A.Mitov Posted January 7, 2014 Share Posted January 7, 2014 Hi Guys, I'm in a pretty big dilemma here and would be really grateful if you can help me out. Which of these two video cards - GTX 770 and R9 290 - would you recommend for the following usage: 3ds max, Vray(production+RT), After effects, Photoshop, Autocad, Rhino 3D? I'm mostly seeing people prefering Geforce. Is there a particular reason for this? I heard that NVIDIA intentionally cripples OpenGL performance on their gaming cards and they have a very bad time dealing with OpenGL software (Rhino 3D, Blender etc..). Is this still the case with the 7 series as it was with the last few generations? I currently have a GTX 460 which is really terrible at OpenGL apps (Rhino 3D, Blender etc.) What about AMD? From what I'm reading the R9 290 is a beast of a card (at least for gaming). Does this translate to 3D software too and does AMD also cripple their OpenGL performance as NVIDIA does? Thanks in advance! Link to comment Share on other sites More sharing options...
Dimitris Tolios Posted January 8, 2014 Share Posted January 8, 2014 AMD cards do have better OpenGL than the average GTX card in their price range (or even outside), but you will lose compatibility with current VRay RT GPU version. Better OpenCL implementation in future versions will supposingly fix that, but that has been said before. Today, any upper level AMD GCN card (e.g. 7950, R9 280 etc) already vastly outperform any GTX offering in pure OpenCL compute. Unfortunately, Adobe stuff that use OpenCL in their latest iterations are CPU limited regardless of doing a tad better with AMD GCN. That said, a 460 is a farcry from either two you are asking for, I doubt that you won't notice a performance increase with any short of upgrade in this price range. Long story short: OpenGL viewports = AMD DirectX viewports = I would say a toss GPGPU = NVidia, till OpenCL catches up (then it will be much like coin mining is now, go AMD or go home. Unless Maxwell architecture changes everything of course. Link to comment Share on other sites More sharing options...
A.Mitov Posted January 9, 2014 Author Share Posted January 9, 2014 Thanks for the broad explanation. So looks like I've got to go with NVIDIA again. Unless someone comes up with a good argument why I shouldn't. Thanks again! Link to comment Share on other sites More sharing options...
Dimitris Tolios Posted January 9, 2014 Share Posted January 9, 2014 Thanks for the broad explanation. So looks like I've got to go with NVIDIA again. Unless someone comes up with a good argument why I shouldn't. Thanks again! Well, if it was for gaming or any workstation work outside CUDA related compute, AMD does great. For OpenCL compute, AMD GCN does AMAZING: I was messing with some benches and the W5000 card, and it actually outperforms any GK104 card by a considerable margin. For example in Luxmark 2.0, a benchmark for the OpenCL Luxrender engine & the "Sala" medium complexity scene: Luxmark V2 Scores Radeon R9 290X : ~2850* Radeon HD 7970 GHz: ~2350+* Radeon R9 280X: ~ 2330* Firepro W9000: ~ 2090* Radeon 7870: ~ 1461* GTX Titan: ~ 1375 Firepro W7000: ~ 1340* GTX 780: ~ 1230 GTX 580: ~ 890 Firepro W5000: ~ 805 Firepro V7900: ~ 660* GTX 770/680: ~ 660* GTX 670 SC ~ 584 Quadro 4000: ~ 295 Quadro K2000: ~ 144 Results with * are taken from review sites. Non * are personal averages using my 3930K @ 4.6 (I do it lower than my 24/7 clock to match the older results in the database with my 3820) The results with a "sub 75W" (PCIe port powered only) W5000, are impressive. It beats the top GK104 parts (770/680 @ 200+ Watt) handily. The actual competitor, the K2000 is a "joke" by comparison, 5.5x slower! Even the "old" V7900 actually matches a 770 in raw compute. BUT...I don't use Luxrender for my projects! Link to comment Share on other sites More sharing options...
eidam655 Posted January 10, 2014 Share Posted January 10, 2014 ...but you will lose compatibility with current VRay RT GPU version. Better OpenCL implementation in future versions will supposingly fix that, but that has been said before. I would like to know more about this. Does this mean that nVidia cards currently outperform AMDs simply because the OpenCL implementation is not as well optimized as CUDA? Is this going to change with VRay 3? Do you have any links to reviews/tech/official sites that discuss or benchmark this? Link to comment Share on other sites More sharing options...
