BobHoward posted:On the state of macOS: iOS already supports it, and that means macOS can too. iOS began life as a fork of macOS, and both still build a lot of components from the same source code. That includes the XNU kernel. While they likely haven't turned on the code paths for AMP support in Intel Mac XNU kernel builds, obviously they've got a relatively easy path to get there for ARM Macs. (AMP = asymmetric multiprocessing, Apple's chosen terminology for this) The problem with it is that while the source is published, it's only published very sporadically (ie. no schedule and no announcements). Nor is there any newer-than-Mach-scheduler code-changes or instances of the word 'AMP' that I've been able to find in a brief look through the codebase, with cscope. Plus, the code hasn't been modified since 2018. Scheduler difference-wise, it doesn't really matter how different the codebases are - a study comparing CFS with ULE found that in large there's no difference despite the fact that CFS is highly heuristics-based whereas ULE is just very optimized (perhaps ULE has a slight advantage over-all, and a bigger advantage in certain benchmarks, but the difference is probably not enough that it's going to make someone choose one over the other). As for taking the source code, that's not really a problem. The only requirements are that you cannot hold Apple liable, and you can't make trademarks based upon the code, that you must include copyright and license, state changes, and disclose the source code. FreeBSD at least would have no problems with any of those, and indeed already contains some code from Apple. Incidentally, about Mac OS X, someone named Lucy did a very interesting talk at 24C3: https://www.youtube.com/watch?v=-7GMHB3Plc8 I'm a FreeBSD contributor who's been threatened with a commit bit more than once, and I've been using it since 2000 BlankSystemDaemon fucked around with this message at 09:42 on Jul 5, 2020 |
|
# ? Jul 5, 2020 09:38 |
|
|
# ? May 8, 2024 06:07 |
|
The cheat for phones is that you can disable the big cores unless the screen is on or the phone is charging. Can't really do that on a desktop or server.
|
# ? Jul 6, 2020 03:52 |
|
Context provided by quoting my post from the AMD thread:SwissArmyDruid posted:Completely and absolutely correct use of your 64 Cores and 128 Threads The Intel camp appears to have fired back. Please be sure to unmute for best experience. Unfortunately, I think this is faked based on the fonts of the task manager? Alas. I believe it was also disclosed that the original AMD one was also faked as well. SwissArmyDruid fucked around with this message at 07:38 on Jul 6, 2020 |
# ? Jul 6, 2020 07:34 |
|
D. Ebdrup posted:The biggest problem with heterogeneous CPU cores is all the schedulers for every single OS that wants to support this has to grow support for tracking energy use / scheduler quanta, and also needs an algorithm for when something should be moved around. BobHoward posted:they can make sure that both support exactly the same ISA, which is what you really want. FunOne posted:The cheat for phones is that you can disable the big cores unless the screen is on or the phone is charging. Can't really do that on a desktop or server. Anyway I read through the Lakeview thing and just came in here to laugh at the fact that Yet Another Naming Scheme is generating "4+1", "4+2", etc. Good job y'all. Is it up to 3 internally? "2+4+2" lmao
|
# ? Jul 6, 2020 17:24 |
|
JawnV6 posted:I dunno if I'd go that far. The sticky wicket on the x86 side is the super wide AVX512, you really really don't want the Atom to have any clue what to do with that opcode, it has to call out for the big one. IMO that's a problem for big.LITTLE x86. Sure, you don't want the Atom core to have the execution unit required for 512-bit wide execution, but you could at least make sure that if a process ever attempts to run AVX512 on it, no SIGILL kill takes place. That's the approach on the ARM side: ARM's wide vector ISA, SVE, is designed with semantics which explicitly support CPU microarchitectures that emulate full width execution with multiple passes through a narrower execution unit. If you want to have your big core support big vectors, you can do that without giving up on identical ISA support and without forcing the little core to be bloated and power hungry.
|
# ? Jul 6, 2020 22:02 |
|
Intel released a bunch of stuff on TB4/USB4: https://www.anandtech.com/show/15902/intel-thunderbolt-4-update-controllers-and-tiger-lake-in-2020 So thinkin about this, specifically the portion about DMA protection, it seems likely that TB4 itself will be exclusive to Intel. I guess that means Macs will probably replace their TB3 ports with USB4 40gb/s ports and call it a day, since it is backwards compatible with TB3. Really seems like the only things Intel is offering over a good USB4 40g implementation is certifications/mandatory features?
