|
Arzachel posted:gently caress, meant Hawaii but brain decided to wander off for a bit, sorry. Still, you're not very good at a listing all those indications because if we go with only heat and density, GF100 was apparently meant for 32/28nm. You really think AMD was stupid enough to believe that TSMC would get 20nm scaled out and yielding well enough for large-ish GPUs in 2013? And that tweaked Bonaire is the GCN successor they're going with for 20nm? You've completely lost me.
|
#
?
Feb 20, 2014 00:36
|
|
|
|
| # ? Jun 4, 2024 01:15 |
|
|
Agreed posted:You've completely lost me. What makes you think Hawaii was backported from 20 to 28nm? Poor cooling and density alone do not make for a compelling argument.
|
|
#
?
Feb 20, 2014 00:56
|
|
|
Arzachel posted:What makes you think Hawaii was backported from 20 to 28nm? Poor cooling and density alone do not make for a compelling argument. Because AMD expected TSMC to be putting out 20nm parts by Q4 of 2012. From 2011:  It's a very open secret that TSMC's process delays have affected both nVidia and AMD. I didn't think that would be a controversial point, I'm surprised that it seems to be.
|
|
#
?
Feb 20, 2014 01:33
|
|
|
Agreed posted:Could definitely be viewed as a shot across the bow at Intel to remind them that they're not the only ones around who know how to build a chip. Kepler was an incredible sea change for nVidia, from its SMX modular construction to its efficiency, but it happened at exactly the right time. Intel has a huge process advantage, but nVidia knows a shitload about making GPUs. It's what they do Agreed (heh) about the competition with intel and the mobile focus. These things were (rightly) absolutely a big consideration for Maxwell and beyond, to the point where their next bottleneck is really going to be memory bandwidth - Sure you can get a great amount of compute power in a 2w envelope with this, but you're still going to end up constrained by slow ram interfaces, and textures are textures and don't really get smaller. Plain FLOPs on a chip are much cheaper in relative power terms than memory/memory bandwidth. On chip cache gives you far superior bandwidth and latency than external ram, but it's still got downsides of competing for die space on a SoC, and off-chip ram is far slower and much higher latency. We're going to need good solutions to power and latency hit of the memory side of the equation sooner rather than later at the rate the GPUs themselves are improving. I have no earthly idea what's on the horizon for RAM/sram/etc though.
|
|
#
?
Feb 20, 2014 02:36
|
|
Once this gets a die shrink it'll be a node behind Intel but it'll still bring (I'm guess....timating... let's go with that) some really significant power savings along with a bit of performance improvement. Intel has been very agile with their GPU work lately, and they can afford to be, in every sense of the word, but nVidia isn't going to just cede the market to them after having enjoyed it for so long.
|
... A press slide from 2011 is your proof that Hawaii was backported from 20nm? With no tapeout announcements or even rumors, no GCN 2.0 implementation or documentation, you'll need some actual evidence to back up that claim. quote:It's a very open secret that TSMC's process delays have affected both nVidia and AMD. I didn't think that would be a controversial point, I'm surprised that it seems to be. Note that "Hawaii is backported from 20nm" is a very different claim from "AMD/Nvidia expected TSMC's 20nm node to be usable by now at some point in the past", you'll hear no objections about the second. I don't think TSMC has released a node on time for the last six years.
|
|
#
?
Feb 20, 2014 13:27
|
|
|
Alright, man, if you think Hawaii is as intended, and that AMD's GPU strategy is on point, you do you. I'll speculate along my lines and you are free as a bird to do so along yours. I don't see how it affects much, especially considering that the emphasis in the post you're trying to tear apart (for _________ reason  ) was more about what's going on -right now- in GPU stuff, and my conjecture regarding AMD's expectations of the 20nm process and Hawaii was comparatively a throwaway line, I have to admit that I feel like you're a bit making a mountain out of a molehill here.
|
|
#
?
Feb 20, 2014 14:40
|
|
|
Has anyone seen 750's or 750ti's in any brick and mortar yet? Amazon had them ready to ship the day they were announced but Microcenter and Best Buy were sort of left out in the cold.
|
|
#
?
Feb 20, 2014 15:33
|
|
|
Agreed posted:Alright, man, if you think Hawaii is as intended, and that AMD's GPU strategy is on point, you do you. I'll speculate along my lines and you are free as a bird to do so along yours. I don't see how it affects much, especially considering that the emphasis in the post you're trying to tear apart (for _________ reason umma do me umma do me umma do me 28 mm in hawaii was the plan by A M D watch me do me watch me do me
|
|
#
?
