- Mar 3, 2017
- 1,747
- 6,598
- 136
Yeah. It’s just fanobyism. Same thing happened with Intel, people for years said 4 cores is all you need. Then Zen came out and suddenly 8, 12, 16 cores became desired.Same here. AMD should introduce 9970x with 1 CCD Z5 X3D and 1 CCD with Z5c. However for some reason there seems to be so much resistance for increasing core count in this forum. For me 4 generations of 16 cores is not a smart decision by AMD.
Zen50% bros are we back?
Zen5 @ 5.7ghz = ~3950points according to that scaling
We have to put SKTcels down before someone eyeballs cluster die area.Zen50% bros are we back?
Here is someone comparing that 5.1GHz Strix to a 5.1GHz 7840HS Phoenix - GB versions match and both runs were done at Windows. IMO, this should be a fine comparison in the 1T department.X Elite and ARL good luck
This run averages 3.9 and peak is like 4.2 so no, this isn't 5.1 Strix.Here is someone comparing that 5.1GHz Strix to a 5.1GHz 7840HS Phoenix - GB versions match and both runs were done at Windows. IMO, this should be a fine comparison in the 1T department.
Yet the difference is not spectacular: https://browser.geekbench.com/v5/cpu/compare/22547333?baseline=22555221
Almost within reach of Phoenix scores within 3.6G!!
2950--> 4000+ @ 5.75G on DT partsZen5 @ 5.7ghz = ~3950points according to that scaling
Almost seem too good to be true 👍
This sample is not having a stable boost clock, it is fluctuating between 3.6 to 4.3G.Here is someone comparing that 5.1GHz Strix to a 5.1GHz 7840HS Phoenix - GB versions match and both runs were done at Windows. IMO, this should be a fine comparison in the 1T department.
Yet the difference is not spectacular: https://browser.geekbench.com/v5/cpu/compare/22547333?baseline=22555221
// Well, it seems Strix ran at ~3.6GHz compared to 5.1GHz Phoenix
x264 can benefit from auto vectorization. See this for instance. I had also found some article that showed benefits on SPEC 2k gzip, but I can't find it any more.Can you point to the particular SPECint benchmark, what compiler will do it, and how much it improves the score? Now SPEC*fp* I can believe, especially the Fortran code, but I could care less how well a compiler will handle Fortran code, or really floating point code in general. That's much less relevant to latency dominated / interactive type performance that forms most people's impression of whether "this PC is slow" or not.
And I certainly won't disagreeBut this is once again the reason I primarily pay attention to gcc/llvm/clang subtests in SPECint and Geekbench rather than the overall score. You aren't going to speed that subtest up with SIMD to any degree, not even if you could hand code it. And to those who object "how often do you compile stuff" that's not the point. The type of work those compilers perform is a fairly good representation of code you see in a lot of modern computing, from handling the GUI, to processing XML/HTML/CSS/etc. in a browser, to other programming/scripting languages like Javascript/Perl/PHP. If you want to know how well a CPU does processing PHP then run a PHP benchmark. If you want to know about its all around integer performance, not dominated by any one factor, not relying on SIMD and other extensions, and not overly influenced by either very large caches or high frequencies, the gcc/llvm/clang subtest is what you want.
People can feel free to disagree, but take they should take their argument to Linus Torvalds. He holds the same position, and would provide a much better defense of it than I. If they can convince him, I'll be convinced too. I've seen people try, and he's demolished them in hilarious fashion every time.
Not until I see timespy run on Strix.So with that all said, are we back?
Here's a third theory for ya - the thermal density is so high, and AMD hasn't bothered with sanding the dies again (in addition to placing and gluing "structural silicon" on top of all dies), so a chiplet can't handle even 65W now at reasonable voltages and clocks.What is up with AMD using considerably lower base clock speeds and TDP's?
Yessir most definitely Zen5%.Who wants to make a bet this is the actual performance of Strix? I’ll be the party pooper and bet this is within the margin of error for final performance.
Here's a third theory for ya - the thermal density is so high, and AMD hasn't bothered with sanding the dies again (in addition to placing and gluing "structural silicon" on top of all dies), so a chiplet can't handle even 65W now at reasonable voltages and clocks.
Well, it's technically +6.5% if you go by the average 7840HS score which is 2389. It's also geekbench, there's a ton of variability and this could be a low scoring sample. Using the napkin math of HXL, 3400 / 2389 nets +42.3% GB6 1T score over a 7840HS. It'd also mean an improvement of 69.5% in nT performance.Yessir most definitely Zen5%.
8955HS scores like 2750-ish in GB6 1t at so that's 23% bump in perf.Using the napkin math of HXL, 3400 / 2389 nets +42.3% GB6 1T score over a 7840HS
See that's where the funny bit crawls in.All of this from a core that AMD claims is a +16% IPC increase running at the same clocks.
One might say they could have run this show back in April.Remember, a real launch has pricing and exact availability.
This was a preliminary launch, nothing was final.
Same with Intel, both were playing games in their presentations.
No it doesn’t. The average for a 8945HS is 2380. I doubt there’s a mythical Hawk SKU that nets that much more 1T performance.8955HS scores like 2750-ish in GB6 1t at so that's 23% bump in perf.
No need to bet, we have AMD on the record showing us just a 5% lead over X Elite in Surface 15". A pity they messed up the end note so we don't know if it's ST or MT score.Who wants to make a bet this is the actual performance of Strix? I’ll be the party pooper and bet this is within the margin of error for final performance.
Please for the love of God, stop using GB averages, they accumulate results that use silly power/performance profiles.The average for a 8945HS is 2380
Hamoa published 1t GB6 was like 3k+ so yea? Not far off.By AMD’s own slides it claims to be a mere +5% better than SDXE in GB6 1T.