4 ms·

I think the guy was just trolling. Esp that TLB stuff thats quite speculative and if true means memory hopping, write heavy, gpu bound apps can't squeeze all of

by aenis 4y ago

I think the guy was just trolling. Esp that TLB stuff thats quite speculative and if true means memory hopping, write heavy, gpu bound apps can't squeeze all of the fastest, niche version of m1 ultra. I mean, thats just ridiculous. Its obvious to any programmer that squeezing all from a processor requires optimizations, usually for rapidly diminishing returns. My m1 max laptop is so good that if it broke today, I'd lament having to wait for a replacement, but wouldnt hesitate. Felt the same way about m1. And thats coming after three decades of low level programming work on dos, windows and linux x86 with the best hardware money could buy.

paulmd 4y ago
If you've hung around some of the previous M1 threads, I really don't think he is trolling. That's actually how some people feel about it. I gotta say it's been really wild watching "epistemic closure" in action. There are clearly two different sets of facts and two different realities based on them that have emerged at this point, one where M1 is slower than x86 and requires everything to be emulated and USB doesn't work and you can't use external displays and it's a closed and locked-down system that is NEVER going to get a linux port, and one where it's got massive IPC and great battery life/etc and everything basically Just Works and the linux people are thundering along and have a basically workable system. People are living in like totally different worlds, one where it's just more apple trash and one where it's the best thing since sliced bread and actually better than the x86 alternatives. I dunno if it's people who just haven't used it, or people who are ideologically opposed to apple's designs, or just super hate macos, or what, but even in the benchmark space it's become completely bimodal, you've got Anandtech where it's showing up to 3.5x the IPC at 0.7x the clock of competing designs in SPEC cpu benchmarks suites, and you've got Hardware Unboxed where it barely edges out current designs at iso-power even in ARM-native builds of Cinebench. https://www.anandtech.com/show/17024/apple-m1-max-performance-review/5 https://www.anandtech.com/show/17024/apple-m1-max-performanc... https://www.youtube.com/watch?v=X0bsjUMz3EM&t=726 https://www.youtube.com/watch?v=X0bsjUMz3EM&t=726 tbh I've been thinking something similar in general - the synthetic stuff seems amazing, but it does seem like it has some problem actually putting power to the ground in real applications. Cinebench is native, so that shouldn't be a factor, and it should be something that makes great use of that absolutely wild FP32 IPC, so why isn't Apple running away with it anywhere near as much as SPEC shows it should? There's something weird there and I can't quite puzzle it out.
- jeffbee 4y ago
  I don't think those final two things are necessarily contradictory. It's very possible to have high theoretical IPC that applications cannot exploit.
  - kergonath 4y ago
    That some applications cannot exploit. Because it’s also possible that other applications actually do. Cherry picking application to discuss performances on a given architecture is not really better than cherry picking benchmarks.
- aenis 4y ago
  I'd guess people who used workflows causing their x86 laptops run hot and loud felt very relieved when switching to dead-quiet-yet-quite-fast m1's. I am one of those people. I travel a lot, and using a $5k xeon laptop meant bringing a heavy 180w brick on flights, and lugging it everywhere with me. And suffering from loud fans. Especially when having a teams or hangouts call with people sharing video. Or using docker compose. Switching to a machine that can do it all, while being quiet and lasting a, say, AMS-SFO flight without charging, felt great. As for usb3 speeds, I mean, oh well. I probably wouldn't notice if the ssd runs at 400 vs. 500mb/s. Minor speed differences are less noticeable than quality of life things like noise, battery life, screen quality, stuff like that. To each their own.
- superkuh 4y ago
  It really is fascinating. No matter how much I say otherwise people always interpret my criticism of the M1 computers in terms of speed. Like I'm denying it's fast and that's the only thing that matters. I'm not and haven't been saying M1 computers are slow. They're obviously, undeniably, fast. I'm saying they're incomplete in terms of hardware/firmware. It's a critism of the hardware not implementing full protocols so that devices work and so that you can change and upgrade your computer. Some people don't want or need to do that, just like I'm not very good with automobile maintainance and don't care about parts availability for my car or that it has all electrical ground points in the engine compartment inaccessible behind plastic. Someone else can deal with that. And that's what Apple computers are. Bad computers that are someone elses problem.

