I've actually done that believe it or not.
1. No change (with efb to texture, efb to ram might see a difference)
2. Depends on the settings. My GTX 260 is able to use 1x and 2x efb scales with 4xSSAA without any difference in speed. 3x efb scale + 4xSSAA shows a drop to about half. My 8800GT sees a difference between 1x efb with 4xSSAA and 2x with 4xSSAA, but small (still fullspeed).
3. HUGE DIFFERENCE! My old pentium D could never achieve playable speed while my core 2 quad laughs.
(01-06-2011, 10:30 AM)NaturalViolence Wrote: [ -> ]Quote:And none of these cards managed to achieve it actually.
I would disagree with that. All of these cards except the 8600gt manage to pull over 60 fps with a 3x efb scale. And bottlenecks from memory bandwidth and cpu throughput vary a lot. You are testing one of the few places where not a lot of cpu throughput is needed, normally achieving 200 fps with framelimiting off is not achievable for most heavier games regardless of how powerful your gpu is.
Quote:For example you can hear all over the place that Mario Galaxy 2 is demanding and you need a highly clocked cpu, where in reality it is a gpu demanding game (there are a lot of other games that are way more cpu demanding than Mario Galaxy 2 actually).
I would be interested to see you prove this. I have tested SMG2 pretty thoroughly and my results do not line up with that.
Out of the games I have:
TP: Usually bottlenecked by memory bandwidth.
WW: CPU heavy
SMG2: CPU heavy and GPU heavy but especially CPU.
NSMB: Not very heavy on either.
DKCR: CPU heavy
MP: GPU heavy and CPU heavy, depends on the area.
Regardless of playable speed in this case they do bring down the cpu performance, which in a game that nears 60 fps or 30fps with efb scale set to 1 matters a lot. Just compare the first and third case, or the second and third in the first post and tell me that cpu overclock seems meaningful with such gpu's. In smg2 personally i get worse results with my core i7 @ 3,2 ghz compared to a core 2 quad at 2,66 ghz when efb scale is set to 3x or integral. The meaning is to find from which point onwards your cpu isn't restrained by your gpu using the program at all. Telling someone to overclock his cpu or buy a new one when the above hurdle isn't passed is a bit pointless to me. Finding the golden ratio of cpu to gpu power if you will is important imo.
Got my 460 GTX now and I did a test with new results:
E6300 @1.86Ghz OC'ed to 3.73Ghz
2x2GB DDR2 1066Mhz Trident G.Skill ram
Nvidia 460 GTX MSI 768MB Core 751Mhz | Memory 900Mhz | Shader 1502Mhz (stock settings)
***Desktop resolution is 1920x1080 on 42'in Samsung TV with Windows Classic theme for maximum fps. DVI-to-HDMI adapter with audio and video on one cable
1x 159
2x 146
3x 117
Quote:Regardless of playable speed in this case they do bring down the cpu performance, which in a game that nears 60 fps or 30fps with efb scale set to 1 matters a lot. Just compare the first and third case, or the second and third in the first post and tell me that cpu overclock seems meaningful with such gpu's. In smg2 personally i get worse results with my core i7 @ 3,2 ghz compared to a core 2 quad at 2,66 ghz when efb scale is set to 3x or integral. The meaning is to find from which point onwards your cpu isn't restrained by your gpu using the program at all. Telling someone to overclock his cpu or buy a new one when the above hurdle isn't passed is a bit pointless to me. Finding the golden ratio of cpu to gpu power if you will is important imo.
I agree but this game/spot does not seem to accurately depict the amount of cpu throughput needed by dolphin in most circumstances. Thus it will artificially inflate the importance of the gpu. I suggest you use SMG or SMG2 next since it's heavy on both the gpu and cpu.
