|
BobHoward posted:yeah, in fact they're doing vastly better than intel at cpu cores designed for phone power profiles. intel completely missed out on that market, the stuff they tried to sell was late and kinda half-assed and eventually they gave up when even heroic marketing techniques (essentially giving the chips away) failed to make a dent. clock-for-clock and watt-for-watt they still aren't quite on par with intel, but when you can get 8c/16t for under $300 that doesn't really matter e: just don't get a used early-stepping chip for development work since there was some kind of hardware bug which caused a segfault under heavy compilation workloads. The_Franz fucked around with this message at 21:45 on Nov 12, 2017 |
# ? Nov 12, 2017 21:42 |
|
|
# ? Apr 26, 2024 04:03 |
|
yeah, amd's new core is a bit better than haswell, both on the factory clocks and ipc sides, but you can get double the cores for similar cash, so it's been selling like hotcakes. these desktop intels that overclock to 5ghz are still quite a bit better in single threaded performance, obviously, but in the xeon market the difference is extremely meh. intel's trying some poo poo, but it appears they simply can't match the high core count prices, and as a server operator, i rather have twice the cores running at ~2ghz than half running at ~4 even before price is considered. the bugs in kvm/qemu have been fixed now too so
|
# ? Nov 12, 2017 21:45 |
|
yeah i was looking at the amd epyc cores for some hpc stuff and supposedly they're pretty decent. might contact supermicro and see if i can get some price quotes and such
|
# ? Nov 12, 2017 21:56 |
|
Best Bi Geek Squid posted:year of linux on the ppc laptop: https://www.powerpc-notebook.org/en/ because the ThinkPad 820/850/860 is rare enough that one of these might actually be cheaper? only reason I can think of…
|
# ? Nov 12, 2017 22:03 |
|
Notorious b.s.d. posted:i think they failed less because of the price, and more because they are a fly-by-night with no product a little bit of both really, of the people actually willing to lay out $4k, very few were willing to give a fly-by-night that cash they’d have done better using a crowdfunding platform where the people laying out the money could be (more) reasonably sure it wasn’t immediately going straight into a stranger’s hookers-and-blow fund, and instead going into that after the project was “fully” funded
|
# ? Nov 12, 2017 22:08 |
|
oh ok i didn't know that open arm stuff was poo poo lmao
|
# ? Nov 12, 2017 22:10 |
|
Best Bi Geek Squid posted:year of linux on the ppc laptop: https://www.powerpc-notebook.org/en/ looked at this in more detail, they appear to be clowns “please send us $$$ via our web site to make this happen” versus “give us $$$ via this crowdfunding platform and when this project is done, you’ll receive a laptop” I mean, this isn’t loving rocket science, and isn’t there even a GNU/crowdfunding thing they can use if they’re obsessing over ideological purity? (my GPD Pocket, funded via Indiegogo, continues to be sweet despite running Ubuntu)
|
# ? Nov 12, 2017 22:11 |
|
i participated in the crowdfunding for that purism open sores smartphone like the complete idiot that i am maybe i'll even get a phone for my $600 in a few years, who knows
|
# ? Nov 12, 2017 22:13 |
|
i'd very much like to have a phone that's as easy to tinker with as a rpi.
|
# ? Nov 12, 2017 22:20 |
|
Sapozhnik posted:i participated in the crowdfunding for that purism open sores smartphone like the complete idiot that i am was their funding target at least realistic? the PowerPC laptop thing is asking for all of €12,600, presumably just to do the design/layout or something, and everything is purely a donation towards that end GPD Pocket, on the other hand, had a funding goal of about US$250,000—and that only realistic because it was from a company located right in Shenzhen with experience producing similar hardware already, and they were obviously going to blow past it if the PowerPC laptop people used a real crowdfunding platform, had a funding goal of $1,000,000 ($850,000 after fees), had a relationship with a contract manufacturer lined up, and $1,000 would get a contributor a laptop at the end, I bet they would actually get enough money that any failure on their part would be hard to attribute to funding
|
# ? Nov 12, 2017 22:26 |
|
Truga posted:yeah, amd's new core is a bit better than haswell, both on the factory clocks and ipc sides, but you can get double the cores for similar cash, so it's been selling like hotcakes. these desktop intels that overclock to 5ghz are still quite a bit better in single threaded performance, obviously, but in the xeon market the difference is extremely meh. intel chose to roll out their new pineapple-in-your-butt pricing the same month that AMD launched their new server chips that cost 1/2 as much as the "bronze" tier of the new gently caress-you scheme it is a lolocaust
|
# ? Nov 12, 2017 22:34 |
|
intel will now charge server customers separately for:
this is exactly as complicated as it sounds picking a 2017 intel sku is the worst purchasing experience i've ever had
|
# ? Nov 12, 2017 22:36 |
|
do intel's single thread performance numbers take into account the thermal throttling if you try to max out all cores simultaneously?
