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

[DisEnchantment]

Speculate at will

 

[poke01]

AMD has server locked.

 

[gdansk]

bidding starts at +25% IPC

My betting strategy is 1 month between release = 1% more 1T performance.

 

[Hulk]

16% average IPC uplift over Zen 4 for Zen 5. That's a winner in my book.
+17% in CB R23 means it should do about 45,000 points.
5.7GHz boost and 170W
AMD has done it again!

 

[DrMrLordX]

What do you mean? No hope in Arrow Lake?

Probably not. If Intel's own leaked slides are any indicator, they're only looking at +5-6% ST and +20% MT total performance gain over Raptor Lake 8+16 at the same PL2. And the way people run their Raptor Lakes, real-world performance gain may not be much better, especially if performance scaling at high power values isn't as good on TSMC N3b. At least Arrow Lake will probably do better at lower power values.

Have to say I'm a little disappointed with the Zen5 numbers, but eh whatever I wasn't going to buy this generation anyway so what do I care?

 

[IronLynx]

Dude, you got duped. Get over it.

If ADROC's claims about server SIR disclosure are correct, there will be some hell to pay. Big players don't like underdelivering. Something is hanging the expected performance/is neutered. Maybe we will only know at the Zen6 launch.IF, for some reason, Zen6 gets a bigger IPC than Zen5, we know something is wrong with Zen5.

 

[poke01]

16% average IPC uplift over Zen 4 for Zen 5. That's a winner in my book.

AMD pulled a sneak tactic and used AES test GB5 in the geomean instead of GB5 1t.

 

[Markfw]

I think 16% is underwhelming, and deff worse than Zen 4, but I also don't think it's atrocious.
Also, do you guys notice that the perf uplift in gaming seems to be... weird? Their comparison vs the 14900k (at least from that videocardz leak) is like 13% better on average, which would mean that it's 25-30% better than Zen 4.

That would be due to the new core IMO. And that makes it better than the 7950x3d in games, so no waiting on a better game CPU

 

[DrMrLordX]

If ADROC's claims about server SIR disclosure are correct, there will be some hell to pay.

Either that, or the SPEC numbers are legit and don't map well to client workloads.

 

[adroc_thurston]

Zen6 gets a bigger IPC than Zen5, we know something is wrong with Zen5.

Zen6 is a boring nothingburger iteration. Stuff not core is what changes there.

 

[RnR_au]

IF, for some reason, Zen6 gets a bigger IPC than Zen5, we know something is wrong with Zen5.

You seem to be suggesting that the RDNA3 engineers had a hand in Zen 5....

 

[gdansk]

Zen6 is a boring nothingburger iteration. Stuff not core is what changes there.

I hope you're wrong 3 times in a row.

 

[Ghostsonplanets]

Didn't PHX' iGPU clock up to 2.7 GHz? 2.9 GHz isn't a groundbreaking change...

Wider GPU though (8 WGP/1024 ALUs x 6 WGP/ 768 ALUs) and they also increased core count. So power budget is more tight to manage.

I expect Kraken or Kraken chop (242) to clock the iGPU higher, like how Ryzen 7540/5U | 8540U clock at 2.8GHz.

 

[adroc_thurston]

I hope you're wrong 3 times in a row.

Hopefully.
This was supposed to be a whole lot more fun than Z4 yet it isn't.
Sad!

 

[CouncilorIrissa]

RDNA3.5 is curiously not clocking that much higher compared to supposedly broken RDNA3, although I guess there's not much point to clock the 8WGP SKUs that high anyway.

 

[poke01]

Don't we get an IPC breakdown like Zen 4 computex?

 

[gdansk]

RDNA3.5 is curiously not clocking that much higher compared to supposedly broken RDNA3, although I guess there's not much point to clock the 8WGP SKUs that high anyway.

I think an 11 TFLOPs iGPU is "enough" and will be bottlenecked by 128 bit memory bus.

 

[adroc_thurston]

RDNA3.5 is curiously not clocking that much higher compared to supposedly broken RDNA3, although I guess there's not much point to clock the 8WGP SKUs that high anyway.

Sustained clocks.

 

[Geddagod]

That would be due to the new core IMO. And that makes it better than the 7950x3d in games, so no waiting on a better game CPU

Im guessing AMD is just cherry picking games or gimping the memory...

Don't we get an IPC breakdown like Zen 4 computex?

Maybe at hotchips?

 

[Geddagod]

Also, lol


edit: read the last sentence

 

[poke01]

Also, lol
View attachment 100279

edit: read the last sentence

where is this?

 

[CouncilorIrissa]

Also, lol
View attachment 100279

edit: read the last sentence

Still the same single intern working at AMD website.

 

[Markfw]

AMD has server locked.

It had that in Zen 4 / Genoa. Now its just embarrassing for Intel.

 

[PJVol]

Why stretch an outdated platform to 2027 and beyond? Socket compatibility is irrelevant to the overwhelming majority of users.

That's good news for me.

 

[RnR_au]

When are Zen 5 reviews expected to be out?

 

[Hulk]

AMD pulled a sneak tactic and used AES test GB5 in the geomean instead of GB5 1t.

AMD just continues to produce solid across-the-board gains with great efficiency.

While I have a 14900K (underclocked), the reality is that I suspect an efficient cool running 9950X will put up 45,000 CB R23 points. A 14900KS has to be volted and clocked to within an inch of it's quickly degrading life to approach such a score and needs a custom loop and chiller as well. Yes I know Zen 5 will compete against ARL but Zen 5 will stand alone for quite some time I think even if Intel manages to push ARL out the door in quantity by the end of the year.

In addition, the enormous rumored IPC gains of Skymont are going to have to be a reality for ARL to compete with Zen 5 considering the former's lack of HT.

Looks like AMD just "called" on Intel. We'll see if Intel is bluffing or not soon enough.

+5% IPC for Zen 5 would have been a stumble. +10% would have been "meh." +16% is good in my book. 20% or more would have been great.