First Gracemont laptop available for sale.
Anybody got disposable $500 to buy and test this laptop?
May be the private caches and the L3 in Zen2 affecting it idk cause for 1T SKT can stretch it legsI think the worst result was getting beat by Zen 2 with HT, of all things.
ZEN TWO!
SMH.
View attachment 113582
I mean it's because it lacks the necessary vector extensions for this test no? It's like saying a 6950xt is on par to a 3060 because they both perform similarly in ray tracing.I think the worst result was getting beat by Zen 2 with HT, of all things.
ZEN TWO!
SMH.
View attachment 113582
If we keeping bashing and putting down Intel perhaps we can help them go out of business?They have a long way to go before they can challenge Zen 6 by converting this E-core into a P-core. Just like Lion Cove, they may end up hobbling their savior unified P-core through various deficiencies and oversights. And they ABSOLUTELY need HT.
I don't want Intel to go out of business. I want them to wake up and start doing the RIGHT things, like AMD did in 2017 with Zen, after 10 years of crap !If we keeping bashing and putting down Intel perhaps we can help them go out of business?
Then Zen 6 can come out in 2036 instead of 2026. No reason to get it out next year if there is no competition right?
I mean it's because it lacks the necessary vector extensions for this test no? It's like saying a 6950xt is on par to a 3060 because they both perform similarly in ray tracing.
No I mean versus Zen4/5 vector performance isn't comparable because it lacks vector extensions and other vector design optimizations. Comparing Skymont to Zen 2 is like comparing a 6950xt to 3060.Which vector extension are they missing compared to Zen 2?
No I mean versus Zen4/5 vector performance isn't comparable because it lacks vector extensions and other vector design optimizations. Comparing Skymont to Zen 2 is like comparing a 6950xt to 3060.
Even compared to Zen2, skymont has a 128-bit SIMD unit width versus Zen2's 256-bit width, which also puts skymont at a disadvantage in vector performance as well.
Relevant sure, but how much vector performance contributes to the performance of the core in your average consumer workloads is debatable.But isn’t that relevant when talking about E-cores being able to compete with P-cores at lower area and power?
The issue is that Mendocino was made with Zen 2...I think the worst result was getting beat by Zen 2 with HT, of all things.
ZEN TWO!
SMH.
View attachment 113582
Relevant sure, but how much vector performance contributes to the performance of the core in your average consumer workloads is debatable. Ray tracing is also a factor when considering the performance of two different graphics cards, but many people would say that rasterization is much more important than ray-tracing thus it's unfair to compare a 6950xt to a 3060 based solely on ray-tracing.
I mean if vector performance is so important why did Intel fuse off avx-512 silicon in multiple generations of golden-cove and it's derivatives.
Workloads such as Lib264 encoding demonstrated here would have benefited a ton from AVX-512. If vector performance was so important for the general use-cases of consumers then intel left double digit performance improvements on the table.
There also is the question of just how easy it is to solve this specific issue. I'm not CPU architect but it seems to me that having relatively weak vector performance is much less of a problem to fix than having weak int/fp performance to begin with. With skymont already being such a good base, why would adding features and improvements to beat a p-core so unfeasible?? It's already 1/3 of the area of golden cove while offering similar IPC in many scenarios while being able to clock 5Ghz without being designed for high clocks. I'd say there's much room left for improvement.
Yeah and SKT is present in exactly 0 value parts. Next.The issue is that Mendocino was made with Zen 2...
No one is debating that, but nonetheless skymont cores are excellent for what they are designed to do and a good showing of what might come in the future.These are general purpose cores meant to be sold in pretty much every segment, if you're going to completely replace the P-core, it needs to be able to do compete in pretty much every category with at least decent competitiveness.
And also maybe because it's not entirely useful for consumers??? I get your argument for all these AVX support for datacenter but I really don't see how it's such a big deal for 99% of consumers buying a cpu for their PC.Because the E-cores didn't support it and they couldn't figure out how to make a mixed instruction set big-little CPU work in software. AVX-512 is all kinds of messy as well for Intel but that's a whole different discussion.
Criticized for wasting die space and money on AVX-512 sure. But I don't think anyone was claiming that somehow giving AVX-512 support for CPUs was going to revolutionize their product. Zen5 is the first full fat AVX-512 implementation and yet it barely improves upon zen4 in any meaningful way for consumers.Yep, and it just led to further confusion and difficulties around adopting AVX-512. They got criticized for it quite a bit.
For consumers, a few tweaks would indeed lead to a world class, small, efficient core while also being much cheaper to produce. I literally linked a benchmark video where they turned off all the P-cores except one and left all 16 e-cores on which led to faster gaming performance than the default configuration. Ofc some leeway should be given here because the default configuration has scheduling issues, but if you compare the results to the 8P test and the new FIXED cyberpunk benchmarks then you see it is pretty close if not on par.I'm not saying it's impossible, or even ill advised, for Intel to move to their E-core architecture entirely, just that the comments of adding a few tweaks and having a world class, small, efficient core top the charts is not realistic. It may compete extremely well in many thread scenarios and in many integer heavy workloads, but will still get smoked in several work loads.
