Overclock.net › Forums › Industry News › Hardware News › [Anand] A Look At Triple-GPU Performance And Multi-GPU Scaling, Part 1
New Posts  All Forums:Forum Nav:

[Anand] A Look At Triple-GPU Performance And Multi-GPU Scaling, Part 1 - Page 13

post #121 of 149
Quote:
Originally Posted by grunion View Post
Where are you getting those numbers?

Because 120 to 169 is not 77% and 169 to 208 is not 60%
That's the percentage of the base fps of a single 580. Each card has diminishing returns.


Quote:
Originally Posted by luke997 View Post
If you read my post, I said:
"There will be no higher res than 1920x1080 in 3D on a single screen"

5760x1080 is 3 screens, each still has the limits due to the reasons I've mentioned.
What I asked was, why did you say you had 3D running at 5760x1080?

Quote:
Originally Posted by luke997 View Post
Frame buffer usage is game dependant, some will use less, some more - seems though more advanced graphics in the game = more usage (which hurts in the game I play like JC2 or Metro)
I haven't run into any problems myself unless I crank the AA up to 32XCSAA for example in BC2. Even still, I can play perfectly fine with well more than 60FPS on average. The review I linked to showed no problems at 4xAA:


AvP was the only one that showed signs of trouble; even that only lost by 1fps to the 2GB card:


The 570s did lose in Crysis Warhead, but didn't exhibit any memory limiting problems.

Quote:
Originally Posted by luke997 View Post
I also believe that quad SLI is not very practical.
Yes, you get better benchmark results with 4 and yes you get a little better performance in few games, but not in most.

By the way, it would be really good to see benchmarks of 2,3,4 SLI in 1.5GB and 3GB in multi monitor resolutions (2D and 3D) by some review site...
Look at CallsignVega's thread. He tested 2-way SLI/Xfire, Tri-fire and 4-way SLI: http://www.overclock.net/graphics-ca...eyefinity.html



His build thread has some videos of gameplay in the second post: http://www.overclock.net/water-cooli...tx580-sub.html
Edited by PoopaScoopa - 4/7/11 at 8:39am
post #122 of 149
Quote:
Originally Posted by PoopaScoopa View Post
That's the percentage of the base fps of a single 580. Each card has diminishing returns.
Still doesn't add up.
Maybe I have a severe case of the stupids this morning, or my calculator needs new batteries.
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
post #123 of 149
Quote:
Originally Posted by grunion View Post
Still doesn't add up.
Maybe I have a severe case of the stupids this morning, or my calculator needs new batteries.
Sorry, math teachers always complained i never wrote down my work.

1-way = 63.9fps = 100%
2-way = 120.2fps | 56.3fps gain = 88% of single 580's 63.9fps
3-way = 169.5fps | 49.3fps gain = 77% of single 580's 63.9fps
4-way = 208.1fps | 38.6fps gain = 60% of single 580's 63.9fps

Each additional card adds less and less of a performance boost. If you prefer the other 1.0 , 1.88 , 2.65 , 3.25 comparison you can just add the percentages together.



It just seems that most review sites like the one in the OP fail to use fast enough CPUs for 3-way/4-way comparisons. Here's some more reviews of tri-sli:

This game could definitely use a 2600K @ 5Ghz to help:





AvP is more GPU limited:







Edited by PoopaScoopa - 4/7/11 at 8:39am
post #124 of 149
Quote:
Originally Posted by PoopaScoopa View Post
Sorry, math teachers always complained i never wrote down my work.

1-way = 63.9fps = 100%
2-way = 120.2fps | 56.3fps gain = 88% of single 580's 63.9fps
3-way = 169.5fps | 49.3fps gain = 77% of single 580's 63.9fps
4-way = 208.1fps | 38.6fps gain = 60% of single 580's 63.9fps

Each additional card adds less and less of a performance boost. If you prefer the other 1.0 , 1.88 , 2.65 , 3.25 comparison you can just add the percentages together.


That's not how I figure scaling when adding cards.



1 to 2 cards 63.9--->120.2 = 88%
2 to 3 cards 120.2--->169.5 = 41%
3 to 4 cards 169.5--->208.1 = 23%

But I do now understand what you're doing.
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
post #125 of 149
ATI's 6900 ARE the better cards for higher resolution and scale better then the GTX 500 series, There's no argument about it.

Instead of posting the average frame rates why don't the people trying to prove that Nvidia is better at extreme resolutions post screen or results showing the minimum frame rate performance as well?

I know from my own experience that if you have 2 cards that are the exact same except for VRAM size that they're will be no difference in the average frame rates at all.

