|
CaptainSarcastic posted:I'm not well-versed on the matter, but I think of ARM processors as being in mobile devices and such. Aren't pretty much all cellphones ARM? I know people have been talking about laptops and servers as possible ARM markets, but it already dominates (if not completely owns) the mobile device market. It seems like that would be where the money is. ARM isn’t going away from mobile, and yes server and laptop are secondary markets currently. They’re probably higher-margin markets too, though I’m not sure how that affects licensing revenue.
|
#
?
Oct 10, 2020 10:12
|
|
|
|
| # ? May 30, 2024 10:09 |
|
|
SCheeseman posted:Buying whole parts is different from investing resources developing upon an architecture that is owned by a direct competitor. Apple could switch suppliers in a single generation, licensing and creating an ARM processor requires far more effort, time and money. They're more likely to move towards RISCV, as a lot of the industry is probably doing with Nvidia absorbing ARM. AMD already had K12, an (now old) ARM design that was entirely in house, and probably was designed with a lot of the lessons from Apples ARM cores in mind. I mean AMD only has to pay for the licensing of the ISA itself, and the licensing from a single client to Nvidia is so small as to be pointless; if at any time it mattered, I'd think we'd be more worried about Nvidia's coming insolvency. But AMD would be a reputable and trusted brand at this point selling those ARM cores on already supported x86 sockets which would probably be a huge boon itself in just getting them sold in the first place. The resulting profit and marketshare that'd come from being able to meet that demand would far outstrip what they'd pay Nvidia IMHO. I mean yeah go nuts on developing a RISC-V solution, or heavens some OpenPOWER solutions as well but it'd be kind of dumb to completely ignore a market because of a few million dollars in licensing to Nvidia. EDIT: Just as a thought, but would there be an advantage for a 2P server running x86 in one socket and ARM in another? Because in theory only AMD seems positioned to potentially do that.
|
|
#
?
Oct 10, 2020 10:42
|
|
|
Cygni posted:x86 has a ton of momentum in the server world, no doubt, but I'm old enough to remember when the idea of x86 consumer parts muscling into big iron was itself considered a far off pipe dream. When people talk about x86 having too much momentum to be overtaken, I think of all those poor early codemonkeys that had to translate untold amounts of poo poo like bank software written for Alphas, SPARCs, Power, and Intel i860 (ironically, all RISC parts!), or even earlier stuff like 68000 CPUs. at this point probably like 90% of the server world is various flavours of open source, and getting that to run on arm is just downloading the arm iso nearly 100% of the time. there's some edge cases here and there, but since it's open source if it's critical for your company, they can fix it on the cheap unless you work with lovely vendors. if anything, servers are gonna be on arm before desktops, seeing how windows on arm is a joke, and windows still holds desktops for now. otoh, on a low level, modern x86 is just risc chips with a bunch of microcode that directs x86 instructions into the correct pipelines on said chips, so the architecture isn't vastly different to arm, they're just bigger. in fact, zen cores running at 2ghz are easily within ARM's power usage ballpark, and that was before the zen3 IPC increase at no increased power which just brings it ever closer. the big difference is, arm generally stops at the ~2ghz sweet spot for power consumption and is a much smaller and generic chip so it's not as good at some specialized tasks, while x86 doesn't care about boosting past 3ghz if it has to and has various potentially very power hungry but extremely useful extensions out the wazoo, poo poo like SSE version 4000 or whatever, and various AVX extensions, that do cost money to have on your chip but with modern power management schemes do not need any extra power unless actually required. i mean, poo poo, zen2 has all the cores completely gated off unless they're doing work, so even all the extra cores are a freebie power-wise until they're needed. i doubt we're gonna see an arm exodus any time soon due to these factors. arm servers have been a thing for like a decade now i wanna say, and while they excel at a few things and are definitely sometimes used over x86, x86 has stayed the leader for general purpose computing for a very good reason until now imo
|
|
#
?
Oct 10, 2020 11:37
|
|
|
If I wanted to upgrade from a 2700X to a new Zen 3 CPU, who would at that point buy a used Zen+ CPU? Just trying to figure out if there would be any kind of market for it at all.
|
|
#
?