Dimitris Tolios Posted January 10, 2014 Share Posted January 10, 2014 We hope will change. After much nagging, AMD cards do work with RT GPU OpenCL mode, but performance is far from indicative of their potential (i.e. much much slower than NVidia cards in either CUDA or OpenCL mode). "Official" sites you can get information from? I guess the RT GPU section in Chaosgroup's forums. VRay 3.0 for 3DS is already in beta testing me thinks, so maybe there will be some news around the matter. Or not...you see, as people were driven away from AMD for so long due to those limitations, it is rare to find early adopters, with current AMD cards, that are simultaneously forum warriors etc to provide regular updates. Link to comment Share on other sites More sharing options...
eidam655 Posted January 10, 2014 Share Posted January 10, 2014 well, that's a shame, especially when a supposedly similar radeon is about 60% price of an nvidia (at least what i found, R9 270X DUAL-X OC BOOST for 204€ versus N770 TF 4GD5/OC for 350€) Link to comment Share on other sites More sharing options...
RyderSK Posted January 10, 2014 Share Posted January 10, 2014 well, that's a shame, especially when a supposedly similar radeon is about 60% price of an nvidia (at least what i found, R9 270X DUAL-X OC BOOST for 204€ versus N770 TF 4GD5/OC for 350€) What do you base this on ? Since both are mostly benchmarked as gaming cards, 270X is not even close to 760 in major benchmarks. Link to comment Share on other sites More sharing options...
A.Mitov Posted January 10, 2014 Author Share Posted January 10, 2014 (edited) I think you might have a point here, guys. Maybe it's better to wait and see what the beta version has to offer before choosing the card. It will be interesting if there's an improvement in opencl performance. The new high end AMD cards certainly pack some serious horse power and it would be great to see it utilized properly! Edited January 10, 2014 by A.Mitov Link to comment Share on other sites More sharing options...
RyderSK Posted January 11, 2014 Share Posted January 11, 2014 I think you might have a point here, guys. Maybe it's better to wait and see what the beta version has to offer before choosing the card. It will be interesting if there's an improvement in opencl performance. The new high end AMD cards certainly pack some serious horse power and it would be great to see it utilized properly! While I am not beta-tester of Vray3.0, I really doubt there is much improvement. All the commercial gpu renderers move along the same thread. If you need the card now, just buy it. If you decide to wait, you might be surprised more by new nVidia Maxwell. Link to comment Share on other sites More sharing options...
eidam655 Posted January 11, 2014 Share Posted January 11, 2014 Juraj: I was looking for the first/cheapest thing to have 4GB GDDR and was struck by the price difference. I didn't look for any benchmarks, since searching for GPGPU ones would be pointless and I didn't think of gaming benchmarks... Link to comment Share on other sites More sharing options...
RyderSK Posted January 11, 2014 Share Posted January 11, 2014 Juraj: I was looking for the first/cheapest thing to have 4GB GDDR and was struck by the price difference. I didn't look for any benchmarks, since searching for GPGPU ones would be pointless and I didn't think of gaming benchmarks... That's very odd :- ) Cheapest nVidia with 4GB: GT630 4GB = 70 Euros (on Alza Zotac,MSI). Cheapest AMD with 4GB: 6570/R7 240 = 70 Euros (on Alza Sapphire) Comparable to 270X is 660Ti/or maybe even 760. Definitely not 770 in any regard. There is currently no "massive difference" in price between AMD and nVidia. I would say after recent falling down of prices by 770/780, price is decently equal on all fronts. I don't think searching for GP-GPU is pointless, you can find great range of both CUDA and OpenCL benchmarks on extremeoverclockers. For OpenCl, try for example the "Folding" part of forum since those guys build massive rigs with multiple cards. Link to comment Share on other sites More sharing options...
eidam655 Posted January 11, 2014 Share Posted January 11, 2014 Thanks, I'll take a deeper look into it btw since you're here, would you have any recommendations for a card Link to comment Share on other sites More sharing options...
RyderSK Posted January 11, 2014 Share Posted January 11, 2014 Thanks, I'll take a deeper look into it btw since you're here, would you have any recommendations for a card You can only choose among nVidia for the reasons above mentioned by Dimitris, and hence move along 660/670/760. For 200 Euros you can only get older 660, and it comes only with 3GB of ram. 670/760 come with 4GB but cost 250 euros at cheapest (Zotac, with reference type cooling), respectively 270 euros for Gigabyte with 3 coolers instead--)more silent (I own 3 of these cards, it's noticeably more silent) [prices from Alza] Link to comment Share on other sites More sharing options...
