- Mar 3, 2017
- 1,684
- 6,227
- 136
That’d be a big enough perf/watt increase to show in a slide.Atta. That's a 105W part (aka 142W).
Versus 88W for GNR.
Well it looks that, yea.That’d be a big enough perf/watt increase to show in a slide.
In case you've missed the keynote, stuff not AI is not relevant.They wouldn’t just forget to show off a >=10% perf/watt increase.
It floats above in AVX-512 too.As was already mentioned, this may be all clock AVX-512 full speed (as opposed to 1 half of AVX-512 at full clock speed in Zen 4)
I’m guessing it’s <10% perf/watt over Zen 4. I don’t think you skip an important detail like that. It was a 90 minute key note filled with partner wank bloat.. they would’ve squeezed it in.Well it looks that, yea.
In case you've missed the keynote, stuff not AI is not relevant.
They did away with GNR in like the first 5 minutes and there you go.
Eh.I’m guessing it’s <10% perf/watt over Zen 4.
It's not important. But AI™ is.I don’t think you skip an important detail like that.
Z4>Z5 is a bigger leap than RPC>RWC+. Turin platform architecture should remain a big differentiator and the gains from what was shown extrapolated to Genoa results is showing another big leap.IMO, the opposite. Client has a better time dealing with the lower than expected IPC gains than server because DIY can cope with 3D V-cache. For servers, Granite Rapids looks much more competitive with Turin now. Both are 128C, Turin should have the per-core perf advantage, but if RWC can manage to clock higher (given a certain amount of power), or at least close to Zen 5....
yeah CWF is the first notable Intel DC product.Z4>Z5 is a bigger leap than RPC>RWC+. Turin platform architecture should remain a big differentiator and the gains from what was shown extrapolated to Genoa results is showing another big leap.
Intel closes the gap a bit, but not enough to change the situation.
Wait for CWF.
IMO, the opposite. Client has a better time dealing with the lower than expected IPC gains than server because DIY can cope with 3D V-cache. For servers, Granite Rapids looks much more competitive with Turin now. Both are 128C, Turin should have the per-core perf advantage, but if RWC can manage to clock higher (given a certain amount of power), or at least close to Zen 5....
Intel wants to keep CSPs loyal since even Meta defected.When it comes to hype, Intel is hyping Sierra Forrest more than Granite Rapids.
It floats above in AVX-512 too.
We will see how the SpecInt Rate turn out for servers.
There are 2 ways to get there. Either higher power, same or lower clock speed and much higher IPC.
Or, if the power consumption is lower, server chip can run at higher all core clock speed and would need lower IPC to get there.
Nah, it's over. If AMD was able to get to the over 30% integer IPC mark, it would have been made known very loudly in their presentation today.
I never believed the crazy high IPC/perf predictions but, like pretty much everyone else, the 16% numbers (still needs 3rd party review) is still a disappointment. Not a bad number, but definitely disappointing given the changes to the core and what they were able to get previously with Zen 3. Based on those things, I was expecting at least 20% but here we are and another hype train has crashed. Looks like my 5900x will get some more life as I'll wait now for at least the 3d cache versions before upgrading, if not Zen 6. Had the hype train been real, I probably would have jumped on the initial release.
On the plus side, it does look like availability will be soon for STX and soon after release. Hopefully this actually comes true in the end consumer market. While I don't care about the NPU, having the top performing NPU for now probably led AMD to get a good chunk of their design wins with all the AI fervor going around. I am also happy to see the lowered TDPs on desktop. It probably will make for an awkward gen on gen performance comparison for each tier that had a significant TDP drop, but I very much disliked how AMD seemed to be following Intel's lead into insance power consumption to eek out a tiny bit higher score on benchmarks. I for one welcome this reset on TDP levels. I just wish they did the same for the top CPU as well, though I understand why they didn't. Server will be fine and I expect they will continue to win marketshare there. Consumer side is quite a battlefield though. I think both Intel and AMD need a A64/Conroe type moment to reset their core designs and stop wasting power and area gunning after crazy high clocks but I don't know if that's really possible for them anytime soon. I was hoping Zen 5 was going to be AMD's first step in that direction, but it didn't turn out that way.
Ahh well, at least we should have some calm before the next hype storm starts brewing. I also want to give a shout out to @Exist50 who seems to have called quite a lot of things with good accuracy on both sides and did so without being abrasive or seeking to demean other posters. Hopefully we can get back to that type of discussion which seemed to be what we (mostly) had for a while until fairly recently.
The IPC is fine they just needed 5% clock rate with it too.
And for the first time Zen clocks are stagnant despite being on a slightly better process.
The claim was SpecInt Rate score not IPC (correct me if I am wrong). IPC was implied.
No it’s not. ARM did 16% YoY vs AMD’s 8% YoY.Even a 15% IPC increase every two years is a lot better than many of their competitors.
That's just one thing.AMD is actually on the low end.
This slide was always considered legit, no surprises there (MLID or not, when they show you stuff like that, that is much more reliable than just those usual quaotes from "sources"). Back then I personally though it would be underguessing more and the final IPC would be +18 %...View attachment 100290
This slide was spot on, seems Zen 6 with 10% IPC would be what I would bet on.
MLID the winning leaker.
AMD took almost two years for a paltry 16% while Apple and ARM iterating on almost yearly cadence.
Nah, there's Sorano.Even in DC it is all Instinct next year.
No it’s not. ARM did 16% YoY vs AMD’s 8% YoY.
Apple did 7% IPC M3 -> M4 in 7 months. That’s also a company that’s lost a lot of CPU talent recently.
AMD is actually on the low end.
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...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"
The stinky Cortex-X thing is power.The latest ARM big core seems to be < 10% though that's still TBD with proper testing.