Oct 10, 2020 12:15
|
|
|
GunnerJ posted:If I wanted to upgrade from a 2700X to a new Zen 3 CPU, who would at that point buy a used Zen+ CPU? Just trying to figure out if there would be any kind of market for it at all. A 3600 equals a 2700X essentially in multicore and beats it single core, TBH I don't see much resale value for anything pre Zen2 at this point. Like the leaps between each iteration on the same socket kind of make the older processors questionable, I think maybe the use case is for someone trapped on the 300 series boards who refuses to budge on buying a newer, better motherboard? Going forward it'll be Zen2 for budget and Zen3 for performance.
|
|
#
?
Oct 10, 2020 12:55
|
|
|
I'm on a 2400G right now and I'd think very hard about getting a 2700X if the price was right (i.e. less than a 3600) not saying I'm gonna buy from you, specifically, but just as a mindset
|
|
#
?
Oct 10, 2020 13:15
|
|
|
CaptainSarcastic posted:I'm not well-versed on the matter, but I think of ARM processors as being in mobile devices and such. Aren't pretty much all cellphones ARM? I know people have been talking about laptops and servers as possible ARM markets, but it already dominates (if not completely owns) the mobile device market. It seems like that would be where the money is. A majority of companies are primarily renting server CPU time rather than buying servers now, and the biggest vendor (Amazon AWS) gives you about a 10% discount for using AMD vs Intel, and about a 20% discount for using their own ARM CPUs. Of course, it's not a fair comparison because they charge you per thread, and the AMD & Intel stuff is 2-way SMT, while the ARM stuff doesn't have SMT at all. Meaning that their ARM stuff is an even better deal than it appears, because 4 of their big ARM cores do pretty drat good vs 2C/4T x86. This sticks AMD in a tough position though, because if you need the fastest per-thread performance, Intel still has that, and if you're cost sensitive you'll go with ARM. The middle option isn't a clear common winner.
|
|
#
?
Oct 10, 2020 14:02
|
|
|
SCheeseman posted:Buying whole parts is different from investing resources developing upon an architecture that is owned by a direct competitor. Apple could switch suppliers in a single generation, licensing and creating an ARM processor requires far more effort, time and money. They're more likely to move towards RISCV, as a lot of the industry is probably doing with Nvidia absorbing ARM. You realize you're talking about AMD, right? The company that started out cloning Fairchild, National Semiconductor, and Intel chips.
|
|
#
?
Oct 10, 2020 14:42
|
|
|
mdxi posted:ARM has always aimed at "really good compute in a small power envelope", which gives them incredible perf/watt but places a cap on absolute performance per core that's a lot lower than x86 chips.
|
|
#
?
Oct 10, 2020 14:48
|
|
|
Cojawfee posted:You realize you're talking about AMD, right? The company that started out cloning Fairchild, National Semiconductor, and Intel chips. Another era. If anything I'd assume having their flagship product inextricably tied to another company's proprietary architecture over the years has been a source of enough legal friction for them to want avoid a repeat of anything like it.
|
|
#
?
Oct 10, 2020 15:05
|
|
|
GunnerJ posted:If I wanted to upgrade from a 2700X to a new Zen 3 CPU, who would at that point buy a used Zen+ CPU? Just trying to figure out if there would be any kind of market for it at all. People on a tight budget. The resale won't be great but depending on how many people upgrade from Zen2 / what stores are unloading them for after Zen3 releases I could see them going for $75-100. Combine it with an RX 570 and motherboard of similar price and baby you got a
|
|
#
?
Oct 10, 2020 15:33
|
|
|
SCheeseman posted:Another era. If anything I'd assume having their flagship product inextricably tied to another company's proprietary architecture over the years has been a source of enough legal friction for them to want avoid a repeat of anything like it. Iirc they hold the right to x86_64, so if Intel sued them for x86, they could counter sue for x86_64
|
|
#
?
Oct 10, 2020 18:28
|
|
|
Got my MSI X570 MAG Tomahawk Wifi mobo this morning, thanks again for the recommendations. Gonna rebuild this afternoon.
|
|
#
?
Oct 10, 2020 18:51
|
|
|
ARM as a licensable ISA-only business is dead, since the ISA itself is meaningless at the high-perf end beyond the software support/cost Qualcomm, Samsung, etc now use cores designed by ARM since it's way cheaper and nobody really cares now that ARM 1st party core design can catch up to Apple (Neoverse cores are within striking distance) They both dumped their 100% custom cpu designs in favor of tweaking as needed since it doesn't matter anymore nVidia otoh needs a real CPU biz to combo with its AI + GPU + NIC RISC-V, now there's a reason to switch, will get a bunch. of $ dumped into it by China to avoid US restrictions
|
|
#
?
