Unfortunately, it's a lot lower in terms of IPC it seems. AMD cherry picked the benchmarks for that IPC calculation. If you remove the GB5 benchmark from that list, geomean score immediately falls below 15%.Regarding 16%.
I cant really get my head around this, i mean whats the motivation to perpetuate those claims here (while often being dismissive and downright insulting to other people). We know why the likes of MLID do it, its grifting for money in their case, but what is to gain from spouting it here, on random tech message boards?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.
More interesting to me is how ideas for how to gain performance on the cores are converging across all the CPU architectures.Regarding 16%. The fruit is getting harder and harder to reach with every generation.
What do you mean by "only"? If you compare results from the various outlets who tested either with JEDEC or AMD recommended specs (expo/xmp with more decent CL30 kits), it is actually disfavouring zen4 as compared to raptor lake(-r), especially in power-hungry nT tests (because enabling xmp/expo or manually OC'ng memory cuts 20 or 30W from the power budget)I can think of another reason why AMD tested with DDR5-6000 only.
It allows much more parallelism with double the registers (newer CPUs has to be designed to make use of all the parallelism). You have to code in ASM to know the current limitations/bottleneck of x64. I always hit the 16 reg bottleneck and wanted more general purpose registers. So your statement doesn't make any sense.APX is just aa64: x86 edition. Except the whole "cleaning the ISA" part.
Wasted opportunity really.
Unfortunately, it's a lot lower in terms of IPC it seems. AMD cherry picked the benchmarks for that IPC calculation. If you remove the GB5 benchmark from that list, geomean score immediately falls below 15%.
If they had shown a DDR5-7200 EXPO kit, for example, a lot of people would've groaned that they have to replace their CPU AND RAM to gain benefits from upgrading to Zen 5. In reality though, most people (except those with really expensive flagship mobos) may have to get expensive EXPO kits and a decent X870 mobo minimum to make the most of their Zen 5 upgrade.What do you mean by "only"?
It will be more interesting to see what the geomean of AT benchmarks ends up being.I think that is the only benchmark that uses AVX-512. I don't think it is inappropriate to add 1 to the mix, given the performance increase of Zen 5 in AVX-512
Yep. As well as pray for an Arrow Lake miracleSo. Wait for 3D?
it’s not a core to core comparison as Zen 4 was weaker in AVX-512. It is a strange addition since the Zen 4 Geomean included GB 1t and not a subtest. This is, say similar to if Apple picked Object Detection subtest to show IPC improvements over M3, where we know object detection is an outlier.I think that is the only benchmark that uses AVX-512. I don't think it is inappropriate to add 1 to the mix, given the performance increase of Zen 5 in AVX-512
Intel might cook something up for Lion cove. Skymont looks like a nice uplift over Crestmont. Hoping there’s something tomorrow about lion cove.Yep. As well as pray for an Arrow Lake miracle
It all depends on ARL competitiveness. And boy it doesn’t look great...
I don't know whether to blame their marketing or "performance labs" folks. If it's the latter, maybe they outsourced it to WTFtech?It is a strange addition since the Zen 4 Geomean included GB 1t and not a subtest.
Good point, but I think vector math has broader applicability than matrix math.it’s not a core to core comparison as Zen 4 was weaker in AVX-512. It is a strange addition since the Zen 4 Geomean included GB 1t and not a subtest. This is, say similar to if Apple picked Object Detection subtest to show IPC improvements over M3, where we know object detection is an outlier.
Intel might cook something up for Lion cove. Skymont looks like a nice uplift over Crestmont. Hoping there’s something tomorrow about lion cove.
I really don’t know. Looking back at Zen 3 and Zen 4 geomean, there were a lot more tests. It seemed like AI took importance?I don't know whether to blame their marketing or "performance labs" folks. If it's the latter, maybe they outsourced it to WTFtech?
It basically uses a single instruction of AVX-512 to compute AES. That's quite a poor usage of AVX-512 that will unlikely affect many users.I think that is the only benchmark that uses AVX-512. I don't think it is inappropriate to add 1 to the mix, given the performance increase of Zen 5 in AVX-512
They couldn't find one decent AI benchmark to replace that GB legacy test to show off the better AVX-512 performance? I feel like grabbing the responsible guy by the collar and asking very rudely, WHY WHY WHY???It seemed like AI took importance?
Most gaming workloads have no power/thermal limitations.Not all workloads are power- and thermally-limited, so nT scores are a poor proxy to estimate gains in those.
At least adroc has now cited server part communications as his source, if I understood it correctly.I cant really get my head around this, i mean whats the motivation to perpetuate those claims here (while often being dismissive and downright insulting to other people). We know why the likes of MLID do it, its grifting for money in their case, but what is to gain from spouting it here, on random tech message boards?
I've seen all kinds of speculation on this, I thought memory speed support was like 90% the controller in the CPU? Certainly overclockers are getting 8000+ on X670E boards today? It would really suck if the current high end boards can't handle a modest 'sweet spot' DDR5 increase to 6400/6800/7200, when the X870 chipset has the exact same hardware.In reality though, most people (except those with really expensive flagship mobos) may have to get expensive EXPO kits and a decent X870 mobo minimum to make the most of their Zen 5 upgrade.
Yes it would and if that happens, AMD can conveniently place the blame on mobo makers for cheaping out. I'm not sure if Det0x or someone in his circle of friends is testing memory OCing with cheap mobos, like suppose the ASUS Prime X670. That bit of info could be very enlightening.It would really suck if the current high end boards can't handle a modest 'sweet spot' DDR5 increase to 6400/6800/7200, when the X870 chipset has the exact same hardware.