Zen: 10 Integer Execution units
Coffeelake: 8 Integer Execution units

Zen: 4 FP units
Coffeelake: 2 FP units

Zen: Float latency 3 Cycles
Coffeelake: Float latency 4 Cycles

And so on....

Zen has 20% to 100% faster / more powerful cores but is bottlencked to about the same performance as Coffeelake by its front end, tweak and improve that front end and Zen will over take Coffeelake in IPC.

This is no doubt the aim of Zen 2.

 

[~>Dg<~]

4.5 ghz as i said for 6 months is my buy point.

 

[Murphy]

Zen: 2048 uOps
Coffeelake: 1536 uOps

Zen: 10 Integer Execution units
Coffeelake: 8 Integer Execution units

Zen: 4 FP units
Coffeelake: 2 FP units

Zen: Float latency 3 Cycles
Coffeelake: Float latency 4 Cycles

Curious where you got those numbers from as some of them don't seem to match with what I've read.

 

[Undesirable]

Can't wait to see the final benches. It'll be nice if I don't have to bother overclocking it at all or tweaking endless RAM timings.

 

[humbug]

Curious where you got those numbers from as some of them don't seem to match with what I've read.

Its in the video you cut out of the quote. and yes they are in that link, i can see them just skimming over it.

The full quote.

Correct.

Zen: 2048 uOps
Coffeelake: 1536 uOps

Zen: 10 Integer Execution units
Coffeelake: 8 Integer Execution units

Zen: 4 FP units
Coffeelake: 2 FP units

Zen: Float latency 3 Cycles
Coffeelake: Float latency 4 Cycles

And so on....

Zen has 20% to 100% faster / more powerful cores but is bottlencked to about the same performance as Coffeelake by its front end, tweak and improve that front end and Zen will over take Coffeelake in IPC.

This is no doubt the aim of Zen 2.

 

[Murphy]

Oh, OK. I've not listened to it.

If that's what it says in the video i think whoever's doing maybe getting some things wrong as Zen doesn't have 2048 uOps (not even sure what that's referring to), (I assume it's referring to this) it has a 2048 uOps L0 cache, it doesn't have 10 Integer Execution units and 4 FP units, it has 4 ALUs and 2 AGUs, and it doesn't have an FP latency of 3 cycles, it has a FP latency of 7-8 cycles.

 

[humbug]

Oh, OK. I've not listened to it.

If that's what it says in the video i think whoever's doing maybe getting some things wrong as Zen doesn't have 2048 uOps, it has a 2048 uOps L0 cache, it doesn't have 10 Integer Execution units and 4 FP units, it has 4 ALUs and 2 AGUs, and it doesn't have an FP latency of 3 cycles, it has a FP latency of 7-8 cycles.

Here we go again with you trying to twist something in to something that it isn't because it doesn't agree with you, how is it that you have so much time and energy to straw-man your way through this thread constantly, don't you have anything better to do or is this what you do?

His information came from an article written by Dr Fog https://www.researchgate.net/profile/Agner_Fog a well respected technology analyst. he's quoting it in the video with the data right in front of him.

If you don't like whats in his article go and disagree with him, stop with your constant arguing in this thread.