|
# ? Nov 12, 2017 22:37 |
|
no, those numbers are mostly for desktop workloads like CAD or games or whatever, those workloads shouldn't throttle unless you have a really lovely case with no airflow even at extreme frequencies. but an 8 core xeon in a half-decent rack chassis will happily run 16 threads forever at factory clocks as long as you don't throw any avx at it, and i don't think amd is any better at that (avx512 isn't even a thing in current zen cpus iirc)
|
# ? Nov 12, 2017 22:44 |
|
pseudorandom name posted:do intel's single thread performance numbers take into account the thermal throttling if you try to max out all cores simultaneously? that's configurable on the server normally when intel publishes numbers for single-thread performance they will tell you whether it's "turbo" (throttling enabled) or just the normal performance (cores maxed out, or throttling disabled)
|
# ? Nov 12, 2017 22:50 |
|
i found this anandtech review of epyc vs skylake: https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/3
|
# ? Nov 12, 2017 22:51 |
|
also, apparently when you use certain intel avx instructions the chip throttles itself to a lower ghz to prevent overheat, slowing down everything on the chip edit: article https://blog.cloudflare.com/on-the-dangers-of-intels-frequency-scaling/
|
# ? Nov 12, 2017 22:53 |
|
eschaton posted:looked at this in more detail, they appear to be clowns you're underselling the clowniness: - they've taken €7100 to produce the block diagram, but they're holding the identity of the chips in the block diagram hostage until they get more money. want a powerpc laptop someday? give us some money, no we won't tell you what might be in it, if it ever happens at all - their next funding goal is just to draw a schematic. not to actually make anything physical, just a schematic. - mechanical engineering? lol what's that, clearly if you can make a motherboard that's the same as a whole laptop right? - this throw-money-at-us website isn't doing anything (not even the block diagram) themselves, they're contracting it out to acube systems. or maybe the funding site is just a front for acube, idk. now, acube systems isn't an entirely unknown quantity. they are a 'company' that has an actual history of 'successfully' designing, manufacturing, and shipping powerpc motherboards. they haven't done anything recently but they have actually shipped 'real' products. but. what market does acube serve, you might ask, and why did i use all those scarequotes? it's because acube serves the amiga 'market'. it is almost certain that acube has never made a real profit unless they found some non-amiga customers for their previous boards. also, modern amigans are more than willing to put up with janky broken poo poo, which is a good thing because companies like acube tend not to have the engineering talent and resources to make things which aren't janky and broken. so. this powerpc laptop project is a pseudo crowdfunding front end to siphon money to amiga fanatics who have long lusted after the holy grail, a powerpc laptop that will show the rest of the world they should be running amigaos too. it has to be powerpc because the clownshoes software company that managed to land itself in pseudo-control of amigaos in the post-commodore era sabotaged attempts to move amiga users to x86 a long time ago. why did they do that? because they're the kind of stuck-in-the-1990s nerd who is ideologically opposed to x86. their customer base isn't any different, it's all people who would die of shame if they ported amigaos to standard x86 hardware and it was better there. they're still clinging to the amiga tribal identity of ~custom chips~ which make amigas better than pcs. it doesn't seem to matter to them if their modern hardware is objectively junk, what they want is something different enough that they can invent a story in their heads which makes it ineffably better. no i am not making this up that's really the way these people think. so, that's why powerpc
|
# ? Nov 13, 2017 00:19 |
|
eschaton posted:was their funding target at least realistic? their funding target is $1.5m. to design something that could in theory be considered a smartphone so lol no they did end up raising about $2.2m though
|
# ? Nov 13, 2017 00:36 |
|
if anyone wants my ibook g4 with a hosed memory controller cost of shipping is fine
|
# ? Nov 13, 2017 00:37 |
|
Cocoa Crispies posted:if anyone wants my ibook g4 with a hosed memory controller cost of shipping is fine P-p-p-p-amigabook
|
# ? Nov 13, 2017 00:41 |
|
Sapozhnik posted:their funding target is $1.5m. to design something that could in theory be considered a smartphone yeah so the company they are working with, acube, has built all their products on the applied micro powerpc 400 series -- an embedded SoC that tops out at 1.2 GHz it's still probably faster than your average smartphone, because it uses "real" ddr2 instead of some garbage low power poo poo... but it ain't gonna give your x86 laptop a run for its money.