The difference comes in the Minimum frame rates.

So while Nvidia cards might produce the same or better average frame rates then the ATI 6900 series the ATI 6900 produce minimum frame rates that are far superior and thus deliver a better experience.

Case in point in the screen below, GTX 580's are the same speed as the 6970's giving the false impression that they have enough VRAM as they're right up there with the ATI 6900 series cards.



Now as I said above the difference between a card that is VRAM limited compared to one that isn't is seen in the minimum frame rate. So lets take a look at it.



What a difference! The GTX 580's have gone from being neck and neck to lagging behind by 19.1fps. 3 cards lag behind 2 cards because for every GPU added the game has to texture swap game assets between 3 cards which is a lot of work.

The 2Gb VRAM will always make sure the 6900 series is the better choice for silly resolutions as they provide a smoother experience.
post #126 of 149
Quote:
Originally Posted by grunion View Post
That's not how I figure scaling when adding cards.



1 to 2 cards 63.9--->120.2 = 88%
2 to 3 cards 120.2--->169.5 = 41%
3 to 4 cards 169.5--->208.1 = 23%

But I do now understand what you're doing.
Well, it seems strange to compare the performance of just a single 3rd card being added to your system against the 2 cards already in it.

Review sites that list 1.0, 1.88, 2.65, 3.25 scaling are showing the same percentages just added together. This way you can say, if I buy another 580 I know I'll only get 77% use out of this card. If I get another, I know I'll only get 60% use out of the fourth card.

Quote:
Originally Posted by scatteh316 View Post
Instead of posting the average frame rates why don't the people trying to prove that Nvidia is better at extreme resolutions post screen or results showing the minimum frame rate performance as well?
You don't want to simply post the minimum framerate that it may of been at for a split second, what you want to do is what [H] does and show the timeline of fps to see if it's frequently dropping down or just simply had a hiccup that won't affect gameplay too much.

As an example, sometimes even ATI cards can have momentary fps drops. It wouldn't be fair to AMD either to try and make any importance about it getting 1FPS in Civ5.




That's why Average fps is more relevant.

I also don't think anyone is saying AMD doesn't scale better. I was simply pointing out that an equally priced comparison of the 570 showed better average fps at 4xAA in all resolutions up to 5760x1080.
Edited by PoopaScoopa - 4/7/11 at 9:32am
post #127 of 149
Quote:
Originally Posted by scatteh316 View Post
ATI's 6900 ARE the better cards for higher resolution and scale better then the GTX 500 series, There's no argument about it.

Instead of posting the average frame rates why don't the people trying to prove that Nvidia is better at extreme resolutions post screen or results showing the minimum frame rate performance as well?

I know from my own experience that if you have 2 cards that are the exact same except for VRAM size that they're will be no difference in the average frame rates at all.

The difference comes in the Minimum frame rates.

So while Nvidia cards might produce the same or better average frame rates then the ATI 6900 series the ATI 6900 produce minimum frame rates that are far superior and thus deliver a better experience.

Case in point in the screen below, GTX 580's are the same speed as the 6970's giving the false impression that they have enough VRAM as they're right up there with the ATI 6900 series cards.



Now as I said above the difference between a card that is VRAM limited compared to one that isn't is seen in the minimum frame rate. So lets take a look at it.



What a difference! The GTX 580's have gone from being neck and neck to lagging behind by 19.1fps. 3 cards lag behind 2 cards because for every GPU added the game has to texture swap game assets between 3 cards which is a lot of work.
The 2Gb VRAM will always make sure the 6900 series is the better choice for silly resolutions as they provide a smoother experience.
Shouldn't that be a red flag when just SLI performs better than 3-Way SLI?

Wouldn't you think there is either something wrong with the setup or with the drivers?


Here is another example

http://guru3d.com/article/geforce-gt...-sli-review/11


Quote:
And in the comparative performance chart, we can start to evaluate again. We are seeing very constant performance every time. Again, since three GPUs eat more CPU cycles than two, in a CPU bottlenecked situation that will give an advantage towards 2-way SLI.

Edited by saulin - 4/7/11 at 9:40am
My System
(13 items)
 
  
CPUMotherboardGraphicsRAM
I7 920 @ 4.3Ghz Asus P6X58D Premium Zotac GTX 480 6GB CORSAIR XMS3 7-7-7-20 
Hard DriveOptical DriveOSMonitor
2X Vertex LE 100GB SSDs/RAID-0 & 2TB Barracuda XT Samsung SH-S243D 24x DVD-Writer Windows 7 LG Flatron W2242TQ & LG Flatron W1942TQ 
PowerCase
Corsair TX750W Cooler Master Haf 932 
  hide details  
Reply
My System
(13 items)
 
  
CPUMotherboardGraphicsRAM
I7 920 @ 4.3Ghz Asus P6X58D Premium Zotac GTX 480 6GB CORSAIR XMS3 7-7-7-20 
Hard DriveOptical DriveOSMonitor
2X Vertex LE 100GB SSDs/RAID-0 & 2TB Barracuda XT Samsung SH-S243D 24x DVD-Writer Windows 7 LG Flatron W2242TQ & LG Flatron W1942TQ 
PowerCase
Corsair TX750W Cooler Master Haf 932 
  hide details  
Reply
post #128 of 149
^^ Different bench run, different iq settings, no listed min fps.

So it's pretty much not possible to compare the g3d results with anand.
Probably a different platform also.

And anand probably have their numbers reversed.
Edited by grunion - 4/7/11 at 10:00am
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
post #129 of 149
Quote:
Originally Posted by grunion View Post
^^ Different bench run, different iq settings, no listed min fps.

So it's pretty much not possible to compare the g3d results with anand.
Probably a different platform also.

And anand probably have their numbers reversed.
Same issue though, CPU bottleneck

Don't 580s still hold world records? I'm sure they scale even in 4-way SLI given the right CPU. When 2-Way SLI = faster than 3-Way SLI, there is a problem. It could be a CPU bottleneck or even a driver issue
Edited by saulin - 4/7/11 at 10:04am
My System
(13 items)
 
  
CPUMotherboardGraphicsRAM
I7 920 @ 4.3Ghz Asus P6X58D Premium Zotac GTX 480 6GB CORSAIR XMS3 7-7-7-20 
Hard DriveOptical DriveOSMonitor
2X Vertex LE 100GB SSDs/RAID-0 & 2TB Barracuda XT Samsung SH-S243D 24x DVD-Writer Windows 7 LG Flatron W2242TQ & LG Flatron W1942TQ 
PowerCase
Corsair TX750W Cooler Master Haf 932 
  hide details  
Reply
My System
(13 items)
 
  
CPUMotherboardGraphicsRAM
I7 920 @ 4.3Ghz Asus P6X58D Premium Zotac GTX 480 6GB CORSAIR XMS3 7-7-7-20 
Hard DriveOptical DriveOSMonitor
2X Vertex LE 100GB SSDs/RAID-0 & 2TB Barracuda XT Samsung SH-S243D 24x DVD-Writer Windows 7 LG Flatron W2242TQ & LG Flatron W1942TQ 
PowerCase
Corsair TX750W Cooler Master Haf 932 
  hide details  
Reply
post #130 of 149
Quote:
Originally Posted by saulin View Post
Same issue though, CPU bottleneck

Don't 580s still hold world records? I'm sure they scale even in 4-way SLI given the right CPU. When 2-Way SLI = faster than 3-Way SLI, there is a problem. It could be a CPU bottleneck or even a driver issue
Sure they do, but world records have nothing to do with game scaling.

Why don't you email anandtech and ask them if they got the min fps reversed for tri and sli?
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
Mighty-iTX
(12 items)
 
One foot in..
(17 items)
 
mATX
(12 items)
 
CPUMotherboardGraphicsRAM
i7-4770K M7-Impact Strix GTX980 Samsung 
Hard DriveCoolingOSMonitor
Crucial M4 256 Glacer 240L W8.1 K272HUL 
KeyboardPowerCaseMouse
Saitek Seasonic X-650 Prodigy MX518 
CPUMotherboardGraphicsGraphics
4770K/45/43 Maximus VI Extreme R290X R290X 
RAMHard DriveHard DriveOptical Drive
Samsung Crucial M4 64GB RAIDR LSSWM 
CoolingOSMonitorKeyboard
Cooler Master Glacer 7 Ultimate 64 SyncMaster Eclipse 
PowerCaseMouseMouse Pad
XFX 1050 BE 600T Silver MX518 Desk 
CPUMotherboardGraphicsRAM
i5-4330 Z87 Gryphon GTX690 G.Skill RipJawsX 
Hard DriveCoolingOSMonitor
WD Blacks Hyper 212 W8.1 ACER 
KeyboardPowerCaseMouse
Logitech Ultra X3 1000w Corsair 230T Orange Razer 
  hide details  
Reply
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Hardware News
Overclock.net › Forums › Industry News › Hardware News › [Anand] A Look At Triple-GPU Performance And Multi-GPU Scaling, Part 1