- Mar 3, 2017
- 1,684
- 6,227
- 136
It was 32% SPECint 1t IPC increase (with the caveat that this was actually a low ball figure because the real number was too big to believe) for one poster and 40+% total SPECint rate performance when comparing core for core. Neither is happening.
The stinky Cortex-X thing is power.
X4 was bad, X5 seems worse.
X3 was the last core where they kept power even.
5Ghz+ Firestorm has to wait another year.I think they mean Zen 5 still hasn't reached Firestorm IPC. And the 9950X will probably be behind a 10W iPad in day-to-day tasks because it didn't achieve that nor did it push clock rates higher. Although I guess it is a fair bit cheaper...
Relative gen on gen power matters.Compared to Apple or prior ARM, sure
Intel yea. AMD not really.Against Intel/AMD, not so much
Server eh, at least AWS strategy since forever is to hug the Vmin to eke efficiency win at area spend.but when competing with Intel/AMD in laptop/server/desktop, they seem to be more than good.
IMO, this is definitely the first miss of the Zen era. I personally don’t buy that the IPC increase is even 16%. If you were to calculate average IPC similar to how they did with Zen 4 (they used a larger group of tests) it would probably end up at less than the 16% figure they gave.
I suppose we’ll get more insight at Hot Chips.
If they wanted to AVX-512 pump, they'd do like y-cruncher.The inclusion of the GB AES test is peculiar for sure but AMD previously has had some peculiar examples in their IPC chart and it still came out pretty much in line with SPEC
They do the number backwards anyway.IMO, this is definitely the first miss of the Zen era. I personally don’t buy that the IPC increase is even 16%. If you were to calculate average IPC similar to how they did with Zen 4 (they used a larger group of tests) it would probably end up at less than the 16% figure they gave.
I suppose we’ll get more insight at Hot Chips.
M4 is also a new core that actually had IPC improvements. Plus Apple stalled. Lost talent.As for Apple... they had only one new core (M3) per four generations of SoCs, since M1. And that core doesn't have much better IPC actually, either. Rest is ramping up clocks (wchih ironically all the fanboys bashed hard as the WRONG and stupid thing to do, in 2020 - I find that very amusing).
AMD still hasn’t caught up to M1 IPC. Yes, AMD is making progress but not that some miracle rate. YoY, AMDs IPC both Zen 4/5 is nothing special.AMD put out Zen 3 at about the same time as M1 and came up with Z4 and Z5 since, both with +10-15% IPC so that's more progress if you believe "gigahurtz dumb"
Relative gen on gen power matters.
X4 was a big stinky (but I bought S24u anyway).
Also ARM itself pushes the suggested clk too.
X5 one is what, 3.8 fmax? Gonna burn.
Intel yea. AMD not really.
Server eh, at least AWS strategy since forever is to hug the Vmin to eke efficiency win at area spend.
It was also the only Zen to be on time so that was nice.Zen 3 was amazing.🤩
Remember, all it takes to hit AMD 1t power for Cortex-X5 is probably like 200-300MHz on top of what ARM suggests.I'm considering ARM's push to encroach on AMD/Intel's turf so comparison to them is what matters to me.
It's a question of clocking stuff down.I still haven't seen anything to suggest that AMD is competitive in terms of 1t efficiency with ARM.
Remember, all it takes to hit AMD 1t power for Cortex-X5 is probably like 200-300MHz on top of what ARM suggests.
It's all relative basically.
Granted, I don't even know if they'll be able to yield them clocks or make it work reliably.
It's a question of clocking stuff down.
Efficiency is relative unless you pick a specific 1t point.
Yeah but that's the thing, no vendor ships ARM recommended.Shipping clocks is the comparison
Also, the last to be fully designed in-office, pre-Covid, with minimum or no remote work.It was also the only Zen to be on time so that was nice.
Yeah but that's the thing, no vendor ships ARM recommended.
They've been shilling like 3.5 since Cortex-X2 or therein, zero stuff ships like that.
But then AMD can just drop 200MHz, and dang, more efficient!That's fine, what ships is the comparison.
🤔 But Zen 1 and Zen 2 were delayed a few months too.Also, the last to be fully designed in-office, pre-Covid, with minimum or no remote work.
Zen1 was basically beta silicon to begin with.🤔 But Zen 1 and Zen 2 were delayed a few months too.
But then AMD can just drop 200MHz, and dang, more efficient!
All relative.
Gotta wait until someone ships C-X5 in laptops so see the lapwarming chops.
Again, it's all kinda relative.I don't think 4.9 GHz will get them to match efficiency but if there are numbers to prove it, I'll gladly be wrong. I don't think they'll be able to match ARM's efficiency this gen without relinquishing their performance leadership claim, so I don't care about discussing hypotheticals of what AMD could do just to match on efficiency.
Again, it's all kinda relative.
ARM itself can beat apple efficiency by just clocking lower (doesn't apply to LITTLEs though).
Yeah but it's all a game where everyone games each other, including the IP provider.And Apple could also clock lower and still be more efficient, but again, I'm not interested in hypotheticals.
I think you're over valuing the ARM and Apple IPC improvements. M4 was like 5% improvement and has been <= 5% for multiple generations now. The latest ARM big core seems to be < 10% though that's still TBD with proper testing. Apple (and ARM to a degree) have mostly been increasing performance through clock gains. Apple is still IPC king because of the lead they had starting several years ago, but IPC increases have not been good as of late.
Yeah but it's all a game where everyone games each other, including the IP provider.
Not really - Apple gained IPC and ALSO gained in frequency. Yes they are operating at a lower frequency so gaining there is easier, but a modest IPC gain combined with a decent frequency gain delivered a very nice performance gain. And while we love us some nice juicy IPC gains here, at the end of the day the total performance gain consisting of both IPC and frequency is what it is all about.
There's a constant tension because both have diminishing returns. Apple seems to be well into the diminishing returns section for IPC, at least barring some major advance, but have headroom for frequency. AMD has headroom for IPC, but appear to be hitting the wall for diminishing returns on frequency gains.