Oct 10, 2020 19:01
|
|
|
EmpyreanFlux posted:EDIT: Just as a thought, but would there be an advantage for a 2P server running x86 in one socket and ARM in another? Because in theory only AMD seems positioned to potentially do that. This sounds kinda cool, but also like an OS design nightmare. You'd either need a kernel (and driver stack) that simultaneously supports multiple architectures, or a way to prevent memory/PCIe/storage/network device contention between two independently running kernels. I'm honestly not sure which would be the harder problem to solve. Seems way easier to just have some x86 nodes and some ARM nodes in your fleet. I have this at home, and manage them all with the same homegrown+Ansible stack.
|
|
#
?
Oct 10, 2020 19:37
|
|
|
GunnerJ posted:If I wanted to upgrade from a 2700X to a new Zen 3 CPU, who would at that point buy a used Zen+ CPU? Just trying to figure out if there would be any kind of market for it at all. The beauty of AM4's longevity is that there is a large market of people who can use those CPUs. It will sell on eBay.
|
|
#
?
Oct 10, 2020 22:47
|
|
|
Looks like a 2700X still gets around $180 used on eBay, not bad at all.
|
|
#
?
Oct 11, 2020 02:21
|
|
|
how is the Windows x86 on ARM emulation working out anyway?
|
|
#
?
Oct 11, 2020 02:27
|
|
|
MaxxBot posted:Looks like a 2700X still gets around $180 used on eBay, not bad at all. yeah, if anything Ryzen Fever has had earlier stuff like the 1700 trending UP in price lately. Not many screaming deals out there anymore!
|
|
#
?
Oct 11, 2020 02:31
|
|
|
shrike82 posted:how is the Windows x86 on ARM emulation working out anyway? like on the "surface book whatever" type stuff that microsoft has been working on on-and-off for a couple years? my understanding is that it's decent for native stuff (UWP), and that the emulation is not quite as disastrous as you might assume (because the kernel itself is running native so the userland is the only thing running in lovely mode) but that it's clearly not a "premium" experience or really anything that's going to fly outside the entry-level tier of laptops where you might otherwise be getting an Atom or whatever. ironically this probably does benefit stuff like java apps and dotnet where all you have to do is port the interpreter/runtime and since the applications are distributed as bytecode you get the application ported "for free".
|
|
#
?
Oct 11, 2020 02:33
|
|
|
Yeah, it's basically user mode dynamic translation with calls into system DLLs and the kernel being trapped into the native equivalent. I'm not sure what exactly the process is from a real low level perspective -- perhaps the virtualization chapters in Windows Internals 7th Edition Part 2 will cover it -- but from what I've heard it functions somewhat like QEMU's user mode emulation system.
|
|
#
?
Oct 11, 2020 02:44
|
|
|
Asrock doing a lot of partnerships these days. First NZXT, now Razer. https://twitter.com/elpewpew/status/1314970528680235010 Honestly anything that starts narrowing down all of the incompatible RGB software, the better.
|
|
#
?
Oct 11, 2020 03:21
|
|
|
Microsoft needs to step in and make a directRGB api goddamn
|
|
#
?
Oct 11, 2020 03:25
|
|
|
seriously, also the RGB software is all trash
|
|
#
?
Oct 11, 2020 03:32
|
|
|
I want ASRock Rack to make GPU cards with the blandest aesthetics possible. Anything that’s not covered by a functional looking cooling solution is just a plain green PCB and be called “ASRock Rack 3D Accelerator - NVidia GA102 chipset - 24GB” and since there’s supply constraints other than getting their own sales guys getting CDW, Techdata, Ingram, etcetera’s sales guys to make IT and retailers purchasing aware of their existence, they can spend zero dollars promoting them.
|
|
#
?
Oct 11, 2020 04:24
|
|
|
That would nice, like those after market heat sinks without a plastic shroud, the branding should just be silk screened on the pcb.
|
|
#
?
Oct 11, 2020 04:29
|
|
|
Cygni posted:Honestly anything that starts narrowing down all of the incompatible RGB software, the better. If you didn't know, keeping everything incompatible is how the companies with $30 ARGB fans make money. WS2812B LEDs are not that expensive, if everything was standardized the ARGB fans would cost about $3 more than a fan with no lights at all. Basically, all the major Brands want to screw you with incompatibility to make sure that a) you're afraid to buy the much cheaper chinesium argb fan, and b) so that when you pick Brand's argb kit and spend $150 on a bunch of crap, you have an investment into buying more from Brand the next time. RGB poo poo is the only place they can get away with this, because it doesn't matter. Everything else is enforced standards. Asrock joining in with razer isn't a step toward standardization, it's asrock being the low bidder because they don't have an RGB brand.
|
|
#
?