(01-07-2011, 04:25 PM)NaturalViolence Wrote: [ -> ]Quote:Regardless of playable speed in this case they do bring down the cpu performance, which in a game that nears 60 fps or 30fps with efb scale set to 1 matters a lot. Just compare the first and third case, or the second and third in the first post and tell me that cpu overclock seems meaningful with such gpu's. In smg2 personally i get worse results with my core i7 @ 3,2 ghz compared to a core 2 quad at 2,66 ghz when efb scale is set to 3x or integral. The meaning is to find from which point onwards your cpu isn't restrained by your gpu using the program at all. Telling someone to overclock his cpu or buy a new one when the above hurdle isn't passed is a bit pointless to me. Finding the golden ratio of cpu to gpu power if you will is important imo.
I agree but this game/spot does not seem to accurately depict the amount of cpu throughput needed by dolphin in most circumstances. Thus it will artificially inflate the importance of the gpu. I suggest you use SMG or SMG2 next since it's heavy on both the gpu and cpu.
As a combination of the two it would be good, maybe in the future. Next i was thinking of a cpu heavy benchmark, haven't really decided which game to be yet.
Btw thanks tuanming for the info, the 1st post has been updated again with the info about the 460GTX 768MB.
Just for fun, I clocked my speed to 2.66Ghz to see how the 460 GTX 768 192-bit stack up against the HD4870 1GB 256-bit.
E6300 @2.66Ghz 460 GTX 768 results:
1x 117
2x 108
3x [color=#FF0000]102[/color]
E6750 @2.66Ghz HD 4870 1GB results:
1x 109 fps
2x 111 fps
3x 82 fps
If my memory serve me right, I believe the E6750 has more cache than the E6300 series. Conclusion? The 460 GTX is the better card, especially at 3x efb scale
(01-08-2011, 12:48 AM)tuanming Wrote: [ -> ]Just for fun, I clocked my speed to 2.66Ghz to see how the 460 GTX 768 192-bit stack up against the HD4870 1GB 256-bit.
E6300 @2.66Ghz 460 GTX 768 results:
1x 117
2x 108
3x [color=#FF0000]102[/color]
E6750 @2.66Ghz HD 4870 1GB results:
1x 109 fps
2x 111 fps
3x 82 fps
If my memory serve me right, I believe the E6750 has more cache than the E6300 series. Conclusion? The 460 GTX is the better card, especially at 3x efb scale
Hmm, indeed, this time the fps drop is only 15% while previously was 25%. The difference in cpu power plays a role to widen the gap it seems, not by much considering the actual difference in power here but it does. Maybe i should put similar powerful processors in the same boat or delete my conclusions altogether and let the numbers speak, i'll think about it.
Use microsoft excel to plug in all the data and generate a line graph where each gpu/cpu combination is a different colored line (x axis is efb scale y axis is framerate). This will help you interpret the data. Or if you don't like microsoft you can use the openoffice equivalent. I will certainly join in on your cpu heavy benchmark if you ever get around to it, I'm much more interested in those results. By the way when you do decide to do that I suggest you have everyone post their results with efb to texture and efb to ram since efb to ram creates a memory read/write bottleneck (and of course make sure everyone has AA, pixel depth, pixel lighting, and efb scaled copy all off to reduce gpu load).
Core i7 930@3.52GHZ, Uncore@3.2GHZ, RAM@1600MHZ
5870 1GB, Cat AI@HQ
1x EFB
NoAA 232
4x 233
9x 202
2x EFB
NoAA 242
4x 129
9x 60
3x EFB
NoAA 206
4x 62
9x 26
Pics
(01-08-2011, 12:35 PM)NaturalViolence Wrote: [ -> ]Use microsoft excel to plug in all the data and generate a line graph where each gpu/cpu combination is a different colored line (x axis is efb scale y axis is framerate). This will help you interpret the data. Or if you don't like microsoft you can use the openoffice equivalent. I will certainly join in on your cpu heavy benchmark if you ever get around to it, I'm much more interested in those results. By the way when you do decide to do that I suggest you have everyone post their results with efb to texture and efb to ram since efb to ram creates a memory read/write bottleneck (and of course make sure everyone has AA, pixel depth, pixel lighting, and efb scaled copy all off to reduce gpu load).
First post is updated with all the new data, i also put an excel file and a separate graph pic to be able to visualize the data.