|
Otakufag posted:But it'll be the year 2019 and info moves way faster than in early 2000's, look at what happened to Nvidia GeForce Partner Program due to all the exposure and nerds crying. fishmech posted:ignoring that the "competitive" part remains heavily qualified by AMD managing mostly to suck significantly less.........There remains a ton of workloads and configurations where AMD is in no way the only sensible choice or otherwise dominant. fishmech posted:People get to posting like it'll just be a massive shadowy Intel conspiracy preventing AMD from having 80% share next year That Intel pulled some secretive back door bullshit to hamstring AMD is a matter of fact at this point and I don't see any reason to give Intel the benefit of the doubt at this point and I don't see why anyone would either. Craptacular! posted:Kind of too late to demand motherboards be white boxes. Craptacular! posted:Intel isn’t the same type of litigious beast it used to be. ItBreathes posted:So with their massively different R&D budgets... PC LOAD LETTER fucked around with this message at 08:14 on Jul 21, 2018 |
# ? Jul 21, 2018 08:09 |
|
|
# ? Apr 25, 2024 12:00 |
|
PC LOAD LETTER posted:Except nothing really sucks at all about current Epyc parts except for maybe the name. By all accounts they're selling well and the main limiting factor is how fast they can make them. They're not as good at Intel for HPC stuff but AMD acknowledged that last year and said they were targeting the remaining 80% or so (according to them I believe) of the server market that isn't HPC so to say there is a "ton" of stuff where AMD isn't sensible/dominant might be technically true in a narrow but is also mostly eye rolling hyperbole in a practical sense. You can only stack like 1/4 as many cores into a system as on the Intel platform (224C vs 64C). That's a pretty big deal in compute-intensive markets (HPC). Rack units matter. Eating more units on the interconnects matters. RAM capacity matters, on different problems than RAM bandwidth. Smaller CCXs have consequences. HPC is a high-margin, high-profile market, and "we're trying guys" is not really an acceptable answer. Mindshare still exists - you have to have a good product for a prolonged period of time if you want to win the leaderboards. Yes, Rome is going to be a lot better in this market. But Epyc has barely ramped, despite 4 straight Quarters Of The Epyc Ramp now. Remember back in November when there was the HPC conference and all the partners wanted to know where the loving Epyc chips were, because they had boards waiting and poo poo, but couldn't get chips from AMD? Yeah, that was 8 months ago. Epyc has been "on the market" for a full year now, it's practically as much a vaporware as Vega. Let's be honest here: this earnings is going to be loving brutal. GPU sales are down 60% according to PowerColor, they sold more GPUs in March than all of Q2. Gigabyte's top-level earnings are down 30%. GPUs are roughly half of the computing group's earnings. Epyc and Ryzen growth is going to be "organic"... meaning 15% growth in 50% of 66% of AMD's revenue. AMD would do well to hold their total revenue to -10% q/q, -20% is possible, holding par would be excellent. AMD super underplayed their crypto exposure. Really hope they socked that money away into Zen2 or next-gen graphics. Paul MaudDib fucked around with this message at 08:29 on Jul 21, 2018 |
# ? Jul 21, 2018 08:16 |
|
PC LOAD LETTER posted:Except nothing really sucks at all about current Epyc parts except for maybe the name. By all accounts they're selling well and the main limiting factor is how fast they can make them. They're not as good at Intel for HPC stuff but AMD acknowledged that last year and said they were targeting the remaining 80% or so (according to them I believe) of the server market that isn't HPC so to say there is a "ton" of stuff where AMD isn't sensible/dominant might be technically true in a narrow but is also mostly eye rolling hyperbole in a practical sense. Everything I've seen says Eypc is actually really good for HPC. A lot of HPC work is memory bound and it's eight channel memory it has an advantage over Intel.
