No Intel has no competition. AMD can put up a pretty good fight with their GPUs unlike with their fail CPUs.
nVidia is quite literally the Intel of GPU's from a technological standpoint as far as I can tell, they just don't execute as well. You can't blame them though, being stuck with the problematic FAB tech we have these days. The work each individual shader can get done is pretty much double that of AMD's which is staggering (although I don't know about the 79xx series). That's how AMD rolls though, MORE COARS. Have to wonder if that will bite them in the ass down the line but I'm not familiar with GPU tech on that level.
I think the higher minimum frame rate argument is pretty telling though, but AMD's execution is pretty stellar as of late.
not quite. AMD's VLIW approach is better for doing raster graphics, while Nvidia's approach is better for general compute. for the task of doing strictly graphics work, AMD's design end up having a higher "IPC".
and six months make a pretty large difference. once again, AMD snagged all the early adopters, who are willing to spend big bucks on the latest and greatest. money Nvidia won't get.
I bet the tessellation performance hurts the 5870 as does 1GB of memory in some of these games.
When I say intel of gpus I'm referring to the IPC advantage Intel has vs AMD on the CPU side. Kinna mimics nVidias approach on the gpu side being that they have the stronger per core performance that's all. Wasn't referring to Intels success or total dominance, just strategy.
Wasn't that 6 month delay entirely because of Fermi? They fixed it in the end and got a generally faster product out of it for a few generations no? I don't see the doom in a 6 month advantage when all AMD is doing with it is charging people heavily for it then getting beat by the competitor in single GPU performance once that six months is up .
I think it was definitely a lead when they came out with first DX11 GPU in all, but there's no way you can match that kind of intensive every product cycle.
Yeah there are some outliers used here that are representing not straight GPU prowess but memory bottlenecks (BF3 ultra 4xmsaa, metro 2033 and others) and multi-threaded rendering driver advantages (Civ5).
Remove those and it's back to the reality of about a 20% gain. The 480 is obviously the better card, but it did come out 6 months later with 6 months more development time behind it than the 5870 and the attenuate thermi, furnace, nuclear-reactor powered, 747 acoustics anecdotes.
One forum member has had them in tri-SLI for so long I guess he must see something worthy about them heat or no heat.Fries eggs better than the 5870 as well
I'm interested to see how those tests would look on a 2gb 5870.
BTW, this is a bit OT, but I'm impressed that AMD finally came out with SSAA on the 79x0 series. I tend to play older games plus a bit of civ 5 (currently mowing through nwn2 again with a barb/frenz berzerker), and having SSAA available for these games on higher-end cards like gtx 480, 7970, etc, is just freakin' awesome. I remember reading about it for years, but now that I've used it on older games I wouldn't want to go back. It's nice to see that a feature that's actually useful/relevant/important to many gamers long-term has been prioritized by both camps. So :thumbsup: to both jhh and Read.
One forum member has had them in tri-SLI for so long I guess he must see something worthy about them heat or no heat.
There are some of us waiting for Lightning edition 7970's, iirc Groove is one of them
Well that's why nVidia keeps coming out with the faster top options no? To lure just those types? Folks who are willing to spend big bucks for the best would migrate if a viable upgrade came about for their 7970's ...
Rasterizing as in photo editing? Isn't that for work station graphics cards?
Yeah there are some outliers used here that are representing not straight GPU prowess but memory bottlenecks (BF3 ultra 4xmsaa, metro 2033 and others) and multi-threaded rendering driver advantages (Civ5).
In BF3 GTX560ti 1024MB is faster than HD5870, no Buffer wall here.
In Civ 5, HD69xx series is faster than HD58xx and it seams that HD79xx scales even better. So, architecture play's a big part in the performance.
In Metro 2033, again GTX560Ti 1024MB is faster, no Buffer wall again.
None of the above games needs more than 1GB at single 1080/1200p resolution.
nVidia is quite literally the Intel of GPU's from a technological standpoint as far as I can tell, they just don't execute as well. You can't blame them though, being stuck with the problematic FAB tech we have these days. The work each individual shader can get done is pretty much double that of AMD's which is staggering (although I don't know about the 79xx series). That's how AMD rolls though, MORE COARS. Have to wonder if that will bite them in the ass down the line but I'm not familiar with GPU tech on that level.
I think the higher minimum frame rate argument is pretty telling though, but AMD's execution is pretty stellar as of late.
BF3 on ultra with 4xMSAA and Metro 2033 on ultra with 4xAA use more than 1GB of VRAM @ 1080P. Without minimums frames you can't use a 560 getting a higher average FPS, a card that in many cases is faster than a 5870, as some sort of indicator that they are not bottoming out and hitting VRAM walls.
Multi-threaded rendering enabled drivers make a massive difference in Civ 5.
Explain this one to me, you think this is just raw performance and not something else ?
I mean really, you think this is just raw performance at work in what is an incredibly underwhelming game if you are just looking on the visuals ? If you want to talk extreme tessellation, the 480 jumping ahead in Crysis 2 looks reasonable considering the ample usage of tessellation in this game, but 100% faster in B:AC ? And you are using this to find your overall average ?
I'm not disagreeing with your sentiment, we knew the 480 was about 20% faster than a 5870 overall when it released almost 2 years ago and again now because you wanted to remind everyone, but your overall summation is incorrect as it's being skewed by said outliers. What is your point in this exactly ? Are you saying you expect when nvidia releases a 28nm DX11 card that this same pattern is going to be followed ? Or do you just want to talk about cards released 2 to 2 1/2 years ago, again ?
If Civilization V was solely a DCL test, then our 2560 results would be impossible – the 7970 is winning by 12% in a game NVIDIA previous won by a massive margin. NVIDIA only regains their lead at 1680, which at this resolution we’re not nearly as likely to be GPU-bound.
So what changed? AMD has yet to spill the beans, but short of a secret DCL implementation for just CivV we have to look elsewhere. Next to DCL CivV’s other killer feature is its use of compute shaders, and GCN is a compute architecture. To that extent we believe at this point that while AMD is still facing some kind of DCL bottleneck, they have completely opened the floodgates on whatever compute shader bottleneck was standing in their way before. This is particularly evident when comparing the 7970 to the 6970, where the 7970 enjoys a consistent 62% performance advantage. It’s simply an incredible turnabout to see the 7970 do so well when the 6970 did so poorly.
Batman Arkham City uses the same Unreal Engine by Epic, as Arkham Asylum, but thanks to the engine's modularity, it has been overhauled, outfitted with the latest technologies, including a graphics engine that takes advantage of DirectX 11.
If the GK100 doubles the compute power of the 580 the die size difference between it and the 7970 will be bigger than between 5870 and 480/580.Most of the DX-11 games will need more compute power in the near future. GNC is the right direction but im afraid that Tahiti is not enough to compete with GK100. Because of that, GK104, a middle end NV card could perform very close with 79xx series in those DX-11 games.
2012 is a DX-11 year, and you wont see the same small performance differences we had in 2009-2010 with DX-9 games in the mix. If GK100 will double its compute power it will have significant more performance in those DX-11 games than Tahiti. AMD will have to design a bigger chip some time in the near future in order to compete in DX-11 games and in GPGPU.
If the GK100 doubles the compute power of the 580 the die size difference between it and the 7970 will be bigger than between 5870 and 480/580.
They don't need more throughput to compete because they will compete in another price bracket. (e.g. IF the GK100 is ever launched and has 2times the compute power). Like they have done since the 58xx serie. (85-90% of the performance on 70% the die size)
Same goes for the 7870 that will have around 70% the die size and probably 90% the performance of the 104. (if i go by the latest rumours)
Also I'm not sure about Civ5, their is a limitation that is also happening on the 7970. It is still cpu limited. (easily seen when upscaling the resolution the frequency stays the same for all resolutions) where the nvidia products can still scale further (65fps vs 87fps). So this game is doing something entiry different on Nvidia and on AMD products (be it GNC or4/5 VLIW). But yeah, the previous VLIW architecture sucked for this game .
This is all wrong, AMD and Nvidia call different things "cores" so your comparison is pretty well pointless. Also, the only reason Nvidia has the top GPU out is that they make a giant GPU. The die size of the 580 is 520 mm whereas the 6970 is 389. If AMD wanted to they could go head to head with Nvidia's top GPU, but they'd rather worry about the upper mid-range. In fact, i'd wager that if both companies had to make a GPU that was the same size, AMD might actually make something faster than Nvidia.
AMD also always has a leg up on the latest fab process. I would argue that AMD is actually the better GPU maker from a purely technical standpoint. Nvidia has an edge in the compute department as well as marketing.
you are assuming that nvidia indeed come up with a faster option, and that it would be a better value preposition than whatever amd will release in the same timeframe. nvidia will have to release a card that's either significantly faster than the 7970 at slightly less cost, or a card that's just as fast as the 7970 but significantly less costly. anything else would result in a big "meh" from the people who already purchased a 7970, since it won't give them any incentive to buy another card. i'm not saying that kepler won't sell, i'm saying that it will have to be pretty damned good to make the big spenders buy into their premium margin top end cards.
"raster graphics" is the current video rendering method being used; the other being raytracing.
No, it's been that way for ages. In the past, it was just AMD on a 1/2 size smaller node. When NV finally "caught up" with gt200, they were always a step behind on process node shrinks, and that's continued to this day. They were late with fermi b/c it came out with the shrink to 40nm, but their greater engineering muscle/budget/etc helpd them to catch up with fermi 1.5. Now that we're going down to 28nm they are behind again, but almost all of us assume that they'll come out with something faster/larger/mostly better. And I wouldn't be surprised if they "caught up" again with the re-spin that will inevitably occur in 12 mos or so. But AMD is quite competitive in the gpu sector, and they're certainly not constrained by a "per clock" performance in gpus as they are in the (losing) war with intel. In fact, "per clock" doesn't mean as much with gpus b/c the architectures are still so different. That was possibly AMD's goal with BD as well, but they didn't execute it very well and their competitor in the cpu space is much more dominant than NV is in gpus.