Feb 20, 2014 15:44
|
|
|
Ignoarints posted:umma do me umma do me umma do me I genuinely don't think it was. GF and TSMC both had planned to have process shrinks accomplished well in advance of this. The development cycle for a GPU architecture runs about four years from conception to launch. AMD definitely expected testable 20nm silicon by Q4 2012. And, yeah, TSMC sucks at hitting their deadlines and node shrinks are really hard, but it is really so far-fetched that a card that basically turns everything else being done in computing, per se on its head with regard to efficiency and heat might have been intended for a process that 1. AMD expected would be established and in production by then, and 2. would have helped substantially lower those problems? Is that ridiculous? I don't think so. Edit: Points for the weirdest reply in a thread  about GPUs ever though. Five points. Six... maybe.
Agreed fucked around with this message at 16:13 on Feb 20, 2014 |
|
#
?
Feb 20, 2014 16:07
|
|
|
The dev cycle on GPUs is about 4-6 years from start to product. Three years ago, AMD was at least a year into Hawaii. I don't think it's a stretch to call it backported from 20nm at all especially with that slide from 2011.
|
|
#
?
Feb 20, 2014 17:30
|
|
|
I have never followed GPU releases before, but is there a general timetable based on the past for card releases after the first one hits the shelves? As in since the 750 maxwell was released, how soon can one expect the 760 maxwell replacement to come? Is it generally spread throughout the year?
|
|
#
?
Feb 20, 2014 19:19
|
|
|
Ignoarints posted:I have never followed GPU releases before, but is there a general timetable based on the past for card releases after the first one hits the shelves? As in since the 750 maxwell was released, how soon can one expect the 760 maxwell replacement to come? Is it generally spread throughout the year? Once you get production-proven architectures, which often launch in the mobile market lately and then make their way into discrete cards, it's just a matter of time and competition. Right now the GTX 760, 770, 780 and 780 Ti are holding the line just fine for nVidia, and it's probably considerably more profitable for them to keep making chips that have likely increased in yield over the time since Kepler's debut than it is to "pointlessly" replace them with Maxwell parts. I say "pointlessly" with scarequotes because it really does kind of come down to the fact that AMD isn't providing a strong market incentive for them to turn this into a race, at any of the more competitive price points or in the high end. I wouldn't be at all surprised to see this one grow from below rather than launch from the top, because nVidia isn't in a position where they have anything to prove, and they can likely get any production difficulties worked out for the more sophisticated cards with less time pressure on them. Kepler GK104 and GK110, like Tahiti/Tahiti XT (and other AMD card names/architectures, really), are proving to be long-lived. While there's no rush, while nVidia can still accurately claim to sell the world's fastest graphics cards and also compete on price, and while they can afford to wait on releasing the higher-performance parts with more transistors, it makes sense for them to wait and maximize the return on the Kepler investment. Still, there's also the Steady March of Progress and all the expectations that come with it, so I wouldn't imagine they'll be holding back exceptionally long. They're just in a good spot to play it by ear, see what AMD does, and respond intelligently where possible rather than fire preemptively when there's nothing to actually preempt at the moment.
|
|
#
?
Feb 20, 2014 19:38
|
|
|
An Upgrade Story: Can the GTX 750 Ti Convert OEMs PCs to Gaming PCs? Not personally interested in it, but for budget HTPCs or just cash-strapped people that don't want to build, this card sounds like the way to go.
|
|
#
?
Feb 20, 2014 19:39
|
|
|
Ignoarints posted:I have never followed GPU releases before, but is there a general timetable based on the past for card releases after the first one hits the shelves? As in since the 750 maxwell was released, how soon can one expect the 760 maxwell replacement to come? Is it generally spread throughout the year? Well, there's a combination of factors at play here: 1. Nvidia is going "bottom-up" with Maxwell, instead of the traditional "top-down". 2. Maxwell was originally meant for 20nm, so the 28nm GTX 750 feels like a sort of treading the waters before TSMC has the throughput to deploy the "real" Maxwell. 3. AMD has been dead silent regarding new architectures, and not in a good way. Maybe we'll start seeing the 800 series in April-June, or they could start releasing more 28nm Maxwell parts before then. Either way, announcements and rumors should start appearing in March.
|
|
#
?