|
# ? Jul 21, 2018 08:17 |
|
Paul MaudDib posted:all kinds of stuff about HPC platforms/market Paul MaudDib posted:Epyc has been "on the market" for a full year now, it's practically as much a vaporware as Vega. Paul MaudDib posted:Let's be honest here: this earnings is going to be loving brutal. I mean at what point did I start talking about their earnings in my post that you're responding to and what does that have to do with anything I said EXACTLY? I mean do you think AMD will go bankrupt here? If you're assuming a worst case revenue reduction of 20% then it sounds to me like you're probably not. So what is the point of bringing this up in response to my previous post? You're posting like you work for Intel marketing at this point. Pablo Bluth posted:Everything I've seen says Eypc is actually really good for HPC. A lot of HPC work is memory bound and it's eight channel memory it has an advantage over Intel. Single thread performance, AVX 512 performance, and numbers of cores matters lots too and Intel can offer 4 and 8 socket systems for HPC which can not only greatly increase the number of cores but also aggregate bandwidth. They're hideously expensive of course and only a relative few will ever buy them but the profit margins are supposed to be absolutely wonderful for Intel's bottom line. PC LOAD LETTER fucked around with this message at 08:33 on Jul 21, 2018 |
# ? Jul 21, 2018 08:29 |
|
PC LOAD LETTER posted:What part of "they're not as good as Intel for HPC stuff" did you not understand or do you have a actual disagreement with? It's really not just HPC. Literally everyone cares about cores-per-rack-unit and RAM-per-box. If you're buying space in a datacenter (let alone building your own datacenter) then you absolutely care. Engineering time isn't free either. "HPC" here means literally anyone who owns or utilizes a datacenter. Those markets don't care about saving $1000 per server or whatever, the RAM alone costs them 5x that, let alone the TCO of a server in a rack unit in a datacenter. Startups who are building their own servers and putting them in their garage are absolutely down for Epyc though. quote:Server markets are slow to ramp and change which is why it took ~2yr for AMD to get 20%+ market share there years ago and its every bit the same way now. According to AMD they're on track to have 4-5% server market share by the end of the year FWIW and nothing I've seen contradicts that. Yup. Just saying, don't expect a turnaround in 6 months. quote:Maybe they will maybe they won't but so what? Lol ad-hominem. You disagree so I'm working for Intel marketing? The whole "AMD would be making bank if it weren't for Intel" thing is exactly the point. There are very legitimate reasons that AMD will go up slowly in the CPU market, and that their computing group (CPU+GPU) revenue may go down quickly due to crypto exposure. Paul MaudDib fucked around with this message at 08:41 on Jul 21, 2018 |
# ? Jul 21, 2018 08:35 |
|
Like, Oracle is what, $25k per core per month? SAP HANA costs $2.7 million for a 1 TB instance? What does a rack unit in a T1 or cloud datacenter cost? Lol if you think that saving $1k on the processor actually mattered even if AMD had an advantage there. Hardware is literally a commodity. You spend it to make software costs (engineer time, 3rd party licenses, etc) go away. And it is very cost effective to do so. Throwing engineers at the task to re-implement an available solution isn't free, and it isn't even reliable. "Well we can white box our own database solution for 1/10th the cost" is something that many, many people have been incorrect about. Paul MaudDib fucked around with this message at 08:53 on Jul 21, 2018 |
# ? Jul 21, 2018 08:44 |
|
Paul MaudDib posted:If you're buying space in a datacenter (let alone building your own datacenter) then you absolutely care. Paul MaudDib posted:"HPC" here means literally anyone who owns a datacenter. Paul MaudDib posted:Yup. Just saying, don't expect a turnaround in 6 months. Paul MaudDib posted:You disagree so I'm working for Intel marketing? Paul MaudDib posted:The whole "AMD would be making bank if it weren't for Intel" All I said was "Realistically speaking though I wouldn't be shocked if Intel uses the same dirty tricks that they used in the P4 days to keep AMD sales down." which isn't the same at all. Paul MaudDib posted:Lol if you think that saving $1k on the processor actually mattered even if AMD had an advantage there. PC LOAD LETTER fucked around with this message at 08:52 on Jul 21, 2018 |
# ? Jul 21, 2018 08:49 |
|
