A 980 here is $700 and a Titan X is $1600. One is a pansy mid range sidegrade, the other is a hilarious rip-off. There is nothing worth updating to here until the 980 Ti.
I'm still surprised you use the 780 ti rather than pick up a new card. I'd have thought your performance tolerance wouldn't be met by the 780 ti at this point haha. Waiting for the 980ti?
I ran some benchmarks on three of my cards, a GTX 980, GTX 780 Ti, and an R9 290, and came up with some pretty surprising findings. Here's my blogpost about it.
In short, it's not Gameworks that's causing the problem for Kepler. But Kepler is definitely under-performing in this game, so something in the game engine is hitting it really hard.
Thoughts?
http://www.eteknix.com/nvidia-gtx-750-ti-2gb-maxwell-graphics-card-review/2/he SM scheduler architecture and algorithms have been rewritten to be more intelligent and avoid unnecessary stalls, while further reducing the energy per instruction required for scheduling.
The organization of the SM has also changed. Each SM is now partitioned into four separate processing blocks, each with its own instruction buffer, scheduler and 32 CUDA cores. The Kepler approach of having a non-power-of-two number of CUDA cores, with some that are shared, has been eliminated. This partitioning simplifies the design and scheduling logic, saving area and power, and reduces computation latency.
Pairs of processing blocks share four texture filtering units and a texture cache. The compute L1 cache function has now also been combined with the texture cache function, and shared memory is a separate unit (similar to the approach used on G80, the first CUDA capable GPU), that is shared across all four blocks
http://devblogs.nvidia.com/parallel...ould-know-about-new-maxwell-gpu-architecture/As with SMX, each SMM has four warp schedulers, but unlike SMX, all core SMM functional units are assigned to a particular scheduler, with no shared units. The power-of-two number of CUDA Cores per partition simplifies scheduling, as each of SMMs warp schedulers issue to a dedicated set of CUDA Cores equal to the warp width. Each warp scheduler still has the flexibility to dual-issue (such as issuing a math operation to a CUDA Core in the same cycle as a memory
operation to a load/store unit), but single-issue is now sufficient to fully utilize all CUDA Cores
First i want to say, great job on the data and the blog. I appreciate the anyone who takes the time to investigate and put in real time to try to analyze things themselves.
We have been discussing some of the potential reasons. For starters, Maxwell is structured differently than kepler and it breaks down different. Surely you have heard that maxwell is 128 cuda cores per SM and kepler was 192 per SMX. But for kepler, that is 192 cores per SMX- which all shared the same cache texture cache, unified cache, and L1 cache).
maxwell 128core SM is different in a pretty large way, the cache (L1,Texture) is shared between two blocks of 64 cores. Also the L2 is much larger. It breaks down much more elegantly,
http://www.eteknix.com/nvidia-gtx-750-ti-2gb-maxwell-graphics-card-review/2/
http://devblogs.nvidia.com/parallel...ould-know-about-new-maxwell-gpu-architecture/
These differences show up from time to time, kepler can be at a disadvantage.
https://compubench.com/result.jsp
We also have AMD GCN in all the next gen consoles, this was supposed to give AMD GCN an advantage in PC as game engines are designed around that specific HW. There have been titles that kepler struggles on, I think maxwell is just better suited in those. And as more titles get built from the ground up, with the next gen consoles in mind, the issue wont go away.
If consoles ports start using direct/open compute, then kepler performance will suffer. We could call it a scheduling issue, kepler just being the odd one out.
There are plenty of engines that are still in use that kepler will still be fine in. And dont get me wrong, I am not saying that this is all that is going on here. AMD has also done a lot to improve their performance these past few years as well. The omega drive gave AMD GPUs about a 10% performance jump across the board. That is pretty awesome right there, I believe that nvidia has also improved Maxwell performance at least 5% or more since launch.
Maxwell is more different than most people seem to think. Its strengths are in the areas where kepler has been weak. So in the end, i think that there will be cases that maxwell and GCN both do well in. But it is not the end of the world, far from it.