Oct 11, 2020 04:50
|
|
|
MeruFM posted:seriously, also the RGB Fixed.
|
|
#
?
Oct 11, 2020 04:53
|
|
|
If you're currently rocking a RGB component you bought because it has RGB I want to invite you to go gently caress yourself.
|
|
#
?
Oct 11, 2020 05:17
|
|
|
AARP LARPer posted:If you're currently rocking a RGB component you bought because it has RGB I want to invite you to go gently caress yourself. great evidence for why RGB is good
|
|
#
?
Oct 11, 2020 05:20
|
|
|
Wow the LED light strips I installed in my computer apparently make people really loving angry.
|
|
#
?
Oct 11, 2020 05:23
|
|
|
It's almost December why y'all getting het up over lights. Get into the spirit of Christmas
|
|
#
?
Oct 11, 2020 05:48
|
|
|
gradenko_2000 posted:It's almost December why y'all getting het up over lights. Get into the spirit of Christmas "Coal" is actually a pretty good description of my computer's esthetic, actually.
|
|
#
?
Oct 11, 2020 06:10
|
|
|
I think I'm gonna hop off the Threadripper train, come the 5950X. I don't really need 24 cores or more, benchmarks differences between the 3950X and 3960X are pretty much within error margin for standard apps, and I'll finally be able to use an AIO. I ditched my NAS, so I don't need additional 8 PCIe lanes for the Mellanox card, altho it may suck not to be able to run more than one NVMe drive directly hooked up to the CPU. Not that I'm doing that right now. --late edit: Also, way less idle power draw. The thing draws like 140W at idle with the displays off (there's also a Pi4 and a DSL modem on my UPS, but their draw is a fart in the wind). Combat Pretzel fucked around with this message at 11:25 on Oct 11, 2020 |
|
#
?
Oct 11, 2020 10:30
|
|
|
Why are XSX and PS5 running RDNA 2 but only Zen 2?
|
|
#
?
Oct 11, 2020 11:36
|
|
|
I don't think there is a way to say for sure why but the most likely reason I can think of is timing. If MS or Sony wanted either of their consoles out in volume (read: low to high 100's of thousands of units ready for sale) for Christmas 2020 then Zen2 was probably the best they could do. AMD squeezing out some relatively low volume parts for DIY'ers in the same time frame doesn't mean that MS or Sony would've been able to launch their consoles in volume within the same time frame
|
|
#
?
Oct 11, 2020 11:41
|
|
|
Josh Lyman posted:Why are XSX and PS5 running RDNA 2 but only Zen 2? MS and Sony have been stockpiling chips for their units since this summer, and their SoCs would have been design complete and gone into tapeout well before then. Zen 3 in likelihood had their specs finalized only in the past month and just entered mass manufacturing.
|
|
#
?
Oct 11, 2020 15:00
|
|
|
MikeC posted:MS and Sony have been stockpiling chips for their units since this summer, and their SoCs would have been design complete and gone into tapeout well before then. Zen 3 in likelihood had their specs finalized only in the past month and just entered mass manufacturing.
|
|
#
?
Oct 11, 2020 15:46
|
|
|
Josh Lyman posted:Couldn’t the same be said for RDNA 2? RDNA 2 discrete GPU is not the same RDNA 2 on PS5 or XBX. Both consoles have custom versions of RDNA 2 that share baseline features (like RT hardware bolted onto TMUs) but the SoCs have been locked down for months and need to ship in the quantity of millions. Meanwhile, RDNA 2 for PC doesn't need to launch nearly as many units. But AMD almost certainly has chips stockpiled themselves while finalizing core clocks speeds and memory. Coreteks tweeted that Navi 21 was being run through its paces in batches in mid-September. This might be why they are confident they will avoid an Ampere paper launch type situation.
|
|
#
?
Oct 11, 2020 16:04
|
|
|
|
| # ? May 30, 2024 10:09 |
|
|
XSX and PS5 are both using APU's not discrete GPU's so you can't use them to judge when discrete RDNA2 dies would be ready and mass produced.
|
|
#
?
Oct 11, 2020 16:09
|
|