PC LOAD LETTER posted:Its got nothing to do with me disagreeing with you and more with your posting. They litterally read like Intel marketing stooge material. And you're known for posting like that Paul, others have called you out on that before, so don't go acting so shocked here at all. Lol. You really, actually can't substantively address my points, it's just ad-hominems all the way down. My points are pretty clear. Feel free to address them. Again, just to make it clear for you: the entirety of the hardware cost is a tiny fraction of the TCO, let alone just looking at the CPU in isolation. There are a lot of people who buy SPARC hardware (of all things) because it makes financial sense. $25k per core per month... and SPARC gets you 4x as many threads per core as any x86 platform. Your boss: All right PCLL, why do you think we should spend an extra $75k per core per month? boardroom eyes shift to you Coworker: But boss, we're not even occupying our current server fully, and if we switch to this 'open sores' software it's going to take a 2-year big-bang rewrite, and then we won't have any support contracts if it breaks! Oracle will give us patches within 24 hours and if they don't work they'll send out a tiger team, that's what we pay them for! And, Postgres doesn't have hotpatches, so we will have to do downtimes to deploy patches, we have a five-year uptime on our Oracle DB right now! That's the reality in a lot of companies. My current employer isn't even big, and that's the deal. Yeah, the hardware is poo poo, but we'd have to talk a government client into giving up their support contracts to switch and lol if you think that's going to happen, just lol. You're not paying for the hardware, you're paying for the whole shebang. Hardware, OS, database, support, and now you're paying for Java too (Oracle just killed CDDL licensing, so fire up that OpenJDK boys). Whatever experience you have home-building rigs, or white-boxing for startups/small businesses, is wildly inapplicable to many, many markets that pay lots and lots of money. (lol if you thought I was shilling for Intel and not Oracle, and I'm not even. That's just an explanation of how it is, and $1000 savings on a processor is not going to change that. And there's plenty of other "business critical" packages that are licensed in the same way. Physically buying a couple TB of memory for a SAP HANA system is not even the biggest cost of building a SAP HANA system. You could echo the same thing for lots of the people who "commercial Unix" is still built for, and yes that is still a thing. Or zOS, and lots of other poo poo you think is dead and never used in the real world. Yes, "real world engineering practices" are loving horrifying.) tldr: your loving Oracle support contracts Paul MaudDib fucked around with this message at 10:23 on Jul 21, 2018 |
# ? Jul 21, 2018 09:19 |
|
Paul MaudDib posted:Lol. You really, actually can't substantively address my points, it's just ad-hominems all the way down. I mean how is "Most datacenters by 1-2 socket systems and that is where AMD has its advantage though." an ad hom? Be exact when you reply too. Paul MaudDib posted:My points are pretty clear. Feel free to address them. I'd also point out that at no point did I say that software licensing costs aren't a bigger expense than hardware or that I don't understand how that software licensing works either so I don't know why you're still blathering on about it. It litterally has nothing to do with what I've said and isn't really relevant to the thread at all because no matter how much the software costs it doesn't mean that hardware price doesn't matter.
|
# ? Jul 21, 2018 10:05 |
|
PC LOAD LETTER posted:Huuuh? There is like one part of that and my previous posts in response to you so far even you are claiming is a ad hom. You might not like or agree with the rest of the responses but if you're gonna call ad hom on everything then you don't know what that term means. Epyc is going to be hugely popular in the insanely cost sensitive cloud markets, where FLOPS/watt are important, and the cost of the box is literally the entire cost, because the entire software stack was developed in house. Most VM instances are memory bound, then IO bound, then CPU bound. A top end Epyc box can have 64 cores, 1 TB of ram, 2 100GbE cards, and 24 4x NVMe drives in it. If 7nm and Rome manage to get the GoFlo/TSMC published 40% power savings and 15% frequency gains, plus an extra 5% IPC, the 4 core CCX would be incredibly compelling. If they move to 8 core CCX then it'll just end up hilarious, since there would be no NUMA node weirdness to deal with on 90% of the commercial parts.