|
# ? Jul 9, 2020 01:19 |
|
Cygni posted:Intel released a bunch of stuff on TB4/USB4: Maybe. Possibly. But also Apple released a big thing today about how they “partnered with Intel to make thunderbolt a decade ago” and would be committing to supporting it on Apple Silicon. That could mean TB3 or it could mean TB4, I doubt they’d keep the TB branding if it remained a generation behind, the whole point of TB4 seems to be to raise the uniformity and standards and whatnot. Also Intel apparently won’t be collecting any royalties or licenses, and there doesn’t appear to be a requirement for hardware that only Intel can produce so....
|
# ? Jul 9, 2020 01:23 |
|
Frankly, universal feature requirements by itself is a pretty nice step forward, instead of the current situation of grab-bag features that has everyone running to the fine print on each different model's spec sheet to figure out what the gently caress those ports actually do. The bump on internal PCIe speeds is also nice, if not really needed quite yet for most things. What I'd also like to know is if it'll let laptops put USB4/TB4 ports on both sides of the body instead of usually just the one closer to the chipset.
|
# ? Jul 9, 2020 01:28 |
|
Ok Comboomer posted:Maybe. Possibly. But also Apple released a big thing today about how they “partnered with Intel to make thunderbolt a decade ago” and would be committing to supporting it on Apple Silicon. That could mean TB3 or it could mean TB4, I doubt they’d keep the TB branding if it remained a generation behind, the whole point of TB4 seems to be to raise the uniformity and standards and whatnot. Huh, i didnt see that Apple announcement. I guess they have to say it considering the Tiger Lake laptops they are about to announce have TB4 included, and would be orphaned before launch if they werent gonna do it. The hardware requirement is the DMA portion, although apparently Intel backtracked and said it was not VT-d REQUIRED... but I'm not entirely sure how else you would do it without it? But I'm a dumbass so!
|
# ? Jul 9, 2020 01:54 |
|
Cygni posted:Intel released a bunch of stuff on TB4/USB4: FYI, I don't think TB4/USB4 can be used interchangeably, in the sense that while the connector will be the same and there will be some interoperability since USB is incorporating TB3 into the USB 4 standard, there are still changes occurring in TB4 that won't be available in USB 4. So if Apple can incorporate it into their ARM Macs, they'd probably still prefer to, so as to maintain the premium image associated with them. Or they'll just fork the USB-C connector for their own purposes
|
# ? Jul 9, 2020 02:09 |
|
Cygni posted:The hardware requirement is the DMA portion, although apparently Intel backtracked and said it was not VT-d REQUIRED... but I'm not entirely sure how else you would do it without it? But I'm a dumbass so! All it needs is some kind of IOMMU sitting between the thunderbolt interface and memory, that can remap what addresses the hardware sees. That abstraction means the OS can lock out sections of memory where it doesn't want random external hardware poking around, just by not mapping those physical addresses to any device-visible addresses. VT-d is Intel's brand name for that virtualization layer. AMD has the same basic functionality under the name "AMD-VI." Apple will come up with their own solution.
|
# ? Jul 9, 2020 20:13 |
|
Cygni posted:Intel released a bunch of stuff on TB4/USB4: The updates to AT's article after AT reached out to the various parties make it clear that "required to support Intel VT-d" was the press release writer repping the brand rather than a hard requirement for Intel's IOMMU and no other. This makes sense as VT-d is a CPU feature that isn't even tightly integrated with TB4. Intel clarification posted:Thunderbolt is open to non-Intel-based systems. Like any other system, devices must pass Thunderbolt certification and end-to-end testing conducted by third-party labs. Thunderbolt 4 requirements include Intel VT-d based or an equivalent DMA protection technology that provides IO virtualization (often referred to as IO Memory Management Unit or IOMMU), as well as OS implementation support. If the equivalent technology supports prevention against physical attacks, then that should meet the requirement. e: f,b guess this was redundant with other posts
|
# ? Jul 9, 2020 20:55 |
I'm just waiting for the hardware privilege checks in the IOMMU to be found lacking, as is/was the case with the privilege checking on the branch predictor which led to part of MELTRE.
|
|
# ? Jul 10, 2020 08:37 |
|
looks like Tiger Lake (and Willow Cove cores, Xe Graphics, and PCIe 4 for Intel by default) will launch on September 2nd https://videocardz.com/newz/intel-has-something-big-to-share-on-september-2nd
|
# ? Jul 15, 2020 19:32 |
|
Good for the laptop market, but wake me up when Intel has something interesting to talk about on the desktop again.
|
# ? Jul 15, 2020 19:52 |
|
I hope Lenovo hurries up and comes up with some Tiger Lake poo poo, I really need an update
|
# ? Jul 15, 2020 20:49 |
|
Does this signal the end of multiple whacky processes being the same Core generation?
|
# ? Jul 15, 2020 21:08 |
|
NewFatMike posted:Does this signal the end of multiple whacky processes being the same Core generation? Nope, they're going to continue cranking out 14nm on the Desktop through 2022.
