Overclock.net › Forums › Graphics Cards › NVIDIA › PhysX - so confused (dedicated card)
New Posts  All Forums:Forum Nav:

PhysX - so confused (dedicated card)

post #1 of 9
Thread Starter 
I'm really lost here. Why is it that a 470 + a dedicated 8800GT has worse PhysX performance then a 470 alone?

Let me explain, GPU usage on the 470 was around 45% and the PhysX card GPU usage was 50%. This was on the particle fluid demo. ~24FPS. CPU usage was also around 30% less.

Here is what I noticed with just the 470 (with no dedicated card for PhysX):

GPU usage was ~72%
CPU usage was around 32% higher
FPS was double what I got had with the 8800GT dedicated for PhysX

---

I understand that the 470 is a more powerful card, but if the GPU usage was only around 50% on the 8800GT, doesn't that mean PhysX performance should be the same?
Yummy Munchies
(13 items)
 
  
CPUMotherboardGraphicsRAM
i5-2500k MSI P67A-C43 (B3) EVGA 570 G.SKILL Ripjaws X Series 1600 2x2GB 
Hard DriveOptical DriveOSMonitor
Seagate Barracuda 7200.10 ST3250410AS LG WH10LS30 Windows 7 2560x1440 
PowerCase
Rosewill Green Series RG530-2 530W Antec Three Hundred Illusion 
  hide details  
Reply
Yummy Munchies
(13 items)
 
  
CPUMotherboardGraphicsRAM
i5-2500k MSI P67A-C43 (B3) EVGA 570 G.SKILL Ripjaws X Series 1600 2x2GB 
Hard DriveOptical DriveOSMonitor
Seagate Barracuda 7200.10 ST3250410AS LG WH10LS30 Windows 7 2560x1440 
PowerCase
Rosewill Green Series RG530-2 530W Antec Three Hundred Illusion 
  hide details  
Reply
post #2 of 9
The 470 is being bottlenecked by the 8800GT. That GPU is not fast enough to keep up with the 470.
Leopold Jr.
(15 items)
 
  
CPUMotherboardGraphicsRAM
i7 930 Gigabyte GA-X58A-UD3R MSI GTX 580 Lightning Corsair Dominator 12GB (3 x 4GB) 
Hard DriveHard DriveOSMonitor
Samsung 830 Series 256GB WD Caviar Black 1TB, Samsung F3 2 x 1TB Windows 7 Ultimate x64 Dell U2713H 
MonitorKeyboardPowerCase
Dell U2311H Nighthawk X8 Custom Backlit - Cherry MX Brown Cooler Master 850W Cooler Master HAF 932 
MouseAudio
Logitech G5 Swan T200b 
  hide details  
Reply
Leopold Jr.
(15 items)
 
  
CPUMotherboardGraphicsRAM
i7 930 Gigabyte GA-X58A-UD3R MSI GTX 580 Lightning Corsair Dominator 12GB (3 x 4GB) 
Hard DriveHard DriveOSMonitor
Samsung 830 Series 256GB WD Caviar Black 1TB, Samsung F3 2 x 1TB Windows 7 Ultimate x64 Dell U2713H 
MonitorKeyboardPowerCase
Dell U2311H Nighthawk X8 Custom Backlit - Cherry MX Brown Cooler Master 850W Cooler Master HAF 932 
MouseAudio
Logitech G5 Swan T200b 
  hide details  
Reply
post #3 of 9
I would think that it's plenty fast, the 8800 is much better than a lower end GT 2XX series (210, 220) etc. I remember people suggesting these for PhysX cards all of the time a few months ago.

As to the problem, I'm not sure. Is it possible that with 2x Graphics cards your board is changing to x8 or x4 on the PCIe slots?
Silent Folder
(14 items)
 
  
CPUMotherboardGraphicsRAM
4770k GA-Z87X-UD4H 7970 2x4GB Samsung Low Voltage 
Hard DriveHard DriveCoolingOS
OCZ Agility 3 120GB OCZ Agility 3 120GB Custom Water Loop Windows 7 Professional 
MonitorPowerCaseMouse
EQ276W PCP&C 850W Modular H630 G500 
Mouse PadAudio
Ratpadz X-Fi Titanium HD 
  hide details  
Reply
Silent Folder
(14 items)
 
  
CPUMotherboardGraphicsRAM
4770k GA-Z87X-UD4H 7970 2x4GB Samsung Low Voltage 
Hard DriveHard DriveCoolingOS
OCZ Agility 3 120GB OCZ Agility 3 120GB Custom Water Loop Windows 7 Professional 
MonitorPowerCaseMouse
EQ276W PCP&C 850W Modular H630 G500 
Mouse PadAudio
Ratpadz X-Fi Titanium HD 
  hide details  
Reply
post #4 of 9
Thread Starter 
Not sure Matt, I've never done SLI or anything with multiple GPU's before. This is the board I'm using:

http://www.newegg.com/Product/Produc...82E16813130265

Since it's designed for crossfire, it should at least have x8 with both slots being used.
Yummy Munchies
(13 items)
 