|
# ? Jul 21, 2018 10:48 |
|
Methylethylaldehyde posted:A top end Epyc box can have 64 cores, 1 TB of ram, 2 100GbE cards, and 24 4x NVMe drives in it. keep going
|
# ? Jul 22, 2018 08:06 |
|
Mofabio posted:keep going It has enough horsepower to provide 100-140 GBE of throughput to an entire rack of compute boxes, without pause or fail, forever. 100+ TB of NVMe storage, fault tolerate and virtualized, ready and willing to be used by anything able to access it's iSCSI mapping API.
|
# ? Jul 22, 2018 08:41 |
|
Methylethylaldehyde posted:fault tolerate and virtualized mmm uuuhnn
|
# ? Jul 22, 2018 09:00 |
|
Paul MaudDib posted:
When you said this on July 2nd, AMD closed at $15.02. Considering that share value has increased to $16.50 in the 20 days since then,
|
# ? Jul 22, 2018 20:25 |
|
Paul MaudDib posted:Lol. You really, actually can't substantively address my points, it's just ad-hominems all the way down. Nobody's addressing your points because they're entirely anecdotal. You're countering statements like "AMD claims 4-5% server market share by year's end, and i can't find any data to counter that claim" with "i can't convince my boss to switch to postgres because they don't want to skip out on their oracle contract." Or, "but look at GPU sales" or "epyc is practically nonexistent in real deployments (CITATION NEEDED)" or "bb... bbut crypto exposure! they exposed to crypto!!!" It's hard to take you seriously when you complain about people arguing in bad faith when that's literally all you do in this thread.
|
# ? Jul 22, 2018 20:29 |
|
Paul MaudDib posted:Again, just to make it clear for you: the entirety of the hardware cost is a tiny fraction of the TCO, let alone just looking at the CPU in isolation. There are a lot of people who buy SPARC hardware (of all things) because it makes financial sense. $25k per core per month... and SPARC gets you 4x as many threads per core as any x86 platform. I mean, like, really, what place does this argument have in the conversation at all? How relevant do you honestly think SPARC's market share is? Lots of shops still run AS400s too, I bet that spells doom for AMD as well somehow, right? Meanwhile, the world has an absolutely insatiable demand for x86 datacenter power for cloud services. China in particular is nearly doubling their capacity every year. That's where AMD is aiming right now, nobody gives a gently caress about sparc except whoever's left at Oracle.
|
# ? Jul 22, 2018 20:46 |
|
How much throughput do modem SPARC processors put out per dollar these days?
|
# ? Jul 22, 2018 22:11 |
|
GRINDCORE MEGGIDO posted:How much throughput do modem SPARC processors put out per dollar these days? About 2-3% goes back to the Oracle sales guy who conned you into buying it. IBM does the same poo poo with POWERPC. Ask me about replatforming an Oracle database from Linux/x86 to AIX/POWERPC in TYOL 2016. The same company also stood up a new DB instance in 2015. They did not always make good decisions.
|
# ? Jul 22, 2018 22:24 |
|
ITT: AMD is doomed because SPARC is clearly the price/performance winner, when Intel isn't. Thanks MuadDib.
|
# ? Jul 23, 2018 10:58 |
|
PC LOAD LETTER posted:Except nothing really sucks at all about current Epyc parts except for maybe the name. By all accounts they're selling well and the main limiting factor is how fast they can make them. They're not as good at Intel for HPC stuff but AMD acknowledged that last year and said they were targeting the remaining 80% or so (according to them I believe) of the server market that isn't HPC so to say there is a "ton" of stuff where AMD isn't sensible/dominant might be technically true in a narrow but is also mostly eye rolling hyperbole in a practical sense. First time reading a post, huh? We've all been there, friend, we've all been there.
