Over the last few weeks I’ve been playing around with MSI Afterburner and EVGA Precision x16 to see what I can meaningfully squeeze out of this card before plunging into the land of the GTX 900 series cards. I’ve had this card in my current machine since January 2012 and it has served me very well. The vast majority of it’s life has spent “idling” at the stock GPU clock of 855Mhz which to be fair is a solid overclock on the stock GTX580 specifications anyway. What little gaming I do it has munched through and it’s only since really starting to work with Blender more recently that I have started to experience that the GTX580 is starting to look a little long in the tooth.
The 900 series cards were announced in September 2014 and they really do look like they would be an immense upgrade over the GTX580 in terms of raw ppw (performance per watt). When the GTX580 Classified was released it really was one of the highest performing single GPUs money could buy (which retaining ownership of all ones organs). I needed the GPU acceleration for lots of video editing work I was doing and in that respect it was an incredible investment. Yes it was expensive (£550) but it was and still is a total workhorse of a component, it still plays recent high end games at max and pretty much deals with it all without any fuss at all which is impressive to say the least considering the GTX580 reference designs are now 5+ years old.
So anyway, I have been playing around with the Classified and thought I’d get this all written up so that when I do decide to upgrade to a 900 series card and I can rerun these benchmarks and see what my cold hard cash has bought me. The overclocking setting I have been playing around with are pretty rudimentary really as I’m simply using the software tools provided by EVGA and no special hardware for the task as I’m still far more preoccupied with stability than pure blistering speeds.
Test Machine
The host machine specifications are:
EVGA GTX 580 Classified Blender Benchmarks
The results of the various benchmarks I ran can be seen in the table below:
|
File
|
OC
Settings
|
Tile
Size
|
Processor
|
Time
|
|
BMW27.blend
|
N/A
|
16×16
|
CPU
|
06:10.03
|
|
BMW27.blend
|
N/A
|
256×256
|
CPU
|
08:05.94
|
|
BMW27.blend
|
–
|
128×128
|
GPU
|
02:08.91
|
|
BMW27.blend
|
–
|
256×256
|
GPU
|
01:49.18
|
|
BMW27.blend
|
GPU +50 Mem +50
|
256×256
|
GPU
|
01:43.82
|
|
BMW27.blend
|
GPU +50 Mem +100
|
256×256
|
GPU
|
01:42.05
|
|
BMW27.blend
|
GPU +50 Mem +100
|
512×512
|
GPU
|
01:42.71
|
|
BMW27.blend
|
GPU +50 Mem +100
|
16×16
|
GPU
|
33:16.59
|
|
BMW27.blend
|
GPU
+55 Mem +100
|
256×256
|
GPU
|
01:41.58
|
|
BMW27.blend
|
GPU
+55 Mem +125
|
256×256
|
GPU
|
01:40.77
|
|
BMW27.blend
|
GPU
+60 Mem +125
|
256×256
|
GPU
|
01:40.63
|
|
BMW27.blend
|
GPU
+70 Mem +125
|
256×256
|
GPU
|
01:40.20
|
|
BMW27.blend
|
GPU
+75 Mem +100
|
256×256
|
GPU
|
–
|
|
BMW27.blend
|
GPU
+75 Mem +125
|
256×256
|
GPU
|
01:40.63
|
|
cycles_bench_272.blend
|
–
|
256×256
|
GPU
|
12:01.27
|
|
cycles_bench_272.blend
|
GPU +50 Mem +100
|
256×256
|
GPU
|
11:12.48
|
|
sponza_cycles_benchmark.blend
|
–
|
256×256
|
GPU
|
10:12:55
|
|
sponza_cycles_benchmark.blend
|
GPU +50 Mem +100
|
256×256
|
GPU
|
09:47.31
|
The end result of this is that I have a rock solid +50 on the GPU and +100 on the memory clock, not bad. My specific card has an ASIC score or 85.4% which is damn fine, I didn’t see the GPU temps rise above 71C which is nice. The final overclocking looks like this:
So what now?
GTX 900 Series
After having had a good look over the 900 series it’s a no-brainer to decide to upgrade. The decision is then between the 980 and the 970. Since the last time I made a decision on a graphics card I had cash to burn and so blindly bought the card I wanted and the cost definitely came second in the decision process. Hence I bought the 580 Classified and was a very happy shopper. The problem was I bought it thinking that I would be buying another in order to then go SLI and have a frankly epic amount of the GPU horsepower to call on when needed. The problem is that cash comes and goes and I never found myself able to justify shelling out an additional £550 for a second 580 Classified.
It just never happened …
Considering that would have been an investment of £1100 on GPUs alone I always backed off from clicking that “buy it now” button. Only when I was doing some intensive video work did I feel that I’d like the extra grunt. Anyway, I’m not making that mistake again and so I’m going to buy both cards at the same time this time and do the SLI immediately. To that end I’ve decided to not go to the bleeding edge of single GPU insanity and instead rein in the spending. That said the sheer processing grunt in a dual GTX 970 setup in SLI configuration will be mind-bending. The most incredible thing is that two 970 will come in at around or a little more than a single 980. It’s just a total no-brainer.
I really haven’t even thought about which 970s to get yet as in reality I don’t have the cash right now and also I’m waiting to see what is announced over the next few months. Waiting for the 8Gb GDDR5 cards is also a no-brainer. Since the RAM on the GPU is so important to Blender it makes sense to wait a little while until the larger RAM capacity cards hit the streets. Seeing as the 900 series 3rd party OC’d offerings appeared very near the release of the Nvidia reference cards it should follow that the larger 8Gb 3rd party cards will appear just as close to reference release.
So, 2x 8Gb GTX 970s in SLI it is then …
Over the last few weeks I’ve been playing around with MSI Afterburner and EVGA Precision x16 to see what I can meaningfully squeeze out of this card before plunging into the land of the GTX 900 series cards. I’ve had this card in my current machine since January 2012 and it has served me very well. The vast majority of it’s life has spent “idling” at the stock GPU clock of 855Mhz which to be fair is a solid overclock on the stock GTX580 specifications anyway. What little gaming I do it has munched through and it’s only since really starting to work with Blender more recently that I have started to experience that the GTX580 is starting to look a little long in the tooth.
The 900 series cards were announced in September 2014 and they really do look like they would be an immense upgrade over the GTX580 in terms of raw ppw (performance per watt). When the GTX580 Classified was released it really was one of the highest performing single GPUs money could buy (which retaining ownership of all ones organs). I needed the GPU acceleration for lots of video editing work I was doing and in that respect it was an incredible investment. Yes it was expensive (£550) but it was and still is a total workhorse of a component, it still plays recent high end games at max and pretty much deals with it all without any fuss at all which is impressive to say the least considering the GTX580 reference designs are now 5+ years old.
So anyway, I have been playing around with the Classified and thought I’d get this all written up so that when I do decide to upgrade to a 900 series card and I can rerun these benchmarks and see what my cold hard cash has bought me. The overclocking setting I have been playing around with are pretty rudimentary really as I’m simply using the software tools provided by EVGA and no special hardware for the task as I’m still far more preoccupied with stability than pure blistering speeds.
Test Machine
The host machine specifications are:
EVGA GTX 580 Classified Blender Benchmarks
The results of the various benchmarks I ran can be seen in the table below:
File
OC
Settings
Tile
Size
Processor
Time
BMW27.blend
N/A
16×16
CPU
06:10.03
BMW27.blend
N/A
256×256
CPU
08:05.94
BMW27.blend
–
128×128
GPU
02:08.91
BMW27.blend
–
256×256
GPU
01:49.18
BMW27.blend
GPU +50 Mem +50
256×256
GPU
01:43.82
BMW27.blend
GPU +50 Mem +100
256×256
GPU
01:42.05
BMW27.blend
GPU +50 Mem +100
512×512
GPU
01:42.71
BMW27.blend
GPU +50 Mem +100
16×16
GPU
33:16.59
BMW27.blend
GPU
+55 Mem +100
256×256
GPU
01:41.58
BMW27.blend
GPU
+55 Mem +125
256×256
GPU
01:40.77
BMW27.blend
GPU
+60 Mem +125
256×256
GPU
01:40.63
BMW27.blend
GPU
+70 Mem +125
256×256
GPU
01:40.20
BMW27.blend
GPU
+75 Mem +100
256×256
GPU
–
BMW27.blend
GPU
+75 Mem +125
256×256
GPU
01:40.63
cycles_bench_272.blend
–
256×256
GPU
12:01.27
cycles_bench_272.blend
GPU +50 Mem +100
256×256
GPU
11:12.48
sponza_cycles_benchmark.blend
–
256×256
GPU
10:12:55
sponza_cycles_benchmark.blend
GPU +50 Mem +100
256×256
GPU
09:47.31
The end result of this is that I have a rock solid +50 on the GPU and +100 on the memory clock, not bad. My specific card has an ASIC score or 85.4% which is damn fine, I didn’t see the GPU temps rise above 71C which is nice. The final overclocking looks like this:
GTX 900 Series
After having had a good look over the 900 series it’s a no-brainer to decide to upgrade. The decision is then between the 980 and the 970. Since the last time I made a decision on a graphics card I had cash to burn and so blindly bought the card I wanted and the cost definitely came second in the decision process. Hence I bought the 580 Classified and was a very happy shopper. The problem was I bought it thinking that I would be buying another in order to then go SLI and have a frankly epic amount of the GPU horsepower to call on when needed. The problem is that cash comes and goes and I never found myself able to justify shelling out an additional £550 for a second 580 Classified.
It just never happened …
Considering that would have been an investment of £1100 on GPUs alone I always backed off from clicking that “buy it now” button. Only when I was doing some intensive video work did I feel that I’d like the extra grunt. Anyway, I’m not making that mistake again and so I’m going to buy both cards at the same time this time and do the SLI immediately. To that end I’ve decided to not go to the bleeding edge of single GPU insanity and instead rein in the spending. That said the sheer processing grunt in a dual GTX 970 setup in SLI configuration will be mind-bending. The most incredible thing is that two 970 will come in at around or a little more than a single 980. It’s just a total no-brainer.
I really haven’t even thought about which 970s to get yet as in reality I don’t have the cash right now and also I’m waiting to see what is announced over the next few months. Waiting for the 8Gb GDDR5 cards is also a no-brainer. Since the RAM on the GPU is so important to Blender it makes sense to wait a little while until the larger RAM capacity cards hit the streets. Seeing as the 900 series 3rd party OC’d offerings appeared very near the release of the Nvidia reference cards it should follow that the larger 8Gb 3rd party cards will appear just as close to reference release.
So, 2x 8Gb GTX 970s in SLI it is then …