|
# ? Jul 15, 2020 21:30 |
|
I've been trying to keep tabs on the arch/product segment/generation of processors that will start supporting 2nd gen Optane Persistent Memory, but the all the 'Lake names keep making it hard to remember - which one was it again, and is it hopefully not another Skylake?
|
# ? Jul 15, 2020 21:57 |
|
Skylake Cores - Skylake/Kaby Lake/Coffee Lake/Cascade Lake/Cooper Lake/Whiskey Lake/Comet Lake (some variations in compute units and stuff but broadly the same family) Palm Cove core - Cannon Lake (mostly still skylake shrunk to 10nm) Sunny Cove core - Ice Lake/Lakefield Willow Cove core - Tiger Lake (10nm) / Rocket Lake (14nm) Golden Cove core - ? Sapphire Rapids? Alder Lake? Cygni fucked around with this message at 22:22 on Jul 15, 2020 |
# ? Jul 15, 2020 22:10 |
|
Which one does Cooper Lake fall under Edit: Ah, here we go, this is what I was looking at: https://www.intel.com/content/www/us/en/products/docs/processors/xeon/3rd-gen-xeon-scalable-processors-brief.html Just wondering what the underlying cores in this architecture are, how they talk to each other, etc. Sidesaddle Cavalry fucked around with this message at 00:33 on Jul 16, 2020 |
# ? Jul 15, 2020 23:46 |
|
I usually dislike Linus Tech Tips but this is a pretty funny watch: https://www.youtube.com/watch?v=Skry6cKyz50 It's about Intel locking out XMP profiles to the Z series chipset in the current generation: The lock is set in the BIOS so the motherboard makers can't do anything about it. The reason for it is "well XMP is an overclocking feature so you need a Z series overclocking chipset to enable it" even though the memory controller is on the CPU it self and the chipset has nothing to do with it. This is yet another anti-consumer thing Intel has done to their product lines and it's pretty fuckin stupid.
|
# ? Jul 22, 2020 19:16 |
|
Is that even new? I thought Intel has done that for ages (locking RAM speeds to "supported" JEDEC speeds outside of Z-series chipsets)?
|
# ? Jul 22, 2020 19:33 |
|
It was always an option for motherboard designers to enable if they wanted to but now you must have a Z chipset.
|
# ? Jul 22, 2020 19:36 |
|
orcane posted:Is that even new? I thought Intel has done that for ages (locking RAM speeds to "supported" JEDEC speeds outside of Z-series chipsets)? it's not new, and not quite accurate (you can exceed JEDEC speeds up to the rated limit of the processor, just not to arbitrary unlimited speeds like Z-series, this is 2933 for Z490 I believe). it's sort of come to prominence now, because AMD is close to a lot of the lower-tier processors due to the locked clocks, and memory overclocking is something that would push Intel ahead another couple percent, but Linus is being deceptive here presenting this as some kind of a "new" lock, you have never been able to break official DRAM speeds on a non-Z board. As we enter year 4 of skylake refreshes, it's really just time for the overclocking limits and memory clock limits to die. Unlock memory clocks across the whole lineup on all boards (or maybe keep like one or two for "business" models), unlock clocks on all processors on Z-series boards. Paul MaudDib fucked around with this message at 19:42 on Jul 22, 2020 |
# ? Jul 22, 2020 19:40 |
|
Yeah I meant the CPU's supported max. RAM speed.
|
# ? Jul 22, 2020 19:54 |
|
Paul MaudDib posted:it's not new, and not quite accurate (you can exceed JEDEC speeds up to the rated limit of the processor, just not to arbitrary unlimited speeds like Z-series, this is 2933 for Z490 I believe). Part of Linus' rant here is that AMD doesn't do any of this limiting at all based on chipset (or seemingly much at all) and overall the rant is "stop segmenting this poo poo, it's dumb now" and he's right on that one.
|
# ? Jul 22, 2020 19:55 |
Don't worry, it's a race to the bottom; AMD will begin segmenting as soon as they think they can. Case in point, they've already dropped ECC support from many lower-end chips.
