rpglord Wrote:Has there been any kind of user benchmarks done which confirm this ?
AFAIK, NaturalViolence's
Minimum specs - video card thread generally gets most of its data from
this discontinued GPU benchmark.Dunno if it's what you're looking for, but it might be worthwhile to read.
Well NV in his thread is mentioning procesing power as well, and not just memory bandwith, which is correct way to judge gpu requirements
Since even 650 do have more then 500 GFLOP/s it would seem its enough for 4x, but I have to say NV was wrong about that, atleast not for all games.
I had 460 gtx 1 gb which I think had around 1000 GFLOP/s and it was not enough for either Last Story, Xenoblade or SMG2 at 4x
Now I have 660 gtx which has around 1800 GFLOP/s and these games are running much better on 4x.
So I would say for 4x more then 1000 GFLOP/s is needed
But, I dont think that even taking into the account both GFLOP/s and memory bandwith is enough, these are not the only 2 things which show gpu perfomance.
For example 650ti has 1400 GFLOP/s which is more then my 460 gtx had, and yet its not really faster, its about the same based on reviews and benchmarks...
Quote:There is really no big difference in power draw for 650 and 650 ti
Maximum Load : 64W vs 110W
GTX 650ti requires 6-pin cable and gtx 650 does not require that
That's a very good question rpglord. I've been waiting for someone to ask me that question since I made the thread (I think that was 2 years ago?).
I typed out a very long and elaborate response but unfortunately I accidentally closed my browser before I posted it. And well, you have all been there before. When something like that happens it's really frustrating and you don't want to to go through writing it all over again. At least not immediately. Plus I've run out of time for the night so if you respond to this post (which you must do otherwise the thread will show up as read and I won't bother to check it) I'll get back to you in a couple of days when I'm back home.
To give you a real quick summary:
My conclusions were based on the benchmark that shonumi linked. That benchmark data is the only real data we gave about GPU performance in dolphin. PC game performance doesn't always correlate well with dolphins video backend performance. The benchmark data showed a heavy emphasis on video memory bandwidth bottlenecks. But that isn't the whole picture. For a more elaborate response you'll have to wait until I get back from spring break.
(03-10-2013, 10:08 AM)NaturalViolence Wrote: [ -> ]That's a very good question rpglord. I've been waiting for someone to ask me that question since I made the thread (I think that was 2 years ago?).
I typed out a very long and elaborate response but unfortunately I accidentally closed my browser before I posted it. And well, you have all been there before. When something like that happens it's really frustrating and you don't want to to go through writing it all over again. At least not immediately. Plus I've run out of time for the night so if you respond to this post (which you must do otherwise the thread will show up as read and I won't bother to check it) I'll get back to you in a couple of days when I'm back home.
To give you a real quick summary:
My conclusions were based on the benchmark that shonumi linked. That benchmark data is the only real data we gave about GPU performance in dolphin. PC game performance doesn't always correlate well with dolphins video backend performance. The benchmark data showed a heavy emphasis on video memory bandwidth bottlenecks. But that isn't the whole picture. For a more elaborate response you'll have to wait until I get back from spring break.
Sure I would be interested to read more elaborate response when you have the time
Because based on my testings this is not the case. There is huge difference in perfomance between 460 gtx and 660 gtx on 4xIR, and they have similiar memory bandwith, it seems everything is as it should be ( or everything is as in pc games ), more powerful card performs better, regardless of memory bandwith ( when its not bottlenecked by cpu which is most of the time as we all know ).
But I only tested Last Story, Xenoblade and SMG2
(03-10-2013, 11:49 PM)rpglord Wrote: [ -> ]There is huge difference in perfomance between 460 gtx and 660 gtx on 4xIR, and they have similiar memory bandwith
The GTX 660 has around 30GB/s more memory bandwidth than the GTX 460, so it should perform better.
(03-11-2013, 12:04 AM)Starscream Wrote: [ -> ] (03-10-2013, 11:49 PM)rpglord Wrote: [ -> ]There is huge difference in perfomance between 460 gtx and 660 gtx on 4xIR, and they have similiar memory bandwith
The GTX 660 has around 30GB/s more memory bandwidth than the GTX 460, so it should perform better.
No I overclocked 460 memory to 2200 I think it was showing over 130Gb/s in gpu-z, and 660 has 144 Gb/s
Its not that big of a difference, plus there is huge difference in games.
I dont have the 460 anymore and I dont remember exact numbers, but I remember it couldn't handle SMG2 at all at 4x, where I was getting 60 fps at 3x it would drop to like 40 or so with 4x.
I am getting fullspeed now with 660gtx, so I would say that is the minimum for 4x, not lowly 650
This should be easy to test, there is no way any of the cards mentioned here like 650 or 7750 could handle SMG2 at 4x IR.
Quote:460 130Gb/s in gpu-z, and 660 has 144 Gb/s
Downclock GTX 660 memory clock then try again . Please use the same Dolphin version , same config to make sure the result is 100% correct
Quote:ere is no way any of the cards mentioned here like 650 or 7750 could handle SMG2 at 4x IR.
Of course , that game requires a strong GPU .
Edit : GTX 550Ti , GTX 650 , AMD 5770 , 6770 , 7750 GDDR5 can handle 4xIR on most game except GPU demanding games such as
SMG 1 &2 , XenoBlade ... ( Xenoblade 4xIR may be possible with those cards but SMG is no way)
NV's memory bandwidth theory is not 100% correct but the accuracy is quite high . I tested many GPU (Nvidia 9800GT , Nvidia 7300GT , AMD 5770 , GT 540M , 8600GT ) all of them prove that NV's theory is correct
(03-11-2013, 12:46 AM)admin89 Wrote: [ -> ]Downclock GTX 660 memory clock then try again . Please use the same Dolphin version , same config to make sure the result is 100% correct
I RMA'd the card because of the faulty fan, I am on intel HD 3000 now.
I will test it when I get the new card, but I got about 50% more perfomance from 660 gtx for only 10-15% more memory bandwith..its just doesnt add up
Its the cuda cores power which is giving me more perfomance in SMG2, not memory bandwith
Thanks for the detailed answers!
Another question: could you recommend me a vga for this system? Just for not to be easy. there are some restrictions:
my config is a htpc in mini itx housing. I built this computer not for gaming, just for watching movies, that's why I chose i3 3223 with intel hd 4000, so it has low power consumption and silent operation. I just realised that it's capable of running wii games, and now I try to make the hardware better for occasional gaming.
So it has FSP 350W power supply, and the box cannot receive top, large VGA-s with huge cooling fan. Is there any point to get a vga with properties that fulfil these requirements or it will be just simply too weak?
Anyway the best would be, if I could make a switchable graphics system as the notebooks have. So when I just watch movies (90%) I could use the integrated gpu, and when I want to play doplhin games (10%), I could switch to the dedicated one with higher performance.
Is this possible with a desktop computer?