Hi!
I'm currently in the process of picking parts for a new computer. Since I do stuff with my GPU frequently (more GPGPU than actual 3D), I have a particular interest in my GPU choice.
I can't really decide if I want a GTX285 or 295, or a newer GTX470 or 480. Wikipedia lists way better fillrates and comparable op/sec for the 200 series, while the 400 come with (tasty) dual precession and a much better memory bandwidth and size. Also, I am uncertain if the 150€ and ~150W intake more for a 480 (as compared to the 470) are really worth it.
The last time I dove into the details on GPUs was way back when I bought my then rather new GF8800GTS, so I am a bit behind. I'd really appreciate if someone more up to date could fill me in with a few aspects that may help me make that decision!
🎉 Celebrating 25 Years of GameDev.net! 🎉
Not many can claim 25 years on the Internet! Join us in celebrating this milestone. Learn more about our history, and thank you for being a part of our community!
GTX 200 or 400?
Quote: (more GPGPU than actual 3D)The 4xx series is flat out designed for general purpose computation, and does a vastly better job than the older generation of GPUs. The absolute newest architecture is actually in the 460, and apparently two 460s in SLI tends to yield better performance for similar money than a single 480.
Still, you'd be crazy to buy a 2xx series at this point. Especially if you're doing anything related to GPGPU, there's an expanded instruction set, much more CUDA/OpenCL/DirectCompute capability, etc.
Go for the GTX 400 series, they support DirectX 11 and OpenGL 4 that gives great visual quality [Tessellation, ...]. AFAIK GTX 200 series should support DirectX 10.1 and OpenGL 3.3 @ max but they work in lower temperatures [if that's a factor for you]
I almost bought a 2 series and then I realized it didn't support hardware tessellation. Get the 4 series as tessellation is such a huge jump in graphics technology.
Author
Well, I guess that's settled then - much quicker than I feared :)
Thanks! I'll now just have to decide whether it will be 2x460 or 1x470, since I too didn't really see that many advantages of the 480 in light of the price and consumption. I also wasn't aware, that tesselation is a new feature of the 400, and the videos I saw with that were rather mind blowing.
Thank you guys!
Thanks! I'll now just have to decide whether it will be 2x460 or 1x470, since I too didn't really see that many advantages of the 480 in light of the price and consumption. I also wasn't aware, that tesselation is a new feature of the 400, and the videos I saw with that were rather mind blowing.
Thank you guys!
Quote: Original post by PromitIf I am not mistaken, that is not quite true. All the nVidia 400 GPUs are, to my knowledge, the same Fermi chip with 512 CUDA cores grouped in 16 stream processors.Quote: (more GPGPU than actual 3D)The absolute newest architecture is actually in the 460
However, they apparently have great trouble manufacturing a Fermi that actually works flawlessly -- on the majority of them, one or several stream processors, and possibly a MMU are broken.
So, instead of sending 90% of their dies down the drain, what they do is they sell the 100% intact ones as "Fermi", the ones with one stream unit broken as 480, and so on (the 460 has 5 broken units). The broken units are disabled and are not visible to the end user.
Nope, not true. 470 and 480 are GF100, 460 is GF104. You can read about GF104, and it's a pretty cool piece of work IMO.
Author
That's quite an interesting article. Impressive what nVidia did there without loudly advertising it.
One paragraph caught my attention: "For GF104, NVIDIA removed FP64 from only 2 of the 3 blocks of CUDA cores. As a result 1 block of 16 CUDA cores is FP64 capable, while the other 2 are not. This gives NVIDIA the advantage of being able to employ smaller CUDA cores for 32 of the 48 CUDA cores in each SM while not removing FP64 entirely. Because only 1 block of CUDA cores has FP64 capabilities and in turn executes FP64 instructions at 1/4 FP32 performance (handicapped from a native 1/2), GF104 will not be a FP64 monster. But the effective execution rate of 1/12th FP32 performance will be enough to effectively program in FP64 and debug as necessary."
Since I do plan to go double precision in some projects, I am curious about how much of an impairment that would mean.
Also, I am not experienced enough to make the decision, if all this means that a 460 is actually generally better than a 470 at GPGPU, or if they both play roughly in the same league when considering everything both architectures bring to the table.
Edit: Never mind. I just found the respective tests in that article *blush*.
One paragraph caught my attention: "For GF104, NVIDIA removed FP64 from only 2 of the 3 blocks of CUDA cores. As a result 1 block of 16 CUDA cores is FP64 capable, while the other 2 are not. This gives NVIDIA the advantage of being able to employ smaller CUDA cores for 32 of the 48 CUDA cores in each SM while not removing FP64 entirely. Because only 1 block of CUDA cores has FP64 capabilities and in turn executes FP64 instructions at 1/4 FP32 performance (handicapped from a native 1/2), GF104 will not be a FP64 monster. But the effective execution rate of 1/12th FP32 performance will be enough to effectively program in FP64 and debug as necessary."
Since I do plan to go double precision in some projects, I am curious about how much of an impairment that would mean.
Also, I am not experienced enough to make the decision, if all this means that a 460 is actually generally better than a 470 at GPGPU, or if they both play roughly in the same league when considering everything both architectures bring to the table.
Edit: Never mind. I just found the respective tests in that article *blush*.
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement
