Wasn't this assuming so OS side patches, or something like that, I remember in footnotes they somehow alluded to tuned software for the highest perf improvement they quoted.It's 36% performance gain, not IPC. An X925 clocked at 3.8 GHz will score ~3000 points in Geekbench 6 Single Core, on par with Lunar Lake and AMD Strix Point.
Ah, that would be unfortunate as it would suggest a cut down die, to be honest, or something that did not pass tests to become Strix Point, hmmOlrak was under impression that 3+5 is cut down strix. Bit it turned out to be Kraken. Here we come fibonacci
https://x.com/Olrak29_/status/1810617934240231540
https://x.com/Olrak29_/status/1810617501593817136
https://x.com/Olrak29_/status/1750579185180119157
Truly measuring IPC involves counting both the actual instructions and the actual cycles. Which is possible to do but somehow hardly anybody does.
Yes; IPC measurements have their uses more (but not exclusively) in microbenchmarking and less in "macro"benchmarking. Hardly anybody performing IPC measurements is largely for the same reasons as hardly anybody performing microbenchmarking.The issue is that IPC might be misleading and not exactly what people are looking for. For example in Zen4 case, and avx512 vs avx2 […]
Sounds as if we both conclude that Geekbench(6?) is not a very practical workload when the idea is to determine the influence of clock speed on performance. One could still resort to fixed-clock GB runs on platforms which support it. But this in turn would not say a lot about practical performance of GB-like workloads on platforms on which largely variable clock speed is part of the designed mode of operation.I'd love to get the average clocks per GB6 subtest with iso-cooling and iso-cache. It's simply hard data to get; that is, I hope I'm not inadvertently neglecting some great data sources that you might suggest? Anyone who's seen a clocks over time chart during a Geekbench run knows it's a mess.
"For most or all games we report on, CPU performance, as an isolated factor, has got only marginal influence on game performance. Get back with us in five years or so for any updates on the state of affairs" is not something many media outlets would be comfortable with to spell out to their readership or viewership.my point was, that 5% gaming improvements weren't something unheard of before even across 2 generations. The whole media coverage of Zen5 sounds like it is an utter failure of a chip. Just going by the titles of some videos one could think it's actually worse than Bulldozer for gaming... It's a bit tiring. That's all.
Isn't there already a cut down version of Strix with 4+6 and 75% iGPU at greatly reduced clocks?Wasn't this assuming so OS side patches, or something like that, I remember in footnotes they somehow alluded to tuned software for the highest perf improvement they quoted.
Ah, that would be unfortunate as it would suggest a cut down die, to be honest, or something that did not pass tests to become Strix Point, hmm
Golden Cove is a 2021 core tho.Raptor Lakes best submission is 452W for 56K.
Looks like another WR for Zen 5. Architecture has potential, just doesnt look like it really can be fully realized with 4nm and current IOD/mem setup. 353W for 60K on 4nm.
Raptor Lakes best submission is 452W for 56K.
Splave`s Cinebench - R23 Multi Core with BenchMate score: 56284 cb with a Core i9 14900KS
The Core i9 14900KS @ 7500MHzscores getScoreFormatted in the Cinebench - R23 Multi Core with BenchMate benchmark. Splaveranks #null worldwide and #1 in the hardware class. Find out more at HWBOT.hwbot.org
All aboard the Zen6 hype train!zen 6 with 2/3nm, DDR6, new IOD and moonshot pathway bugs fixed will destroy the benches themselves 🥋🥋
announcement should be just 12 months away
Is that speculation or does he have a source 🤔if Zen6 is indeed 2027 as Kepler said
Kepler definitely has a source, but we all have to keep in mind that roadmaps are dynamic and plans can change very quickly.Is that speculation or does he have a source 🤔
Or sources can sometimes give wrong information!Kepler definitely has a source, but we all have to keep in mind that roadmaps are dynamic and plans can change very quickly.
* Wrong informationsOr sources can sometimes give wrong information!
Indeed.The moment it is given to someone, it may already altered or completely wrong. A lot of people have trouble with this part.
Raptor Cove is not the same as Golden Cove. In addition, Raptor Lake is 24 cores, which is the only reason it is as close as it is to begin with.Golden Cove is a 2021 core tho.
yes it is , it´s just GC with 2 MB L2 cacheRaptor Cove is not the same as Golden Cove. In addition, Raptor Lake is 24 cores, which is the only reason it is as close as it is to begin with.
Adding L2 makes it different by definition. Its also using Intels 3rd gen enhanced superfin 10nm process while Golden Cove used 1st gen. That alone lets it clock >10% higher. Its not the same, lol. Thats like saying Zen 3 and Zen 4 are the same because Zen 4 has more L2 as well.yes it is , it´s just GC with 2 MB L2 cache
Well, is Zen 5 in Strix actually Zen 5 then?Adding L2 makes it different by definition. Its also using Intels 3rd gen enhanced superfin 10nm process while Golden Cove used 1st gen. That alone lets it clock >10% higher. Its not the same, lol. Thats like saying Zen 3 and Zen 4 are the same because Zen 4 has more L2 as well.
Yes. Why wouldnt it be?Well, is Zen 5 in Strix actually Zen 5 then?
Yes with cutdown AVX-512 to save die spaceWell, is Zen 5 in Strix actually Zen 5 then?
Because of what you wrote before using L2 size as a reason to differentiate cores. 🤷Yes. Why wouldnt it be?
Zen 5 in Strix is far more different than RPC compared to GLC if anything.Because of what you wrote before using L2 size as a reason to differentiate cores. 🤷
Number of 512b entries is noticeably smaller. For 128/256 the number is either exactly the same or very close to Granite Ridge depending on the source.Zen 5 in Strix is far more different than RPC compared to GLC if anything.
It's not only the lack of 512-bit wide datapath, the vector PRF for 128- and 256-bit operations is smaller and the FADD latency is higher.