People are trying to spin this into some huge conspiracy when there are very real and technical reasons right there in front of them. There is data showing kepler weaknesses all around and in these same areas, Maxwell does very very well.
The game is under 2GB VRAM for obvious reasons, have u seen the horrible vegetation?
Yeah, although I cannot remember where, I've read that while the devs at some point said the install was over 50GB, it now is way less.
One can take an educated guess what parts of the game took a hit from that (and why they did it in the first place if you fancy a tinfoil hat)
Interesting to see how Tonga (285) is performing vs Tahiti (280x) here. 11% faster at 1080p with 12.5% less shaders and some amount of a frequency disadvantage.
Interesting to see how Tonga (285) is performing vs Tahiti (280x) here. 11% faster at 1080p with 12.5% less shaders and some amount of a frequency disadvantage.
I finally got around to playing the game last night. I messed with the settings quite a bit, and I ended up between 40-60 FPS @ 1440p with no AA (even with it off, I'm not seeing jaggies), SSAO (not HBAO+) everything set to medium, except water on High and textures on Very high. All of the PP options are enabled as well, except for vignette.
This is with my GTX 780 clocked around 1215/1656. The game is very playable to me, and even with these mediocre settings the game is gorgeous IMO. I enabled Hairworks on Gerault, and honestly, I could barely see any difference in IQ, just a frame rate loss. Personally, I was more impressed with TressFX in Tomb Raider. I also must say, it's a breath of fresh air to have such little VRAM usage on a beautiful open world game.
On another note, I could never select the fullscreen option, as it always reverts back to windowed mode -borderless. Any ideas?
(1) you might try HBAO+, as it only took a 1fps hit versus SSAO in my testing.
I'll give HBAO+ a try, but do you really see a large difference with it enabled?
I'm not seeing much here-
http://international.download.nvidi...ve-comparison-4-nvidia-hbao-plus-vs-ssao.html
http://international.download.nvidi...ve-comparison-3-nvidia-hbao-plus-vs-ssao.html
...
"As you can see, Witcher 3 is very modest in its VRAM requirements, which is quite surprising given how beautiful the graphics are. To me this looks like good coding, so users with 2 GB cards will not have to worry at all".
Beautiful graphics ... LOL
On another note, I could never select the fullscreen option, as it always reverts back to windowed mode -borderless. Any ideas?
That is what is causing lots of people bad fps, i suspect lots of benchmark posts are doing it as well. The game resets to fullscreen boarderless after every restart of game, you have to put it fullscreen each time.
I don't understand why people are knocking the vegetation in game, it looks great to me. Its 3d, it blows in wind, moves when you run into it. What else do you want to watch it grow? lol
I don't understand why people are knocking the vegetation in game, it looks great to me. Its 3d, it blows in wind, moves when you run into it. What else do you want to watch it grow? lol
I'll give HBAO+ a try, but do you really see a large difference with it enabled?
I'm not seeing much here-
GTX670 vs 970 on same clock.1920x1080 All max with hairworks+HBAO+ and with shadows tweak for better shadows from
Maybe not in the first shot - second one you linked shows a big difference. Vegetation looks very flat without HBAO+ enabled. It's a setting that I would not disable personally, too big of a difference for outdoor areas.
The patch fixes stability issues, adds support for DLCs, and improves the overall performance of the game on all platforms. However, it seems like at the same time, the patch also scrapes some of the in-game effects on the Xbox One as well as the PlayStation 4 version of the game.
The day 1 patch adds notable improvements to the game, such as less geometry pop-in during cut-scenes, fewer instances of flickering shadows, slightly better frame-rate during gameplay, and other minor tweaks. However, these improvements come with a cost. As seen in the GIF comparison below, courtesy of EuroGamer, anisotropic filtering has been dropped a little with day 1 patch. Texture filtering has also been very slightly downgraded. Tail-off in clarity across the floor in the post-patch build of the game is quite visible in the comparison below.