This gen on gen is very convenient for AMD comparisons since it’s every 2 years unlike for Intel, Apple and ARM which is every year.That is a pretty significant Gen-on-Gen increase!
Higher than Apple has done in 4 years.
Can we like wait for tests. I mean in tests without AVX-512 it won’t be that impressive. We already know CBR23 and CB 2024 scores for Strix Point. Nothing groundbreaking.Also, that 16% was an average of apps and games listed. There are some apps and games that have seen obscene 40-60% uplifts, and NOT due to AVX-512.
I did some quick math and got the following:
View attachment 101191
For myself as somebody who so far uses Zen chips only in the form of server chips (in compute nodes), Zen 3 did not stand out at all. The one thing good about it from my perspective is that it laid the groundwork for Zen 4.it looks like after the OG Zen, only Zen 3 stands out.
Did you miss the part where I said the YoY gains over the last 4 years. If we take that metric and we should if we are comparing with others, AMD comes of as average.But I just read that Zen 5 had the lowest IPC increase other than Zen+! Perhaps I misread it because it actually has the second highest other than Zen 3.
This just proves AMD YoY IPC gains are mediocre after Zen 3. I’m not referring to total performance uplifts but IPC.I did some quick math and got the following:
View attachment 101191
Actually AMD told us that 16c Zen 5 @ 4GHz are on average 1.16 as fast as 8c Zen 4 @ 4GHz.But I just read that Zen 5 had the lowest IPC increase other than Zen+!
They said in the press briefing pre Computex that there are clock increases in many workloads.What I said still stands and Zen 5 is worse overall because of no clocks increase too.
What's Intel's, Apple's, and ARM's —for instance— 9654 competitor then? (Edit 2: I wanted to say, increasing mad performance by 8%/y is not the same as increasing middling performance by 8%/y. Similar to the argument by others in this thread when Apple's deceleration after their big M1 jump was brought up.)All I am saying is AMD is not doing any better than Intel, Apple and ARM. No need to give AMD a special pat on the back.
I’m referring to maximum clocks. All core clocks over Zen 4 should be greatly improved tho.They said in the press briefing pre Computex that there are clock increases in many workloads.
This just proves AMD YoY IPC gains are mediocre after Zen 3. I’m not referring to total performance uplifts but IPC.
What I said still stands and Zen 5 is worse overall because of no clocks increase too.
All I am saying is AMD is not doing any better than Intel, Apple and ARM. No need to give AMD a special pat on the back.
I was perplexed by that when I read the endnotes. Shouldn’t the result be the same if they compared 7950X v 9950X instead of 7700X v 9950X?Actually AMD told us that 16c Zen 5 @ 4GHz are on average 1.16 as fast as 8c Zen 4 @ 4GHz.
Not quite. They alluded to having higher boost clocks. Base clocks doesn't really have any bearing on what we expect to do with these.There were comments made in an interview with two AMD reps that alluded to this.
yep thats right. Thats both impressive and sad lol. Impressive that Apple still holds the IPC lead after the slowdowns and sad that AMD still cannot beat Apple despite executing better than Apple.Compared to Apple at least, AMD's CAGR in IPC is around twice that of Apple's (I get 5.5 CAGR IPC for Apple and 11.8 for AMD).
Edit: Starting in 2020 with Zen 3 and A14.
yep thats right. Thats both impressive and sad lol. Impressive that Apple still holds the IPC lead after the slowdowns and sad that AMD still cannot beat Apple despite executing better than Apple.
Still to early to tell, if Zen5 can get closer to max clocks for longer in nT it will be a win.I’m referring to maximum clocks. All core clocks over Zen 4 should be greatly improved tho.
I would be interested in a chart showing IPC vs transistors used per core. Found it hard to get a good source of the data though... maybe my google-fu sucked on that day.I did some quick math and got the following:
View attachment 101191
Compared to Apple at least, AMD's CAGR in IPC is around twice that of Apple's (I get5.56.23 CAGR IPC for Apple and 11.8 for AMD).
Edit: Starting in 2020 with Zen 3 and A14.
I’m sure someone has the xtor count compiled, but I’m not convinced it will be worth much vs IPC because not every transistor is made equal, e.g. a lot of transistors get used for AVX2 and AVX512 but those don’t translate to IPC particularly well. A core which focuses on general purpose compute and not on HPC or server, like Skymont, will naturally have higher “IPC/xtor” so the metric doesn’t mean much unless your feature set is equal.I would be interested in a chart showing IPC vs transistors used per core. Found it hard to get a good source of the data though... maybe my google-fu sucked on that day.
It wasn't for a comparison vs other companies, just to see if if there was a growth in the needed transistors for +1% IPC. ie is the curve roughly a straight line still, or is the transistor needs slowly growing to keep the IPC needs going? Perhaps percentage units wouldn't be optimal either, but rather an absolute performance unit would be better.I’m sure someone has the xtor count compiled, but I’m not convinced it will be worth much vs IPC because not every transistor is made equal, e.g. a lot of transistors get used for AVX2 and AVX512 but those don’t translate to IPC particularly well. A core which focuses on general purpose compute and not on HPC or server, like Skymont, will naturally have higher “IPC/xtor” so the metric doesn’t mean much unless your feature set is equal.
Fwiw, a bunch of xtors get used to increase max clock speeds as well, and those don’t translate to IPC directly. If there’s one thing to compare, it might be total ST performance vs xtor count, but I suspect it will just roughly follow the Square Root Law, i.e. 2x xtors for 1.4x performance.It wasn't for a comparison vs other companies, just to see if if there was a growth in the needed transistors for +1% IPC. ie is the curve roughly a straight line still, or is the transistor needs slowly growing to keep the IPC needs going? Perhaps percentage units wouldn't be optimal either, but rather an absolute performance unit would be better.
I was going to go back decades... maybe too hard to measure such a marketing thing as IPC across such a long time. A single benchmark might be better, but again there would be issues there too over the long haul.
If only eekBench was available back 20 years ago...
I don't even know if V Cache with higher clocks / Voltages is a good thing. It's biggest strength, the low power consumption, is literally because of that limitations. So maybe ZEN5 X3D will be another like 5% faster but lose it's power consumption advantage? Doesn't sound like a good tradeoff for me. Also Both CCDs with V Cache is unnecessary.Even MLID warns possible sandbagging especially V-cache from AMD in his latest video: