|
Cygni posted:yeah 8mb of vram is enough for anyone Oops, sorry!
|
#
?
Jul 3, 2018 02:07
|
|
|
|
| # ? May 17, 2024 21:01 |
|
|
Gamers Nexus presents: Everything You Wanted To Know About Memory Timing But Were Too Afraid To Ask. Just another reason why they're one of the best.
|
|
#
?
Jul 3, 2018 02:24
|
|
|
Craptacular! posted:Honestly since going from Intel graphics to AMD graphics on my HTPC there are moments on the desktop where the space around the arrow seems to briefly go corrupt in a flash and then correct itself. This is just sitting on the desktop with no windows open. Given it's purpose the only fullscreen app that machine ever runs is Kodi. Yikes gently caress AMD that's horrible. My condolences.
|
|
#
?
Jul 3, 2018 13:13
|
|
|
My HTPC with a rx570 experiences no such corruption ever  Maybe Vega drivers < Polaris drivers
|
|
#
?
Jul 3, 2018 14:52
|
|
|
ufarn posted:Nvidia being bad doesn't make AMD good, and you can't tell me I would have had a worse experience with an Nvidia GPU than my 270X. (I would probably have flipped such a GPU during the mining boom, but that's another story.) How about the time you couldn't update drivers on 560Tis for a year because it caused constant display driver crashes?
|
|
#
?
Jul 3, 2018 15:48
|
|
|
Arzachel posted:How about the time you couldn't update drivers on 560Tis for a year because it caused constant display driver crashes?
|
|
#
?
Jul 3, 2018 16:07
|
|
|
Risky Bisquick posted:My HTPC with a rx570 experiences no such corruption ever It's also possible that the APU-specific driver was/is bad. Don't know if that will continue now that APU's use the same installer as RX add-in cards instead of being a separate download. I'll admit I haven't seen it happen lately.
|
|
#
?
Jul 3, 2018 19:20
|
|
|
Video drivers are horrible hack jobs and I'm constantly amazed that anything runs. There's a very good reason both AMD and Nvid have people that swear one has superior drivers. In my opinion they have trade offs. If they don't bother you pick the one that does best in benchmarks, but benchmarks are not everything in GPUs due to feature sets and the fact that you will sometimes have AMD's driver be really good in one game, but then 3 months later Nvid has optimized it and now they are better. The opposite happens too, and sometimes they preform well from the get go, and other times someone loses performance. But this is the CPU thread, It's far easier to look at Intel vs AMD and go multi thread and single threaded for comparing 2 chips. You can't go FPS and RPG or any kind of broad category like that (maybe if you could group them based on source, unreal and unity or something, but almost everyone is going to play a mix of all 3 engines, and very possibly have no idea what engine each game uses).
|
|
#
?
Jul 3, 2018 19:35
|
|
|
Craptacular! posted:It's also possible that the APU-specific driver was/is bad. Don't know if that will continue now that APU's use the same installer as RX add-in cards instead of being a separate download. I'll admit I haven't seen it happen lately. I thought Raven Ridge was using a separate driver package that released quarterly?
|
|
#
?
Jul 3, 2018 19:40
|
|
|
Paul MaudDib posted:I thought Raven Ridge was using a separate driver package that released quarterly? Ah, is that what it was? Okay then. I don't know how to support the hardware I've purchased because they're so consumer-unfriendly to people who don't want to do homework on driver downloads like it's still two decades ago. It doesn't help that marketing keeps stepping in with other phrases like "ReLive" and "Adrenalin" to try and make all these driver versions distinct. All that said, I'm just glad I'm not running Obscure Microsoft Update Database Driver anymore. Craptacular! fucked around with this message at 19:56 on Jul 3, 2018 |
|
#
?
Jul 3, 2018 19:52
|
|
|
pixaal posted:I get the messed up cursor occasionally, it's happened in both 7 and 10 over several motherboards and graphics cards (all happen to have been AMD because I didn't like how Nvid handled 3+ monitors). It only happens when there are multiple monitors and only happens on the primary monitor. The mouse cursor ends up looking like a stack of lines and is really odd looking. Holy poo poo, so that's what it is. I've wondered why this happens sometimes on my PC.
|
|
#
?
Jul 3, 2018 20:37
|
|
|
Can someone help me calculate something, please? I downclocked my Fury X by 10% at every power state, and undervolted it by 10% also. It seems fully stable at this. What estimated power is it likely to be drawing now? I know there's equations and poo poo.
|
|
#
?