by aenis 4y ago

paulmd 4y ago

If you've hung around some of the previous M1 threads, I really don't think he is trolling. That's actually how some people feel about it. I gotta say it's been really wild watching "epistemic closure" in action. There are clearly two different sets of facts and two different realities based on them that have emerged at this point, one where M1 is slower than x86 and requires everything to be emulated and USB doesn't work and you can't use external displays and it's a closed and locked-down system that is NEVER going to get a linux port, and one where it's got massive IPC and great battery life/etc and everything basically Just Works and the linux people are thundering along and have a basically workable system. People are living in like totally different worlds, one where it's just more apple trash and one where it's the best thing since sliced bread and actually better than the x86 alternatives. I dunno if it's people who just haven't used it, or people who are ideologically opposed to apple's designs, or just super hate macos, or what, but even in the benchmark space it's become completely bimodal, you've got Anandtech where it's showing up to 3.5x the IPC at 0.7x the clock of competing designs in SPEC cpu benchmarks suites, and you've got Hardware Unboxed where it barely edges out current designs at iso-power even in ARM-native builds of Cinebench. https://www.anandtech.com/show/17024/apple-m1-max-performance-review/5 https://www.anandtech.com/show/17024/apple-m1-max-performanc... https://www.youtube.com/watch?v=X0bsjUMz3EM&t=726 https://www.youtube.com/watch?v=X0bsjUMz3EM&t=726 tbh I've been thinking something similar in general - the synthetic stuff seems amazing, but it does seem like it has some problem actually putting power to the ground in real applications. Cinebench is native, so that shouldn't be a factor, and it should be something that makes great use of that absolutely wild FP32 IPC, so why isn't Apple running away with it anywhere near as much as SPEC shows it should? There's something weird there and I can't quite puzzle it out.

jeffbee 4y ago
I don't think those final two things are necessarily contradictory. It's very possible to have high theoretical IPC that applications cannot exploit.
- kergonath 4y ago
  That some applications cannot exploit. Because it’s also possible that other applications actually do. Cherry picking application to discuss performances on a given architecture is not really better than cherry picking benchmarks.
aenis 4y ago
I'd guess people who used workflows causing their x86 laptops run hot and loud felt very relieved when switching to dead-quiet-yet-quite-fast m1's. I am one of those people. I travel a lot, and using a $5k xeon laptop meant bringing a heavy 180w brick on flights, and lugging it everywhere with me. And suffering from loud fans. Especially when having a teams or hangouts call with people sharing video. Or using docker compose. Switching to a machine that can do it all, while being quiet and lasting a, say, AMS-SFO flight without charging, felt great. As for usb3 speeds, I mean, oh well. I probably wouldn't notice if the ssd runs at 400 vs. 500mb/s. Minor speed differences are less noticeable than quality of life things like noise, battery life, screen quality, stuff like that. To each their own.
superkuh 4y ago
It really is fascinating. No matter how much I say otherwise people always interpret my criticism of the M1 computers in terms of speed. Like I'm denying it's fast and that's the only thing that matters. I'm not and haven't been saying M1 computers are slow. They're obviously, undeniably, fast. I'm saying they're incomplete in terms of hardware/firmware. It's a critism of the hardware not implementing full protocols so that devices work and so that you can change and upgrade your computer. Some people don't want or need to do that, just like I'm not very good with automobile maintainance and don't care about parts availability for my car or that it has all electrical ground points in the engine compartment inaccessible behind plastic. Someone else can deal with that. And that's what Apple computers are. Bad computers that are someone elses problem.