|
# ? Nov 13, 2017 00:55 |
|
Condiv posted:i found this anandtech review of epyc vs skylake: https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/3 too long, didn't read it barely matters whether skylake-ep is good or not, because the pricing is such a loving mess that customers will run away. some will run to amd, others will just make huge buys of broadwell-ep while they can still get the parts.
|
# ? Nov 13, 2017 00:57 |
|
Notorious b.s.d. posted:yeah so the company they are working with, acube, has built all their products on the applied micro powerpc 400 series -- an embedded SoC that tops out at 1.2 GHz i was referring to the purism smartphone
|
# ? Nov 13, 2017 01:37 |
|
BobHoward posted:you're underselling the clowniness: lmao outstanding
|
# ? Nov 13, 2017 02:48 |
|
Truga posted:i'd very much like to have a phone that's as easy to tinker with as a rpi. have u heard of Google Project Ara
|
# ? Nov 13, 2017 03:05 |
|
atomicthumbs posted:have u heard of Google Project Ara quote:Google ATAP requires their employees to sign a 2-year contract. Limiting their time at Google allows employees to take risks and move at a faster pace than other employees. Paul Eremenko's contract expired after the 2nd Developers Conference in 2015, along with many other Google ATAP employees. Rafa Camargo, lead of the original Motorola Droid, has now taken over as project manager for Project Ara. lol let's try to build something that will require a huge amount of new technology to ship anything and it might not even be a viable product because nobody but a handful of nerds cares about swapping modules in and out of their smartphone AND while we're taking on this huge engineering and manufacturing challenge let's also shoot ourselves in the foot by only letting ourselves have people for 2 years max (in reading some articles about ara's cancellation this whole 2 year contract thing was because motorola's atap group was ~inspired~ in some way by how darpa works) (lol darpa) (lol motorola) (lol google)
|
# ? Nov 13, 2017 03:54 |
|
BobHoward posted:(in reading some articles about ara's cancellation this whole 2 year contract thing was because motorola's atap group was ~inspired~ in some way by how darpa works) darpa limits managers to five years, not two, so something like the wizard grand challenge (that took about that long from proposal to execution) can actually happen and it's contractors doing all the work anyways
|
# ? Nov 13, 2017 06:18 |
|
been researching server hardware all day and wow, the amd epyc stuff is way better than it should be considering the price: https://www.spec.org/cpu2006/results/res2017q2/cpu2006-20170529-47127.html - AMD EPYC 7601 CINT https://www.spec.org/cpu2006/results/res2017q2/cpu2006-20170529-47126.html - AMD EPYC 7601 CFP vs https://www.spec.org/cpu2006/results/res2017q3/cpu2006-20170807-48055.html - Xeon Platinum 8176 CINT https://www.spec.org/cpu2006/results/res2017q3/cpu2006-20170807-48003.html - Platinum 8176 CFP the platinum 8176 kinda holds its own in the integer benches, but in floating points it's beaten on all but 3 benches. meanwhile, the cost of the epyc is wayyyyyy lower than the platinum (i estimate the system i want to build is around $40k with the epyc and 4x tesla p100s, it's $64k for the same with the platinum 8176)
|
# ? Nov 13, 2017 23:25 |
|
be warned,spec rate scores are kinda bullshit as a basis for making decisions about many loads. rate works by running N copies of the single threaded spec benchmarks (N chosen by whoever’s running them) so there is literally zero ipc going on, it’s a textbook embarrassingly parallel load. sometimes this tells you what you need to know, sometimes it really really does not. (single threaded spec has its issues too, a couple of the benchmarks like libquantum have been thoroughly broken by compiler autopar and that’s legal to do under the “peak” rules so they’re not really single threaded results)
|
# ? Nov 13, 2017 23:46 |
|
that's more akin to what we're using our current hardware for though, so it's a decent measure in our case. also, even if the epyc performed a little worse on average, i think it'd be our choice considering how ridiculously expensive the platinum 8176 is.