Feb 20, 2014 19:40
|
|
|
Nephilm posted:Maybe we'll start seeing the 800 series in April-June, or they could start releasing more 28nm Maxwell parts before then. Either way, announcements and rumors should start appearing in March. The last official word I've read was http://wccftech.com/nvidia-maxwell-geforce-gtx-750-ti-gtx-750-official-specifications-confirmed-60watt-gpu-geforce-800-series-arrives-2014/ posted:NVIDIA also confirmed during the conference that they are planning to introduce the GeForce 800 series which is fully based on the Maxwell architecture in second half of 2014. This means that we will see the proper high-performance GPUs such as the replacements for GeForce GTX 780, GeForce GTX 770 and GeForce GTX 760 in Q3 2014. We have already noted codenames of the high-end Maxwell chips which include GM200, GM204 and GM206, however NVIDIA didn’t mention what process they would be based on but early reports point out to 20nm. I'm not sure if NVIDIA just confirmed second half of 2014 or Q3 but I wouldn't expect high-end parts in Q2.
|
|
#
?
Feb 20, 2014 19:47
|
|
|
My 860   Guess I'll just wait for more news.
|
|
#
?
Feb 20, 2014 19:53
|
|
|
Pimpmust posted:My 860 Yeah this is what I'm really looking forward to. I'm on the fence about getting another 660 Ti to SLI but the prices are just not great and generally getting higher (at least at the price I bought mine at). From the poor stock situation on the internet I assumed it was no longer being made and had a glimmer of hope that perhaps the 860 is on the near horizon but I realized it's probably because the current 760 more or less seems to be the 660 Ti replacement (for the time being). I dunnooooooooooo... pony up? 
|
|
#
?
Feb 20, 2014 20:06
|
|
|
I would wait, if only because the 192-bit bus on the 660Ti would probably bottleneck it in SLI.
|
|
#
?
Feb 20, 2014 20:17
|
|
|
How power-efficient is a GTX 860 likely to be? This new Steambox-style Silverstone case is super-tempting, but SFX power supplies top out at 450W. Will Maxwell make this case much more capable?
|
|
#
?
Feb 20, 2014 20:38
|
|
|
A 450W power supply is still enough to run a quad core i5 or i7 with an R9-290X or 780 Ti. How much more capable could the case be?
|
|
#
?
Feb 20, 2014 20:56
|
|
|
Hace posted:I would wait, if only because the 192-bit bus on the 660Ti would probably bottleneck it in SLI. Yeah it was my first thought but then I see things like this http://www.legionhardware.com/articles_pages/gigabyte_geforce_gtx_titan,4.html tl;dr on par with a titan I think it would be worthwhile I just feel lovely buying a phased out card, probably used card for basically the same price I bought it new when something shiny is on the horizon. It'd be an easier choice if I had ever SLI'd before and understood the quirks I hear about it too. Regardless, it seems like performance is pretty good, as long as it works. edit: And I see things like this that I missed not even a month ago quote:NVIDIA GeForce GTX 660 Ti 2GB GDDR5 Video Card $113-119.99+ Free Shipping and there is nothing to be found like that anymore Ignoarints fucked around with this message at 21:03 on Feb 20, 2014 |
|
#
?
Feb 20, 2014 20:57
|
|
|
Factory Factory posted:A 450W power supply is still enough to run a quad core i5 or i7 with an R9-290X or 780 Ti. How much more capable could the case be? I was just following the guidelines in the OP of the parts-picking thread, and assumed it wouldn't quite cut it. So 450W should be able to drive an i5 Haswell and a GTX760 just fine? Edit: Not really the right thread for this, I guess.
|
|
#
?
Feb 20, 2014 21:02
|
|
|
Yep. 84W of processor plus 170W of graphics card. You aren't fitting enough extras into that tiny case to come anywhere near a 450W PSU's limits.
|
|
#
?
Feb 20, 2014 21:04
|
|
|
cbirdsong posted:I was just following the guidelines in the OP of the parts-picking thread, and assumed it wouldn't quite cut it. So 450W should be able to drive an i5 Haswell and a GTX760 just fine? If there is anything people overbuy on its power supplies.
|
|
#
?
Feb 20, 2014 21:24
|
|
|
Don Lapre posted:If there is anything people overbuy on its power supplies. Speaking off, I have a vastly overpowered power supply (750W for an i5-4670k, 1 660ti, 1 SSD, 1 hard drive) and if I don't end up going SLI I always wondered if a large power supply naturally wastes power by being on or does it throttle itself efficiently enough to not matter a lot. If it matters in this case it's a 80 plus gold rosewill capstone
|
|
#
?