Jul 4, 2018 19:32
|
|
GRINDCORE MEGGIDO posted:Can someone help me calculate something, please? I'll note that you should underclock the card to where you are comfortable first and then undervolt it as low as it will go while remaining stable, you can often undervolt a lot more than you would think on those cards, past a certain point every extra Mhz of speed uses more power than the previous and those cards were pushed well into the exponential power use curve. Anyway, if it's undervolted by 10% it should just use 10% less, though it's a bit more complicated than that and you would need testing equipment to really be able to tell so it's easier to just say 10% less and leave it at that. If you want specific wattages I remember it was something like 250W-350W at stock depending on the load, so 10% less than that should be a pretty safe bet.
|
|
|
#
?
Jul 4, 2018 19:47
|
|
|
AVeryLargeRadish posted:I'll note that you should underclock the card to where you are comfortable first and then undervolt it as low as it will go while remaining stable, you can often undervolt a lot more than you would think on those cards, past a certain point every extra Mhz of speed uses more power than the previous and those cards were pushed well into the exponential power use curve. Anyway, if it's undervolted by 10% it should just use 10% less, though it's a bit more complicated than that and you would need testing equipment to really be able to tell so it's easier to just say 10% less and leave it at that. If you want specific wattages I remember it was something like 250W-350W at stock depending on the load, so 10% less than that should be a pretty safe bet. Thanks. I'm on my phone, so I CBA to find the equation calculating power use taking clock speed and voltages into account. I know about them being clocked to the edge, hence undervolting it. GRINDCORE MEGGIDO fucked around with this message at 20:05 on Jul 4, 2018 |
|
#
?
Jul 4, 2018 19:50
|
|
|
GRINDCORE MEGGIDO posted:Can someone help me calculate something, please?  or https://www.hwinfo.com/
|
|
#
?
Jul 4, 2018 20:08
|
|
|
Oh, Hwinfo is accurate? It's indicating about 211W when loaded at 100%. Cheers.
|
|
#
?
Jul 4, 2018 20:23
|
|
|
GRINDCORE MEGGIDO posted:Oh, Hwinfo is accurate? It's indicating about 211W when loaded at 100%. Cheers. Compare it stock vs under volted. I used hwinfo when I had a  deep learning cluster
|
|
#
?
Jul 4, 2018 20:31
|
|
|
Got a new 2600X and I'm loving it so far, except for the fact that, by default, it literally never downclocks / lowers voltage even when idle. I take a look into the "Ryzen Balanced" power plan and see that the minimum processor state is set to 90% instead of the 5% of the regular "Balanced" plan. Changing this setting immediately causes the Ryzen to start dynamically clocking the same way my old overclocked 2500k did / any modern processor should. Except as I googled this problem I keep coming across some vaguely explained "core parking" boogeyman that is apparently avoided by running your CPU balls out 24/7 like a drat bitcoin rig. Has this core parking problem been fixed or is my processor from 2018 just never supposed to use it's power saving modes? edit - Further reading suggests Ryzen handles power states in a way that Windows doesn't show properly. Despite the clock speed not changing much visibily the power draw actually does vary wildly depending on load. FallenGod fucked around with this message at 03:57 on Jul 7, 2018 |
|
#
?
Jul 7, 2018 03:37
|
|
|
You don’t need the Ryzen Balanced power plan any longer, just set it to Balanced and it’ll be fine.
|
|
#
?
Jul 7, 2018 04:39
|
|
|
Thanks, I remembered reading something like that a while ago but when I installed the AMD chipset drivers it still defaults to their power plan. Figured it must have been needed.
|
|
#
?
Jul 7, 2018 04:52
|
|
|
Should I install AMD chipset drivers?
|
|
#
?
Jul 7, 2018 05:55
|
|
|
LRADIKAL posted:Should I install AMD chipset drivers? I do because the Windows update version sometimes hangs for me.
|
|
#
?
Jul 7, 2018 05:57
|
|
|
I'm betting I'm just slower than molasses on this one and everyone else got the joke first, but maybe AMD didn't pick EPYC as a server CPU name out of a hat? Based on their "zero fucks" marketing team, I feel like there is a strong possibility it's meant to invoke Intel's failed Itanium/EPIC (or Explicitly Parallel Instruction Computing). Because really nothing could be more "gently caress you Intel" than using Intel's dead baby as their server CPU brand name.
|
|
#
?