|
# ? Nov 13, 2017 23:56 |
|
Condiv posted:been researching server hardware all day and wow, the amd epyc stuff is way better than it should be considering the price: the real joke is that intel has hosed up their pricing that badly the amd epyc stuff is priced correctly, but intel is gouging us with their latest poo poo
|
# ? Nov 14, 2017 00:00 |
|
BobHoward posted:be warned,spec rate scores are kinda bullshit as a basis for making decisions about many loads. rate works by running N copies of the single threaded spec benchmarks (N chosen by whoever’s running them) so there is literally zero ipc going on, it’s a textbook embarrassingly parallel load. sometimes this tells you what you need to know, sometimes it really really does not. i am sometimes fond of the specjbb benchmark one the one hand, it's kinda dumb. on the other hand, it looks a lot more like real applications than specint or what have you
|
# ? Nov 14, 2017 00:01 |
|
Intel had the market cornered for years, now AMD finally has a decent competitor, so hopefully prices will come down a bit.
|
# ? Nov 14, 2017 00:05 |
|
looking at that pricing I’m wondering if intel is having serious yield issues
|
# ? Nov 14, 2017 00:11 |
|
so i decided to get btrfs on my laptop's partition with kernel 4.14 and zstandard compression. so far it's p nice, and compresses better than lzo and faster than zlib, so it seems like it might work well. thinking about setting the fileserver at work up with this plus btrfs raid 10 with a scrub as a cronjob and hopefully it will work p well. i just need to test quotas and make sure it works well
|
# ? Nov 14, 2017 02:21 |
|
hobbesmaster posted:looking at that pricing I’m wondering if intel is having serious yield issues nah they’ve been shipping 14nm products forever, it is the definition of a mature process. 10nm is where they’ve been having problems, which they’ve been dealing with by using 14 longer than they would have for new product generations. xeon gold/silver/etc is purely an attempt at monopolistic profit taking. slice your customer base up as finely as possible, extract as much as you can from each niche by using fuses to disable all the good stuff for customers who can’t afford to pay as much. while this is always done in the chip industry to some extent and can even be beneficial to customers, intel takes it to the next level
|
# ? Nov 14, 2017 02:30 |
|
BobHoward posted:nah they’ve been shipping 14nm products forever, it is the definition of a mature process. 10nm is where they’ve been having problems, which they’ve been dealing with by using 14 longer than they would have for new product generations. yeah, seeing how bronze had hyperthreading disabled blew my mind. that's been a standard on intel chips forever right? even celerons had it right?
|
# ? Nov 14, 2017 03:04 |
|
Condiv posted:yeah, seeing how bronze had hyperthreading disabled blew my mind. that's been a standard on intel chips forever right? even celerons had it right? no they’ve been doing market segmentation for a while with it. see the i5s and i7s that only differed by the i7 having hyperthreading turned on
|
# ? Nov 14, 2017 05:48 |
|
|
# ? Apr 26, 2024 04:03 |
|
Condiv posted:so i decided to get btrfs on my laptop's partition with kernel 4.14 and zstandard compression. so far it's p nice, and compresses better than lzo and faster than zlib, so it seems like it might work well. i guess this is the right time to mention my pet peeve with btrfs remember that when taking backups you need to take note of what folders/subvolumes that has the nocow file attribute set, because btrfs send/receive does not restore that: https://btrfs.wiki.kernel.org/index.php/Design_notes_on_Send/Receive#Send_stream_v2_draft nice script for uncowing (mentioned in the faq) after restoring from backup: https://github.com/stsquad/scripts/blob/master/uncow.py
|
# ? Nov 14, 2017 13:16 |