The only integer IPC improvement they state is 7.3%.
Not sure where you got the 4.6% integer IPC improvement from the video. Happy to be corrected.
Let’s move this to the Apple thread. I’ll have a look at this video and Geekerwan’s prior videos too. They clearly been using different datasets.He put out 2 videos. You can see this post for the IPC calculation from the first video.
Keplar can reply himself if he wants to, but I believe he already explained that he wasn't meaning that Apple would get lapped this year, but that the trend for AMD/Intel against Apple would mean that they would get lapped within the next few years and he believes they will struggle to catch back up. He very well could be wrong about that, but he is at least correct that Apple's improvements for the last few generations have lagged their competitors, some times quite significantly.
Releasing a one off is one thing. Delivering regular, sizable performance increases is another. Apple has not delivered. Qualcomm has also struggled in this area. AMD has not. (no comment on Intel lol)
L3 b/w was already class-leading, rest not so much.That's a significant upgrade in L1 and L2 cache bandwidth. L3, not so impressive.
It's kinda expected to support twice the FP throughput over Zen 4, but let's wait and see how much it benefits INT workloads.That's a significant upgrade in L1 and L2 cache bandwidth. L3, not so impressive.
Another benchmark for GNR. 5.1ghz clockspeed for 9600x.
L3 got that shiny new macro though.L3 b/w was already class-leading, rest not so much.
It is easier for Intel/AMD to get bigger IPC improvements because they're behind. Let's see how big their improvements are when they reach M3's IPC. Making things wider has a smaller and smaller impact the wider you are.
Further there's the law of large numbers working against this. Let's say you stop making any increases in frequency, so IPC was all there was. IPC improvements would then easily be calculated by percentage increase in performance. If your GB6 score is 2000, and you gain 200 points, that's a 10% improvement. If your GB6 score is 4000 and you gain 200 points, that's only a 5% improvement.
I don't think the law of large numbers has anything to do with it as we always count improvements in terms of percentages and no one counts by raw score increases.
It’s not like, hey, we’ve also added X3D to a chip. We are working actively on really cool differentiators to make it even better. We’re working on X3D, we’re improving it.” — Donny Waligorski, AMD
You mean that's why its cpu-z score is so pathetic? Why should ES be any different from retail ones on behalf of ST boost capability?Engineering Sample
You mean that's why its cpu-z score is so pathetic? Why ES should be any different from retail ones on behalf of ST boost capability?
It seems as if AMD tech PR has a KPI that forces them to do their best to lower initial sales and dampen hype. Not that I'd bought vanilla Z5 anyway, since it seems there won't be any bw improvements at all (so no extra perf for me).Things just got very, very interesting. Hoards of gamers will now skip the initial Ryzen 9000 SKUs once this news spreads like wildfire.
SPEC uses x264 with --disable-ASM, which disables all pre-written aseembly optimizations, meaning you are only getting autovectorization. It's due to the policy of the benchmark but note that this is 100% crack mode of usage, it's completely unrealistic for actual x264 encoding, which relies MASSIVELY on hand-optimized code and the C code it uses doesn't really get autovectorized very usefully. Too complex and needs too much clever handling to extract the performance x264 needs/achieves.I made a run without and with native gcc flag on a CPU supporting AVX-512. Quickly looking at the disassembly it looks like only AVX2 was used (but I only looked at x264, so I might be wrong). The result is a mixed bag but it proves that some integer tests were auto vectorized.
Eh, what? Doesn't Zen5 on 8-series motherboards offer support for higher memory clocks?since it seems there won't be any bw improvements at all (so no extra perf for me).
No? Who told you that? If IO die hasn't changed, you won't see anything better than 8000 mt/s memory and 2200 ish IF (and i'm already using 7800 / 2166 on a 2dpc board)Eh, what? Doesn't Zen5 on 8-series motherboards offer support for higher memory clocks?
Poor souls wasting their time. How bored they must be coz their favorite brand launches their nextgen in October and then they have to spend something like $1000 to get new mobo and maybe a not very interesting Core Ultra 7 whereas existing AM5 users spend less than that and enjoy a 9950X. Really think Intel should swallow their pride and release ARL on LGA1700 with a socket adapter or something. In a month. To prevent a mass exodus of their gamer base jumping onto AM5.
I'm not interested in measuring encoding maximal performance. I don't care about it at all. If I want max encoding speed, I expect the encoding SW to use the platform HW block not the CPU. While SPEC is a CPU benchmark which is what I'm interested in.SPEC uses x264 with --disable-ASM, which disables all pre-written aseembly optimizations, meaning you are only getting autovectorization. It's due to the policy of the benchmark but note that this is 100% crack mode of usage, it's completely unrealistic for actual x264 encoding, which relies MASSIVELY on hand-optimized code and the C code it uses doesn't really get autovectorized very usefully. Too complex and needs too much clever handling to extract the performance x264 needs/achieves.
What is the highest speed listed on the specs page for that mobo? My Z790 mobo lists DDR5-7200 as maximum and QVL also has maximum 7200 MT/s modules listed under Raptor Lake. Max for Alder Lake is 6200 but I can run 7000 MT/s with my 12700K. Similarly, QVL list for ASROCK's X670E mobo lists max DDR5-7000 kits. It is possible that X870 chipset mobos get 8000+ MT/s support through higher quality traces.(and i'm already using 7800 / 2166 on a 2dpc board)
Apple has been delivering consistent year to year performance improvements. Apples M3->M4 transition alone is comparable to Zen4->Zen5 in absolute performance increase. I too think that AMD has been making great progress with the Zen family. I don’t see however how their momentum is more sizable than ARM’s or Apple’s. Not to mention the still large (but shrinking) difference in power consumption.
Most of us here got into PC's because of games
Software has to enable and empower great hardware to the end user, and not in a closed wall eco-system.