Jul 9, 2018 20:51
|
|
|
WTF would you ever want the name of your product to make someone think of Itanium?
|
|
#
?
Jul 9, 2018 21:04
|
|
|
To lord it over Intel that AMD owns the rights to the x86-64 ISA? I've known some goddamn petty people in the industry who would absolutely do something like that.
|
|
#
?
Jul 9, 2018 21:21
|
|
|
Petty? The last time AMD had a superior product, Intel literally tried to force AMD to bankruptcy through illegal backroom deals which kept AMD processors from gaining market share. AMD had to do rounds of mass layoffs and cut operating expenses to the bone as a direct result of Intel's attempt to sabotage them out of the market. AMD has a right to be cocky; they have again managed to develop a superior product, despite having only a fraction of the R&D budget to work with.
|
|
#
?
Jul 9, 2018 22:42
|
|
|
There could be something to it. Beyond the petty revenge thing, there also might be a fair bit of marketing benefit to it. Personally 'epyc' didn't sound all that nutty as a processor brand name to me, and some of that might be lingering memory of intel's branding. Kinda forgotten about it until now. It's kinda piggybacking on the long-dead recognition that intel spent a lot of money on. But EPIC was never an official trademark or anything.
|
|
#
?
Jul 9, 2018 23:00
|
|
|
I don't think anyone but Intel is really supposed to get it if there is anything to it, which is why it is a "gently caress you Intel". I seriously think there is something to it, because the Zen team has been on an absolute relentless pace with this poo poo. They may have good reason to as well - http://www.agner.org/optimize/microarchitecture.pdf I recently stumbled on this while looking for details on K10 but it also has information on Zen and Sky(Coffee)lake, and Zen looks like a winner since each core has more resources than an Intel counterpart, which is why Zen is better at multithreading than Skylake. It is also able to execute up to five instructions per clock cycle, compared to Intel's four. Also that paper proves IMHO that Phenom/K10 would have been better @ 28nm than Bulldozer ever could dream of. K10 had an enormous amount of potential throughput, but it could rarely be utilized because the despite having 9 execution units it only had retire 3 instructions at a time. Also the FP scheduler on it was all kinds of hosed, and it didn't have full OoO execution. It also still soundly beat any Construction core in IPC, and a mild redesign to fix the FPU and the EU:Retirement (as well as making all instructions the same latency), plus making the pipeline fully out of order might have actually kept AMD competitive. Or just look at the Jaguar entries, it's a design without any serious bottlenecks. Doubling it's resources, adding an L3 and focusing on higher power operation still seems like it'd have come out better than Bulldozer did.
|
|
#
?
Jul 10, 2018 02:59
|
|
|
AMD starting to sponsor Twitch streamers which is a rare sight to say the least: https://twitter.com/Dyrus/status/1016595724639731713 Both CPU *and* GPU. I assumed this would come with Zen 2, but it's nice to see happening already; it was either doing it on this side of Intel's 8-core CPU or after the Zen 2 launch. They might have helped a dude out a bit with the production value, though ...
|
|
#
?
Jul 10, 2018 17:00
|
|
|
lol, according to dyrus' twitch profile he had a 1080ti before really taking gaming to the next level with that rx580
|
|
#
?
Jul 10, 2018 17:07
|
|
|
The 24C Threadripper is apparently labeled the 2950X. If it's snatching the 16C TR1's successive label, I sure hope that it has the same launch price.
|
|
#
?
Jul 11, 2018 09:18
|
|
|
Combat Pretzel posted:The 24C Threadripper is apparently labeled the 2950X. If it's snatching the 16C TR1's successive label, I sure hope that it has the same launch price. You know you want one 
|
|
#
?