  
CPUMotherboardGraphicsRAM
i5-2500k MSI P67A-C43 (B3) EVGA 570 G.SKILL Ripjaws X Series 1600 2x2GB 
Hard DriveOptical DriveOSMonitor
Seagate Barracuda 7200.10 ST3250410AS LG WH10LS30 Windows 7 2560x1440 
PowerCase
Rosewill Green Series RG530-2 530W Antec Three Hundred Illusion 
  hide details  
Reply
Yummy Munchies
(13 items)
 
  
CPUMotherboardGraphicsRAM
i5-2500k MSI P67A-C43 (B3) EVGA 570 G.SKILL Ripjaws X Series 1600 2x2GB 
Hard DriveOptical DriveOSMonitor
Seagate Barracuda 7200.10 ST3250410AS LG WH10LS30 Windows 7 2560x1440 
PowerCase
Rosewill Green Series RG530-2 530W Antec Three Hundred Illusion 
  hide details  
Reply
post #5 of 9
You can download GPU-Z and it will tell you at what bandwidth your cards are running at just to make sure.

http://www.techpowerup.com/gpuz/
post #6 of 9
I know I've seen countless benches of this. There comes a point where it is better to just do the PhysX on the CPU, but 8800GTs and 9800GT i think are often considered one of the better cards to use as stronger cards don't provide much more benefit.

Here is one one OCN from a couple days ago
Edited by Vagrant Storm - 5/26/11 at 12:51pm
My System
(16 items)
 
  
CPUMotherboardGraphicsGraphics
2600K @ 4.7GHz Asus P8P67 B3 GTX 580 EVGA Hydro Copper 2 GTX 580 EVGA  
RAMHard DriveOptical DriveOS
4x4GB G.Skill Ripjaws bunch of 'em Blu-Ray For movies Windows 8.1 
MonitorMonitorKeyboardPower
24.4" Hans G HH251 X2 Yamakasi DS270  Blah 1000watt Super Flower 
CaseMouseMouse PadAudio
Built into Desk Microsoft SideWinder X8 Comfy one... Creative Extreme Gamer 
  hide details  
Reply
My System
(16 items)
 
  
CPUMotherboardGraphicsGraphics
2600K @ 4.7GHz Asus P8P67 B3 GTX 580 EVGA Hydro Copper 2 GTX 580 EVGA  
RAMHard DriveOptical DriveOS
4x4GB G.Skill Ripjaws bunch of 'em Blu-Ray For movies Windows 8.1 
MonitorMonitorKeyboardPower
24.4" Hans G HH251 X2 Yamakasi DS270  Blah 1000watt Super Flower 
CaseMouseMouse PadAudio
Built into Desk Microsoft SideWinder X8 Comfy one... Creative Extreme Gamer 
  hide details  
Reply
post #7 of 9
Here's the thing: with the Fluidmark bench, the rendering piece of the equation is SO easy that your rendering card really has almost nothing else to do aside from the physX calcs.

So what you're effectively doing in the tests you're conducting is comparing the physX performance of the 470 to the physX performance of the 8800GT ... Obviously, the 470 is going to kick the 8800GT's rear in a head-to-head matchup. Do you follow my logic here?

I would try doing some game benches, like Mafia 2, Batman AA (both games have demo's w/benchmarks), and Metro2033. Run them maxed out, so that the main card has a lot of work to do aside from the physX ... that's when you'll see the value of the dedicated card

I understand your confusion, esp. w/regards to the 8800GT only running at 50%, but I think the bottom-line is ... Fluidmark is not an accurate bench in terms of calculating the value of a dedicated physX card. I believe that the results you're seeing in your tests are not reflective of real-life gaming performance.

AFA the 50% thing goes, that number may just not be 'accurate' in terms of expressing how much of the cards 'physX power' is available. The usage number may well be calibrated based on using the card as a GPU, not a PPU ... ergo, 50% 'gpu-usage', when a card is doing physX, is actually equal to '100% physX usage'.

Now, that's totally just guess on my part, but it wouldn't shock me if the 'scale' of GPU usage is inaccurate when the card is used as dedicated physX.
Edited by brettjv - 5/26/11 at 6:17pm
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
xeon X5675 6-core @ 4.1ghz (1.29v, 20x205 +ht ) rampage iii extreme msi rx470 gaming X (the $159 budget king) 3 x 2gb corsair xms3 pc12800 (9-9-9-24-1T@1600MHz) 
Hard DriveOptical DriveCoolingOS
hynix 250gb ssd (boot), 2tb deskstar (apps),1tb... plextor px-712sa - still the best optical drive... corsair h8o v2 aio W10 home 
MonitorPowerCaseAudio
asus vw266h 25.5" (1920x1200) abs sl (enermax revolution) * single 70A rail 850w silverstone rv-03 XFi Titanium 
  hide details  
Reply
post #8 of 9
Quote:
Originally Posted by brettjv View Post
The rendering piece of the equation is SO easy in Fluidmark that the results you see from your tests are not going to be reflective of real-life gaming experiences.
Exactly this.