No one claimed that designing an E-core to replace all the functions of the P-core would result in a core 1/4 the size of the current P-core. I agree, the core would look much bigger than an actual e-core maybe more like 60-70% the size of a P-core rather than 25-33% which we see now. To fully replace the P-core would take significant work but at the same time what CPU architecture doesn't take significant work?To get to a more fully competitive design will significantly increase power and area. Adding full width support for wider vector operations not only increases power and area on its own, but you have to be able to feed the beast as well which means expanding other parts of the core. Even in lightly threaded integer workloads, you either need to increase IPC or frequency enough to compete with designs clocking 20+% higher, neither of which is easy and will add additional area and power. The point is, by the time you do all that, the E-core is not going to look so much like an E-core anymore. If done right, it will probably be much better than what the P-core team can put out, but it's not going to be a few tweaks here and there and you get an E-core competing in the competitive P-core landscape. That's just not realistic.
It was disabled because of scheduling problems on hetero-ISA (and also because it's bad optics to have a CPU that performs better in a small subset of tasks with E-cores disabled, because some mobos actually allowed enabling AVX-512) CPUs IIRC.I mean if vector performance is so important why did Intel fuse off avx-512 silicon in multiple generations of golden-cove and it's derivatives. Workloads such as Lib264 encoding demonstrated here would have benefited a ton from AVX-512. If vector performance was so important for the general use-cases of consumers then intel left double digit performance improvements on the table.
U behave like an imbecile in every thread I see you in lol, at least stay in AMD threads if you're gonna behave like this. Arrow lake is shit, BMG is largely unprofitable as it stands, lion cove is dog and P-cores have no future. Foundry likely has made huge advances but is unlikely to be profitable for the next half-decade if not longer.Yeah and SKT is present in exactly 0 value parts. Next.
Yea but how much performance did that really leave on the table for the average consumer?? It'd be nice if those products did include AVX-512 support but once again unless you're doing heavy productivity work that also happens to rely heavily on vector performance, you'd be perfectly fine. It's like how some people are very adamant on Intel releasing a 12-core HEDT product with only P-cores and AVX-512 support but to be perfectly honest most people don't care and it's a tiny segment of the consumer market.It was disabled because of scheduling problems on hetero-ISA (and also because it's bad optics to have a CPU that performs better in a small subset of tasks with E-cores disabled, because some mobos actually allowed enabling AVX-512) CPUs IIRC.
Sapphire Rapids had AVX-512 support.
Meds.U behave like an imbecile in every thread I see you in lol, at least stay in AMD threads if you're gonna behave like this
It's fine. Z4c-ish with castrated SIMD is pretty good but not exactly jack off materiel either.Skymont is indeed impressive
Clock it at 4.9 instead of 4.6 and it's way close to Z4 rather than Z4c. Regardless for x86 it certainly is exciting especially given it's rumored to be the future of Intel products, and compared to the P-cores and previous E-cores it is jack-off material. Clearwater forest may be the first time Intel has had a competitive data center product in 5-10 years.It's fine. Z4c-ish with castrated SIMD is pretty good but not exactly jack off materiel either.
If you want a real impressive LITTLE, look at AAPL stuff.
It is not so simple. Skymont has 4x128b symetrical SIMD units, Zen2 has asymetrical units so for inner loops of things like matrix multiply they are equally matched.Even compared to Zen2, skymont has a 128-bit SIMD unit width versus Zen2's 256-bit width, which also puts skymont at a disadvantage in vector performance as well.
Past stupid decision when Intel thought it owned the marked, so they thought market segmentation will be good for them as they would be able to milk avx512 as a bonus feature. So it got completly ignored in consumer market until Rocket/Tiger lake and then Intel cut the support after one gen as they needed E-cores to save face in Cinebench.I mean if vector performance is so important why did Intel fuse off avx-512 silicon in multiple generations of golden-cove and it's derivatives
Not a lot, that's correct. I was mostly arguing that Intel didn't leave performance on the table because it was irrelevant, they did because they had no other option.Yea but how much performance did that really leave on the table for the average consumer?? It'd be nice if those products did include AVX-512 support but once again unless you're doing heavy productivity work that also happens to rely heavily on vector performance, you'd be perfectly fine. It's like how some people are very adamant on Intel releasing a 12-core HEDT product with only P-cores and AVX-512 support but to be perfectly honest most people don't care and it's a tiny segment of the consumer market.
Agree. But given Intel's history, they may release at most an 8-core Skymont die if they actually decide to release an ADL-N successor. They seem to be allergic to making enthusiasts happy.I don't believe that an E-core product with only a few tweaks would outright beat the current lineup of Zen5/Arrowlake either. But if you can get 90% of the performance, essentially N-1 performance of Zen4/12-13th gen, and then sell that 16-core product for 200$ it'd already be an insane product for pretty much everyone with a cpu older than 3-4 years or a cpu with 8 or less cores. For example currently I am on a 9600k with 6-cores and around half of the single thread performance of the newest CPUs, not to mention the nT performance. If i could pay 200$ for a 16-core CPU (even without hyperthreading) that has Zen4 performance for most consumer applications and gaming, and much better power efficiency compared to even leading edge I would immediately upgrade no issue.
OFC this product is all theoretical because even though it may be a good product on paper, it may cost lots of engineering time and money for a low-margin budget product that might not sell well due to not being the newest and greatest thing. Also it might mess up Intel's own segmenting of their products and overall would not be good for their product stack. Regardless it is absolutely the case that an E-core only cpu using skymont with a few tweaks/higher clocks would be great for MOST consumers especially since the die cost of such a cpu would be a fraction of other 16-core or 24-core CPUs.