Feb 20, 2014 21:28
|
|
|
Ignoarints posted:Speaking off, I have a vastly overpowered power supply (750W for an i5-4670k, 1 660ti, 1 SSD, 1 hard drive) and if I don't end up going SLI I always wondered if a large power supply naturally wastes power by being on or does it throttle itself efficiently enough to not matter a lot. If it matters in this case it's a 80 plus gold rosewill capstone No it will only output what your system needs, if anything it probably draws less power than a smaller supply would that might be running in a worse spot on its efficiency curve with the same system.
|
|
#
?
Feb 20, 2014 21:32
|
|
|
A PSU draws only as much power as it needs, modified by its efficiency. PSUs are most efficient at 50-60% loading. So strictly speaking, your PSU is drawing a handful of watts more than would a PSU that was sized more appropriately, but we're talking about as much power as a single hard drive uses - not a lot.
|
|
#
?
Feb 20, 2014 21:33
|
|
|
Factory Factory posted:Yep. 84W of processor plus 170W of graphics card. You aren't fitting enough extras into that tiny case to come anywhere near a 450W PSU's limits. It also fits 3/2.5in drives, 1/3.5in drive an optical drive and 3 fans. KingEup fucked around with this message at 21:38 on Feb 20, 2014 |
|
#
?
Feb 20, 2014 21:35
|
|
|
KingEup posted:It also fits 3/2.5in drives, 1/3.5in drive an optical drive and 3 fans. So up to an extra 50W.
|
|
#
?
Feb 20, 2014 21:53
|
|
|
Factory Factory posted:A PSU draws only as much power as it needs, modified by its efficiency. PSUs are most efficient at 50-60% loading. So strictly speaking, your PSU is drawing a handful of watts more than would a PSU that was sized more appropriately, but we're talking about as much power as a single hard drive uses - not a lot. New power supplies while still most efficient at half load can still maintain 90%+ at full load. http://www.jonnyguru.com/modules.php?name=NDReviews&op=Story4&reid=326
|
|
#
?
Feb 20, 2014 22:03
|
|
|
Is there some chart showing the fps difference between older or ancient GPUs like the 8800GTX/4850/GTX 460 vs the new 750? I remember tomshardware had something similar.
|
|
#
?
Feb 20, 2014 22:31
|
|
|
DaNzA posted:Is there some chart showing the fps difference between older or ancient GPUs like the 8800GTX/4850/GTX 460 vs the new 750? I remember tomshardware had something similar. You're probably thinking of this. http://anandtech.com/bench/GPU14/815 It's amazing, doesn't have any of the really old cards on there though.
|
|
#
?
Feb 20, 2014 23:14
|
|
|
Just use the 2012 version: http://anandtech.com/bench/GPU12/372 The 750 has very similar performance to the 650Ti, so just compare and contrast with that instead.
|
|
#
?
Feb 20, 2014 23:41
|
|
|
Hace posted:Just use the 2012 version: http://anandtech.com/bench/GPU12/372
|
|
#
?
Feb 21, 2014 00:29
|
|
|
cbirdsong posted:I was just following the guidelines in the OP of the parts-picking thread, and assumed it wouldn't quite cut it. So 450W should be able to drive an i5 Haswell and a GTX760 just fine? This is my exact combo: i5 4570, GTX 760, and a 450w PSU (Rosewill Capstone). Built in early January. No issues.
|
|
#
?
Feb 21, 2014 02:40
|
|
|
Sure hope none of you wanted an 8xx series card for msrp
|
|
#
?
Feb 21, 2014 04:15
|
|
|
Don Lapre posted:Sure hope none of you wanted an 8xx series card for msrp Wish there was a combination of  and  Even accounting for VAT, AMD cards are cheaper here than in the states - I expect the same to hold true for the 800 series, though there probably will be a delay in availability.
|
|
#
?
Feb 21, 2014 05:29
|
|
|
Don Lapre posted:Sure hope none of you wanted an 8xx series card for msrp Can someone do the math and see if the cards have settled price/wise to match this chart?
|
|
#
?
Feb 21, 2014 05:42
|
|
|
Purgatory Glory posted:Can someone do the math and see if the cards have settled price/wise to match this chart? Price-wise to what? This is efficiency, and the initial investment doesn't matter because mining bitcoins is pure profit, so better performance per watt is more pure profit.
|
|
#
?
Feb 21, 2014 05:55
|
|
|
|
| # ? Jun 4, 2024 01:15 |
|
|
Upside: Maybe the prices for lower end AMD cards will finally go down?
|
|
#
?
Feb 21, 2014 06:03
|
|