|
# ? Jul 23, 2018 16:23 |
|
I want to go back in time to see fishmech defend Netburst
|
# ? Jul 24, 2018 08:31 |
|
Munkeymon posted:First time reading a post, huh? We've all been there, friend, we've all been there. Yeah I fell for that trap but I guess I was on a posting roll that night edit: Leaked Intel HPC roadmap that looks very not good for them in 2019 and 2020: https://imgur.com/a/r29ceIp Man if AMD can really manage to deliver what the leaks are saying performance wise I think they're gonna do great in 2019 and probably 2020 too. PC LOAD LETTER fucked around with this message at 18:28 on Jul 25, 2018 |
# ? Jul 25, 2018 17:58 |
|
Sell some fabs Intel
|
# ? Jul 25, 2018 23:54 |
|
AMD's earnings are in, and it's actually really good. Computing/graphics group revenue is down a bit and income is down about 15%, but enterprise/semicustom revenue increased by 26% and income more than quadrupled, which more than offset the C/G group. https://www.marketwatch.com/press-release/amd-reports-second-quarter-2018-financial-results-2018-07-25 Per the earnings call, crypto went from 10% of revenue in Q1, to 6% in Q2, and they're predicting 0% in Q3. I'm guessing that also implies they expect revenue to shift from graphics to the CPU side of the C/G group. Which is probably a smart expectation at this point, the GPU firesales are happening as we speak. It sounds like they don't really expect the semicustom windfall to continue into Q3, but it sounds like Ryzen and datacenter graphics are going to be able to pick up the slack. Paul MaudDib fucked around with this message at 00:49 on Jul 26, 2018 |
# ? Jul 26, 2018 00:38 |
|
The latest Intel rumor, which I bring up only because of AMD's sordid past decade: Intel's Alleged Core i7-9700K Coffee Lake Refresh 8-Core CPU Hits SANDRA Sans HyperThreading Intel, are you guys doing okay over there?
|
# ? Jul 26, 2018 03:15 |
|
Am I an idiot if I plan on getting the 9900K preferably binned and run it at like 5.2 on a fuckoff expensive board with fuckoff expensive RAM. I've already got an overkill custom loop for cooling, and I just want the best gaming performance. It seems nothing else on the market will challenge that in the areas of single threaded esport games and also online CPU killers simultaneously. AMD is going to come a lot closer with performance ~gaming~ next year, are there signs that they will actually take the top end (5GHz+ CPU 4000+ men)? I do also do nerdy physics poo poo but we have a uni supercomputer for that, never on my personal computer. E: I also do not care for ethics in game processors BurritoJustice fucked around with this message at 03:49 on Jul 26, 2018 |
# ? Jul 26, 2018 03:47 |
|
If you're gaming at 1080p and want to try and max out your 240hz monitor I can see that making some sense because then yes that sort of a set up might actually give you enough fps to really matter towards that goal. BUT if you game at 1440p and/or are "only" shooting for maxing a 120/144hz monitor I don't see how that would make sense really because at that point the difference in performance between current Intel and AMD CPU's is going to be minimal almost no matter what you do. IMO saving your money for the NV 1180, or whatever they call it, seems way more sensible AND practical if your goal is max fps this year. Maybe by mid next year that'll change but it sounds like you might not be able to hold out until then. And yeah if the performance rumors are spot on then AMD could definitely meet or beat even top end OC'd Intel next year with Zen2 for gaming. Even then GPU will probably still matter a whole lot more.
|
# ? Jul 26, 2018 05:16 |
|
SwissArmyDruid posted:The latest Intel rumor, which I bring up only because of AMD's sordid past decade: Intel's Alleged Core i7-9700K Coffee Lake Refresh 8-Core CPU Hits SANDRA Sans HyperThreading I remember back in the early K7 days everyone was racing to 1 GHz and Intel "clenched" it by giving marketing the reigns, throwing engineering out the window, and blasting a Coppermine to 1.03GHz or something stupid it couldn't realistically maintain for retail sale and they had to recall them. It didn't matter, they got the headline. That's Intel for you. I really like AMD's new marketing of being snarky but present about their strengths and limitations. Although I would enjoy a useless ad or two poking fun at NetBurst architecture just to troll Intel that little bit more. Poke a little more and make the bully on the playground consume themself.