When you are benchmarking with Fluidmark, there is so little in the way of graphics rendering that your card is ONLY doing PhysX processing. This represents the performance differential between the 8800GT and GTX470, which is why you see much lower performance with the 8800GT as a "dedicated" PhysX card.

However, once you are gaming, with the GTX470 doing all the graphics work, the required power for PhysX does not subtract (much) from the performance of the primary card while the secondary is utilized for PhysX (which creates more details for the primary card to render). This way, your overall performance is improved with the 8800GT dedicated to PhysX, rather than having your GTX470 do both tasks, which would severely limit your graphics and PhysX performance combined.
    
CPUMotherboardGraphicsGraphics
Core i7-920 C0 4.00GHz HT Asus P6T Deluxe V2 Radeon HD5970 Radeon HD5870 
RAMHard DriveHard DriveHard Drive
12GB Mushkin DDR3-1600 Samsung Spinpoint F4 HD204UI Samsung Spinpoint F2 HD154UI Samsung Spinpoint F1 HD103UJ 
Hard DriveHard DriveOptical DriveCooling
Samsung Spinpoint F3 HD103SJ Samsung 840 Pro SSD Samsung 22X DVDRW ThermalRight Venomous-X 
OSMonitorKeyboardPower
Win7Ultimate x64 Asus VH222H Micro Innovations Antec TruePowerQuattro 850W 
CaseMouseOther
Silverstoned FT01 BW Logitech M100 Logitech G25 
  hide details  
Reply
    
CPUMotherboardGraphicsGraphics
Core i7-920 C0 4.00GHz HT Asus P6T Deluxe V2 Radeon HD5970 Radeon HD5870 
RAMHard DriveHard DriveHard Drive
12GB Mushkin DDR3-1600 Samsung Spinpoint F4 HD204UI Samsung Spinpoint F2 HD154UI Samsung Spinpoint F1 HD103UJ 
Hard DriveHard DriveOptical DriveCooling
Samsung Spinpoint F3 HD103SJ Samsung 840 Pro SSD Samsung 22X DVDRW ThermalRight Venomous-X 
OSMonitorKeyboardPower
Win7Ultimate x64 Asus VH222H Micro Innovations Antec TruePowerQuattro 850W 
CaseMouseOther
Silverstoned FT01 BW Logitech M100 Logitech G25 
  hide details  
Reply
post #9 of 9
(A) On FluidMark
GTX470 GPU: 45%
8800GT PPU: 50%
FPS: 24
CPU: 30%

GTX470 GPU and PPU: 72%
FPS: 48
CPU: 32%

Brief recap of what has been mentioned:
a. CPU usage is not much relevant

b. FPS got doubled when 470 did both jobs (GPU and PPU).
From one perspective, the 8800GT is not doing its Physx job fast enough and it is holding back the 470, thus yielding a lower FPS.

On another perspective, it is the 470 which has such an extremely easy job of its own (evidenced by 45% GPU) that allows him the luxury opportunity to wait for the 8800GT.

c. This is not a common situation in actual gaming unless (eg) an 8600 is used as a Physx card.


Member Kush had also recently ran Fluidmark with his GTX480 and 9800GTX+.
(http://www.overclock.net/nvidia/1024...omputer-2.html)
Interestingly, given the actual settings are different, his result is not as extreme as the FPS being doubled.
His result:-
FPS: 31 (480 as GPU and 9800GTX+ as PPU)
FPS: 39 (480 as GPU and PPU)


(B) On % Usage figure

I am guessing the same thing as brettjv on the accuracy (or rather, the inaccuracy) of the GPU % usage number.

I had tested an 8800GTX and a GTX285 as dedicated Physx card along with a GTX570 as the main card for Mafia II Game benchmark. Obviously, GTX285 is a much stronger GPU than an 8800GTX (240 shaders vs 128 shaders).

Maximum GPU usage for the 8800GTX is 60-70%. This number led me to think that the card is not fully stressed by those Physx effects and has abundant power reserved to do more.

But when I switched to a GTX285, the maximum GPU usage % figure remained in the same range of 60-70%. If my thinking above was correct, this number should be lower. But it is not.

So, I reckon brettjv is correct about the GPU usage % figure is not calibrated properly to reflect PPU activity accurately.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › PhysX - so confused (dedicated card)