Jul 11, 2018 15:01
|
|
|
Hi. How you been doing? Been a while since there's been some really interesting AMD news. https://www.theregister.co.uk/2018/07/10/amd_china/ quote:Thanks to a licensing deal with AMD and a complex joint-venture arrangement, the Chinese chip producer Chengdu Haiguang IC Design Co. (Hygon) is now producing x86-based server processors that are largely indistinguishable from AMD's EPYC processors—so close in design that Linux kernel developers had to do little in the way of patching to support the new processor family, called "Dhyana." The server chips are being manufactured for domestic use only—part of an effort to break China's dependence on foreign technology companies. China-based THATIC and AMD formed CHMT, which is 51% owned by AMD. CHMT licenses Zen cores to the THATIC-owned Hygon (30% ownership stake by AMD) to design the peripherals, power control, and I/O around the Zen Cores, then passes the work back to CHMT. CHMT goes to pure play foundries to make not-AMD chips. I know I said I expected AMD to kill in the Chinese server market, I just didn't expect them do go about it like THIS.
|
|
#
?
Jul 12, 2018 06:07
|
|
|
.
sincx fucked around with this message at 05:50 on Mar 23, 2021 |
|
#
?
Jul 12, 2018 06:33
|
|
|
sincx posted:I wonder how much China paid AMD for the most modern x86 implementation in existence. $239 million, plus additional royalties per quarter (not sure if this is known yet, but in 2017 it was order-of 25-50 million per quarter). On one hand yeah, AMD just handed the crown jewels to the Chinese. On the other hand AMD has no marketshare in the Chinese market to lose (Intel owns the market almost entirely) and there's now an additional x86 player on the market.
|
|
#
?
Jul 12, 2018 06:59
|
|
|
AMD must either be a) very certain that the underlying technology is not trivially reverse-engineerable 2) licensing blackboxes with "power delivery goes here" and "PCIe controller goes here" and "memory controller goes here" stubs, or iii) none of the above and whistling past a graveyard.
|
|
#
?
Jul 12, 2018 07:34
|
|
|
My understanding is that AMD isn't giving it to the Chinese at all, rather it's two sets of puppet companies with modifications to the security design of the processor for Chinese specifications. It's AMD on either end, so at worst it seems to be Option 2, the Black Box. There is no way AMD is just going to let Zen get completely reverse engineered like that as it'd be within a year where China would cut AMD loose. I mean, supposedly smarter companies have been dumb and burned by China before, guess we'll see.
|
|
#
?
Jul 12, 2018 09:09
|
|
|
FaustianQ posted:There is no way AMD is just going to let Zen get completely reverse engineered like that as it'd be within a year where China would cut AMD loose. Alternate hypothesis: AMD let Zen get completely reverse engineered and China cut AMD loose within a year. Say, how long has it been since Epyc "launched"? Oh right, a year. Ask Micron what China thinks of IP agreements (after a week of denial, they confirmed the Chinese ruling yesterday, locking them out of the Chinese market). Especially after the US keeps isolating itself internationally, and provoking trade wars. Actions have consequences, and IP theft is one of the easy ones. There is no way AMD made this agreement without knowing that it was a scorched-earth deal. They don't get the Chinese market, but they do get unimpeded access to foundries, and Intel doesn't get the Chinese market either. AMD has ~0% marketshare in China, Intel has ~100% marketshare in China. Guess who loses when the inevitable happens. I mean, competition in the marketplace is good for consumers! Now we have a third x86 player. I wouldn't buy them if I didn't want my PC to become part of a botnet, but they're there, and maybe eventually the x86 thing will break apart. If not who cares - the whole "you get exclusive rights to IP for 30 years" thing is stupid in this era when products are obsolete in 5 years or less. I'd like to see cheap RAM before I'm 60 and I don't care if it's a Chinese brand that does it. Paul MaudDib fucked around with this message at 09:39 on Jul 12, 2018 |
|
#
?
Jul 12, 2018 09:22
|
|
|
yeah the end effect of this is practically scorched earth -- even if the US stops its trade war after repeated poor sales of US domestic product start delivering the second Great Depression to the doorsteps of actual human beings (well, moreso than it already has), china stops being an x86 market to basically anyone and the knock on effect of this is that if AMD does get screwed out of this deal, it would still harm Intel even moreso than it would hurt AMD
Anime Schoolgirl fucked around with this message at 09:43 on Jul 12, 2018 |
|
#
?
Jul 12, 2018 09:31
|
|
|
|
| # ? May 17, 2024 21:01 |
|
|
SwissArmyDruid posted:AMD must either be ④ Knows from past experience that despite having a good product with Zen, Intel was going to gently caress them over with anti-competitive deals & monopoly power. So this time they're holding a live hand grenade with a pulled pin and yelling "I swear to loving god I'll do it!"
|
|
#
?
Jul 12, 2018 13:04
|
|