|
# ? Jul 26, 2018 05:47 |
|
SwissArmyDruid posted:The latest Intel rumor, which I bring up only because of AMD's sordid past decade: Intel's Alleged Core i7-9700K Coffee Lake Refresh 8-Core CPU Hits SANDRA Sans HyperThreading 8 core coffee lake likely beats 6c/12t coffee lake in general even if we're generous about HT performance gains. It's probably also responsible for the higher all core turbo. Winks fucked around with this message at 09:23 on Jul 26, 2018 |
# ? Jul 26, 2018 09:20 |
|
PC LOAD LETTER posted:If you're gaming at 1080p and want to try and max out your 240hz monitor I can see that making some sense because then yes that sort of a set up might actually give you enough fps to really matter towards that goal. I'm currently only ever CPU bottlenecked (2600 non-k with 4x4GB 1600c8 memory) in absolutely everything. So the plan is get the best CPU upgrade currently possible and ride that out. GPU upgrade will come after, probably when the mid generation 1180ti or equivalent comes out. It seems like a short wait for the intel option, and my money is worth the year(?) it will take AMD to catch up.
|
# ? Jul 26, 2018 11:53 |
|
Paul MaudDib posted:Per the earnings call, crypto went from 10% of revenue in Q1, to 6% in Q2, and they're predicting 0% in Q3. I'm guessing that also implies they expect revenue to shift from graphics to the CPU side of the C/G group. Which is probably a smart expectation at this point, the GPU firesales are happening as we speak. By 'firesales' do you mean prices are where they ought to be for a ~three year old nV design* without the craptobubble driving prices up or are prices actually below where you'd otherwise expect? *no, I don't care about
|
# ? Jul 26, 2018 15:02 |
|
1080Tis are starting to hit ebay at 400. Gonna upgrade my 980ti with a 200bux 1080ti at the end of the year hopefully
|
# ? Jul 26, 2018 15:11 |
|
!!L@@K Lightly Used* 1080Ti in ORIGINAL BOX L@@K!! *used only to mine Litecoin
|
# ? Jul 26, 2018 15:56 |
|
BurritoJustice posted:I'm currently only ever CPU bottlenecked (2600 non-k with 4x4GB 1600c8 memory) in absolutely everything. If you feel you have to use the liquid metal thermal paste (which you don't, even just using OK generic stuff will be a big step up) make sure you protect the components on the CPU package with nail polish first though. Truga posted:1080Tis are starting to hit ebay at 400. PC LOAD LETTER fucked around with this message at 17:25 on Jul 26, 2018 |
# ? Jul 26, 2018 17:22 |
|
The Asrock X470 Master SLI's VRM is, uh, pretty wonky to say the least: https://www.youtube.com/watch?v=mjoa2mU4uMw It doesn't have working overtemp protection on the VRM and when the temp sensors reaches 125° C it rolls over to zero This is on a testbench with no airflow over the VRM at all though so not really a scenario most people will run into, but still, that's pretty bad! TheFluff fucked around with this message at 18:16 on Jul 26, 2018 |
# ? Jul 26, 2018 18:13 |
|
PC LOAD LETTER posted:If you feel you have to use the liquid metal thermal paste (which you don't, even just using OK generic stuff will be a big step up) make sure you protect the components on the CPU package with nail polish first though. Intel has reportedly finally decided to start soldering i7s and i9s again, which makes these a pretty obvious move for anyone interested in peak gaming performance who isn't willing to have conductive liquid sloshing around in their processor in exchange for an extra degree of temps.
|
# ? Jul 26, 2018 18:17 |
|
TheFluff posted:The Asrock X470 Master SLI's VRM is, uh, pretty wonky to say the least: So it also cannot tell if the temp is negative, which is also a dangerous place to be. That seems really dumb. 125C is a really weird spot, I'd kind of expect it at 127/128 and using a signed byte.
|
# ? Jul 26, 2018 18:44 |
|
https://twitter.com/AMDNews/status/1022235553838903296
|
# ? Jul 26, 2018 19:13 |
|
Paul MaudDib posted:Intel has reportedly finally decided to start soldering i7s and i9s again,
|
# ? Jul 26, 2018 19:35 |
|
|
# ? Apr 25, 2024 12:00 |
|
TheFluff posted:The Asrock X470 Master SLI's VRM is, uh, pretty wonky to say the least: The bios on the asrock x470 itx is the worst most unfeatured beta style bios I've ever used. The board was £172, I'm giving asrock a hard pass next time.
|
# ? Jul 26, 2018 19:37 |