I know of someone - an ex, actually - running a couple dozen instances on Graviton for exactly that (PostgreSQL and a Rails app.) But that's just anecdote and you have no particular reason to believe it.
All I can say is that I'm seeing real Graviton use in my industry, and I sure as hell don't...
I feel obligated to point out that I've benchmarked three gens of Graviton on real applications (EDA synth flows) and found it compelling. Single-thread perf (mainly talking about 3/3E here) is great. Perf across whole chips (bare metal instances) is less great but very good.
They're selling...
It doesn't matter if AMD or Intel sell at cost. What matters in terms of gaining market share is what they're selling for externally. Clearly, judging by the three largest cloud providers opting to roll their own, that's cheaper. Being able to spend a couple of tens of millions of USD on...
Baffled by claims that ARM isn't cheaper for hyperscalers... and if hyperscalers build their own ARM SoC's because it's cheaper, that doesn't count for some reason. Ooooookay.
Whatever. I will continue to think the industry is fundamentally silly until they bring me my Itanium roadmap back...
ARM most assuredly has been cheaper for hyperscalers. Hence Graviton, Cobalt, and Axion. Google said outright that they've been running a bunch of workloads on ARM servers for a while - do you really think they'd be doing that if it weren't cheaper?
Apple's cores have been well-characterized across a wide range of workloads. They are serious cores, with superb general-purpose performance. This isn't a great take IMO.
This kind of tone is beneath you, Maynard.
Your linked paper says that manual vectorization of a test suite of loops explicitly designed to be vectorizable gets another 40% boost over the tested vectorizing compiler. I don't find that enormously whelming given my experience with real-world...
Nobody in industry describes "wider" as anything but "more op throughput and execution resources" - there are clearly many ways to scale performance without increasing those, as enumerated above
Not going to respond to the Apple hagiography because I don't think it's very relevant
This isn't true at all.
One of the largest performance limiters in modern microarchitecture is correct branch prediction, which isn't about executing more ops at all, but rather about running ops as early as possible and wasting as little work as possible.
Cortex-X4 is wider than M3, and slower.
Likely. There are lots of tricks that boost your general core perf without going wider.
Deeper structures - schedulers, ROB, PRF size
Larger or faster caches (like Z, which has usually not been on the cutting edge of width but has generally had very large L1/L2)
Non-committing runahead (like...
Unfortunately not - but I'd be happy to do a run if someone has a system I can use remotely.
In my experience, even autovectorization on SPEC is very limited. I agree with your doubts about it being able to usefully emit SME.
Oryon is going to be presented at this year's Hot Chips.
I won't be attending this year - I have less than zero interest in how AI-heavy the conference has become since I last went - but I'll be interested to see what comes out of that presentation.
Sure, SPEC is a compiler benchmark to a degree, but that's not a bad thing. It means you can build it in whatever configuration approximates what the software you'll be shipping or running will use - rather than having to just roll with whatever Primate Labs gives you. That's valuable! Vendor...
I've seen it, and I agree with some of their points, though with reservations. The tl;dr version of my opinions on Geekbench is that it's pretty good, better than other free benchmarks available, but is not a replacement for SPEC even if they tend to correlate decently well with each other. It...
Geekbench isn't bad, in a pinch. SPEC is better, and now we have SPEC numbers for M4, demonstrating small but real clock-normalized gains and a decent generational clock bump.
Nobody is outraged. Nobody is saying Apple somehow cheated. People are trying to figure out what kind of general-purpose improvements can be expected from this microarchitecture. Early SPEC results have made that far clearer.
Nobody is persecuting you or Apple, and your (repetitive) reaction to...
I like gcc and perlbench - xalanc is often a good proxy for my workloads too. Not so much of a fan of xz, exchange2, or x264. Some of my reasons for loathing xz are not fully rational (it takes forever to run, and it's the only subtest that allows OpenMP parallelization in the official...
Thanks for posting the full subtests! The original video wasn't loading for me for whatever reason.
I don't really know what to make of it, honestly. I will say that xz is not my favorite subtest, for a lot of reasons, and it's the only regression. For the rest - this looks to me like...
I put in 3.8 and 4.4 (3.78 is more precise, 4.4 is from Geekbench results for M4 - like this one.)
Isn't 4.05GHz the M3 clock speed? Or am I talking crazy? (This probably doesn't matter, I'm just wondering if it's where you got 4.04 maybe?)
This thread is rapidly getting a bit silly, so I'll just make a couple of comments and then I'm probably going to step away from it a little.
Cinebench is a potato. It may be useful for... something... and I know Chipsandcheese uses it periodically, but I've never seen it act as a reliable...
Fine by me. I know you're trying to make some kind of point about people being unfair to Apple, and maybe there is some of that going on; I think my posting history makes clear I'm not one of those - but there absolutely is value, a lot of it, in knowing what a core is likely to do with a random...
Oh. So they aren't preventing anyone from running Windows - they simply aren't taking it upon themselves to do the port themselves. It isn't their operating system, so it's not their problem.
Likewise, Qualcomm is not preventing anyone from running MacOS on their hardware.
Is Apple preventing you from running Windows? How, exactly?
I don't use Apple's stuff, and I'm not the target market for it - but Apple has done absolutely nothing to stop Microsoft from supporting their hardware, as far as I'm aware. Also, the Win/ARM virtualization story on ARM MacOS is, from...
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.