|
|
# ? Jul 22, 2020 20:41 |
|
Number19 posted:I usually dislike Linus Tech Tips but this is a pretty funny watch: this is what I should do whilst my product line is under assault from all sides
|
# ? Jul 22, 2020 21:48 |
|
so you buy a locked CPU, and decide to pair it with an H410, B460, or H470 motherboard, since you don't need the overclocking capabilities of a Z490 except you're leaving performance on the table because your memory speeds are capped/can't be tuned as tightly so it pushes you towards wanting a Z490 board instead except pairing a locked CPU with an overclocking board is dumb but getting both an unlocked CPU and an overclocking board is also that much more expensive but kneecapping your memory by using a non-Z490 chipset is also dumb is that about right? seems bad
|
# ? Jul 23, 2020 03:11 |
|
gradenko_2000 posted:so you buy a locked CPU, and decide to pair it with an H410, B460, or H470 motherboard, since you don't need the overclocking capabilities of a Z490 yeah, intel is suffering from the paradox of choice, even if you found one of those combinations to perform acceptably... they are rubbing it in your face that you are missing out on something. AMD gives you a fully enabled system, where overclocking is largely pointless and memory overclocking is necessary to even catch up to a locked Intel configuration, but it's "fully enabled", isn't that just mentally easier? it is the same thing as why the "hyperthreading DLC" produced a backlash and the 4700U or 9700K does not. Give people a choice and they hate it. Intel persisting in over-segmenting their lineup in the face of an incredibly competitive AMD (especially in the lower-end segments where this becomes a price+performance concern) is loving stupid but on the other hand it's pretty easy to talk yourself into circles like that on all kinds of products, and the paradox of choice is the term for that. Paul MaudDib fucked around with this message at 03:33 on Jul 23, 2020 |
# ? Jul 23, 2020 03:28 |
|
movax posted:this is what I should do whilst my product line is under assault from all sides STILL THE FASTEST IN SINGLE THREADED PERFORMANCE!
|
# ? Jul 23, 2020 03:31 |
|
Also lol @ their lack of ECC support. Lets have 32GB+ of ram but no ecc, just buy our xeons if you want ecc memory to go with our gob of cores.
|
# ? Jul 23, 2020 03:40 |
|
I'm a VFX artist and Intel's behaviour over the last decade has made sure I'm sticking with AMD as long as their CPUs are within ~15% of the performance and similar or better power requirement. The fact Intel stuck with 4c/8t and 6c/12t as the only affordable options for well over a decade has made me despise them. The higher core processers were prohibitively expensive and our work relies heavily on multithreading. After a decade of using 2 or 4 core processors, I bought a 5820k in 2015 and still found it lacking for multithreading. I couldn't afford a Xeon so had to suck it up. Being able to buy a 12 (or more) core processor for my home workstation seemed like a fantasy as long as AMD weren't a threat to Intel. I now have a 3950x and it's made me bitter I didn't have a 12 or 16 core processor from Intel available at an affordable price in the past. I could have saved quite a bit of money over the years from not using render farms. Sorry for the rant, but most artists I know were jumping for joy when AMD's 12+ core processors were shown to wipe the floor with Intel's offerings. Everyone I know in the industry that's upgraded their machines has a 3900x or 3950x.
|
# ? Jul 23, 2020 04:08 |
|
Why do we even need to do "warranty breaking" XMP overclocking of RAM anyway? We had 3200 in 2014, how come poo poo doesn't just work at 3600+ in 2020? Is it just that DDR5 is taking kind of a long time to arrive?
Carecat fucked around with this message at 12:33 on Jul 23, 2020 |
# ? Jul 23, 2020 12:27 |
|
Encrypted posted:Also lol @ their lack of ECC support.
|
# ? Jul 23, 2020 13:20 |
ECC is confirmed-working up to 3200MHz, so what's the issue exactly? Once you go much higher than that, the CAS latency timings begin being a problem for memory-intensive workloads which would otherwise benefit from the higher bandwidth, if memory serves.
|
|
# ? Jul 23, 2020 13:44 |
|
Combat Pretzel posted:ECC pisses me off. Why the gently caress is no one making modules with XMP profiles? If not that, JEDEC, get the gently caress cracking on higher frequencies! it's my understanding, though I could be wrong, that by DDR5 we're just going to have ECC all the time because the tolerances are so tight that you need to have ECC just to make the thing work
|
# ? Jul 23, 2020 13:46 |
gradenko_2000 posted:it's my understanding, though I could be wrong, that by DDR5 we're just going to have ECC all the time because the tolerances are so tight that you need to have ECC just to make the thing work Several studies have shown this, and although it's a pity I can't find it, I also have vague memories of some statistics published by Microsoft based on their system information gathering, that showed that ECC memory accounts for absolutely staggering amounts of lost work over the decades. EDIT: Here's one more thing from all the way back in 2000, in addition to the five things I already linked. BlankSystemDaemon fucked around with this message at 14:36 on Jul 23, 2020 |
|
# ? Jul 23, 2020 14:30 |
|
|
# ? May 8, 2024 06:07 |
|
D. Ebdrup posted:ECC is confirmed-working up to 3200MHz, so what's the issue exactly?
|
# ? Jul 23, 2020 21:49 |