Valtiel Posted January 26, 2014 Share Posted January 26, 2014 Just a few things to toss in here. GPU Rendering: In addition to what Dmitris has mentioned, I've been reporting my experience with trying to get VRAY RT to work on my Tahiti cards here and on the Chaos Group forums. The short of it is they’re not usable. The long story is they finally work again after being broken for so long but they’re still gross under-performers; Apparently even old HD58xx hardware outperforms the 79xx’s so something is up at the software/driver/API levels. Vlado and Chaos Group say the OpenCL kernel is scratch built rather than a port of the CUDA version and they insist AMD’s drivers are to blame. AMD has yet to respond to my inquiries/complaints. I’m inclined to believe Chaos Group but it is apparent that Nvidia has simply been more active in working with them. I'm not a 3.0 Beta tester but I don't expect it to employ all of this wasted horsepower. I’ve asked a few review sites to look into trying APU models, 290’s, S10000s, and W9000s to see if the issue is just Tahiti drivers but so far nothing’s come of it. AMD, or anyone else, do not seem to be pushing hard enough to get OpenCL on a solid rendering platform. Next Limit and Pixar have not visibly committed to GPU applications and few people use Luxmark or RatGPU outside of benchmarks as far as I know. Thinking about what happened to the Brazil/Caustic team’s efforts just makes me angry. Nvidia is certainly doing their part to stay on top in both a partner-support capacity and straight-up owning one of the most prevalent renderers; I expect Mental Ray/iRay will never be OpenCL, effectively locked to Nvidia hardware. It looks like Octane, Thea, and finalRender 4’s implementation are CUDA-only as well. Just read this article and ctrl+f : ”opencl”. http://www.fxguide.com/featured/the-state-of-rendering-part-2/ The find count is sad pandas. While AMD’s been very keen to show off OpenCL/HAS advantages with physics or stereoscopy in Maya, I think rendering is what we professionals are watching and Nvidia has it cornered. I honestly think catching up in Vray RT is AMD’s best bet in turning it around. Prices: High end AMD cards (79xx, 280X, 290, 290X) are all selling well above MSRP and are globally difficult to find due to cryptocurrency miners snapping up more high-end cards per system than CG pros or gamers put together. NVidia cards, known to be poorer performers in mining, are well in stock and easier to find good deals on. AMD must be both laughing and struggling to keep up with demand. I’m late to the mining party but I’m glad I finally have a good use for my two 7950’s. They’re great for folding proteins with the new core_17 client but again, I don’t expect them do be great gpu rendering cards for awhile; it’s a CUDA game. Dmitris: Regarding my Firepro driver curiosity above, have you tested your W5000 on VRay RT by chance? Any better results than consumer cards? Do Intel iGPUs at least perform proportionately to their feeble potential (to shed light on the CUDA-bias vs. AMD’s driver neglect debate). Link to comment Share on other sites More sharing options...
Dimitris Tolios Posted January 26, 2014 Share Posted January 26, 2014 (edited) Dmitris: Regarding my Firepro driver curiosity above, have you tested your W5000 on VRay RT by chance? Any better results than consumer cards? Do Intel iGPUs at least perform proportionately to their feeble potential (to shed light on the CUDA-bias vs. AMD’s driver neglect debate). There is not real difference I am afraid...sorry for not coming up with more detailed tests, I am literarly sitting on a pile of cards atm, but I have little time to organize myself into doing the actual test/research for a more elaborate test. This is what lies within arm's reach from my desktop @ home. That's: FirePro V5900 / V7900 / W5000 Quadro K600/K2000/2000/4000 (a few more not shown, all older) GTX 660 + Titan (inside "FooBox", it is "wet" so doesn't get out for picks anymore - edit: you can actually see the lil illuminated waterblock on the bottom left ). Guesstimating from past experience at least a couple of hours with each card, going through various benchmarks, trying to optimize that new capturing routine + 25M poly + 100M poly animation scene for 3DS viewport I was trying to devise, and I have to admit I feel overwhelmed...with my current schedule it means a card / per day of free time at best. Edited January 26, 2014 by dtolios Link to comment Share on other sites More sharing options...
Valtiel Posted January 26, 2014 Share Posted January 26, 2014 Ah, all good. Thanks for being willing to do so either way. So AMD's not simply holding real functionality to pro cards...Good news I suppose. The image is appearing broken for me. Is it just a photo of gpus? Link to comment Share on other sites More sharing options...
Dimitris Tolios Posted January 26, 2014 Share Posted January 26, 2014 Yeah, fixed the link to the pic...quality is "amazing", thanks to my phone I don't know how all "old"/"pre-GCN" OpenCL engines work fine with AMD cards including the much newer Adobe CC OpenCL routines, but Chaosgroup has issues with AMD drivers... Maybe they are right, maybe AMD is working closely with everybody but them. Not impossible, but hard to believe that there is nothing to blame on their side. Link to comment Share on other sites More sharing options...
Valtiel Posted January 26, 2014 Share Posted January 26, 2014 Agreed, though I guess I can't blame them for putting their resources towards the vendor who reciprocates. Nice choice of speakers by the way. Rocking a pair myself. Link to comment Share on other sites More sharing options...
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now