Overclock.net › Forums › Graphics Cards › NVIDIA › GTX570 vs GTX570 + 8800GTX Dedicated Physx: brief study
New Posts  All Forums:Forum Nav:

GTX570 vs GTX570 + 8800GTX Dedicated Physx: brief study

post #1 of 10
Thread Starter 
A few months have now passed since my previous thread (http://www.overclock.net/nvidia/7723...benchmark.html) and a couple things have changed.

1st : Just when I was ready to welcome a pair of GTX470s, my target Galaxy GTX470 'Gundam edition' were no longer available in local stores due to the imminent arrival of the GTX570. So, I now have a single GTX570.

2nd : Around this same time, one of my 8800GTX suffered an unexpected serious illness. It is currently lying in the morgue awaiting post-mortem examination. Hopefully after the autopsy, it can be resurrected.

So, now I have a GTX570 + 8800GTX to conduct my Physx tests. Hopefully this test can be a reference material for fellow members who are comtemplating adding a Dedicated Physx card to compliment a single GTX470/480/570/580.

Testing conditions:-
CPU: i7 965 at 4GHz
GPU: GTX570 at default factory pre-overclocking: 780/1560/2000. 8800GTX at default speed.
Benchamark: Mafia II Demo Built-in BM, Highest settings, Physx High, AA on, 16xAF, 1920x1200
Driver: 263.09 WHQL / 266.35 Beta
Mobo: Asus P6T6 WS Revolution. Primary card and Physx card are both in x16 speed slots.

Test 1: Apex Physx Off
Test 2: CPU doing Apex Physx
Test 3: GTX570 doing Apex Physx
Test 4: 8800GTX doing Apex Physx

Results:
Driver 263.09 WHQL (all are ave framerates)
Test 1: 55
Test 2: 26 (4GHz)
Test 3: 40
Test 4: N/A (263.09 is for GTX570/580 only and cannot recognize the 8800GTX)

Driver 266.35 Beta
Test 1: 54
Test 2: 17 (4GHz), 13 (3.2GHz)
Test 3: 38
Test 4: 43 (8800GTX's max usage = 68%)


Observations/Comments:
1. From no Physx to High Physx, framerate drops 11 (54 to 43). Is the visual Physx effect worth this framerate drop? You make your own choice.

2. Offloading the Physx from the GTX570 to the 8800GTX yields 5 fps (38 to 43) more. Quite significant percentage wise. Having a Dedicated Physx card is worth serious consideration. I note that the max usage of the 8800GTX was only 68%, meaning it still has some reserve to handle more Physx effects. But Physx coding/content varies from 1 game to another.

3. 'Open source' Apex Physx is reported to be modern multi-core-multi-thread CPU friendly. 26 fps in Test 2 under 263.09 is quite remarkable. However, it drops to 17 fps under 266.35 for some unknown reason. It drops further to 13 fps when the i7 965 runs at default 3.2GHz.

4. 266.35 Beta seems to yield a slightly lower graphic performance than 263.09 WHQL. However, it is the only current Nvidia driver to recognize both GTX5xx and other cards without any modding.

For further reading:
http://www.overclock.net/nvidia/8011...s-strange.html

http://www.overclock.net/nvidia/9046...70-vs-gtx.html

Edit 1:
I was going to do Batman AA but its demo does not contain a Benchmark and I do not have the game. But I do have Metro 2033 and I will include it as soon as I have some spare time to kill.

Edit 2:
And, let me add some info about temperatures:
1. See pic for setup: 'Top' card is GTX570, 'Bottom' card is 8800GTX, ambient 19-20C
2. Without the 8800GTX installed, the GTX570 breathes freely and idles at 33C (fan 40%)
3. After adding the 8800GTX, GTX570's temp rises to 38C
4. With the enhancement of the front Antec Spot Cool (Med setting), GTX570's temp drops 3C to 35C
5. With the enhancement of the rear 140mm fan (at 1100rpm), both cards drop 1-2C. The GTX570 idles at 33-34C. I imagine if I change to a higher speed 120mm fan, this will give even better result.

Temp conclusion:
Doing a 'push-pull' to dual graphic cards setup helps. The smaller the gap is, the bigger help will be. And, a fan with higher static pressure is preferred.

Edit 3:
Driver 266.58 WHQL
Test 1: 60
Test 2: 16 (4GHz)
Test 3: 40
Test 4: 45 (8800GTX's max usage = 69%)

14th March , 2011 Update (stock GTX285 as the Dedicated Physx card)
Driver 266.58 WHQL
Test 1: 59
Test 2: 17 (4GHz)
Test 3: 39
Test 4: 50 (GTX285's max usage = 65%)
Comments:Compare with the above numbers of 8800GTX as Dedicated Physx card, it is noted that Tests 1,2 and 3 yield similar numbers (as expected).
For Test 4, GTX285 gives 50fps vs 8800GTX's 45fps, with usage % in the same range (60-70%).

Edited by windfire - 4/4/11 at 8:28pm
post #2 of 10
very nice and helpful

test 3 is 570 alone ?...
My build :)
(15 items)
 
little sis
(12 items)
 
 
CPUMotherboardGraphicsRAM
I7 2600K @ 4.9 1.41v Z77 Gigabyte G1.Sniper 3 (Gigabyte P67 UD7 old) Gigabyte GTX580 SLI 2x4GB HyperX Fury  
Hard DriveHard DriveCoolingMonitor
crucial mx100 256GB 3x1TB WD black rasa block + 2xGTX360 rad's + 2EK waterblocks DELL P2414H 
PowerCaseMouseAudio
corsair 1200AX HAF stacker G400,Performance MX, Roccat Kone XTD beyerdynamic DT 770M 
CPUMotherboardGraphicsRAM
intel core i7 920 gigabyte x58 extreme integrated water cooling nvidia gtx285 2GB 3x2GB Kingston 1333mhz 
Hard DriveCoolingOSMonitor
2x1TB WD black XSPC 360AX win7 ultimate DELL U2311H 
PowerCaseMouseAudio
corsair AX1200 HAF-X logitech G400 audio technica ath-m50 
  hide details  
Reply
My build :)
(15 items)
 
little sis
(12 items)
 
 
CPUMotherboardGraphicsRAM
I7 2600K @ 4.9 1.41v Z77 Gigabyte G1.Sniper 3 (Gigabyte P67 UD7 old) Gigabyte GTX580 SLI 2x4GB HyperX Fury  
Hard DriveHard DriveCoolingMonitor
crucial mx100 256GB 3x1TB WD black rasa block + 2xGTX360 rad's + 2EK waterblocks DELL P2414H 
PowerCaseMouseAudio
corsair 1200AX HAF stacker G400,Performance MX, Roccat Kone XTD beyerdynamic DT 770M 
CPUMotherboardGraphicsRAM
intel core i7 920 gigabyte x58 extreme integrated water cooling nvidia gtx285 2GB 3x2GB Kingston 1333mhz 
Hard DriveCoolingOSMonitor
2x1TB WD black XSPC 360AX win7 ultimate DELL U2311H 
PowerCaseMouseAudio
corsair AX1200 HAF-X logitech G400 audio technica ath-m50 
  hide details  
Reply
post #3 of 10
Defiantly getting a Physx card.

it would have to be higher than a GTX8800 though. GTX9800+ or a GTX260.
    
CPUMotherboardGraphicsRAM
|i7 2600K| |EVGA P67 SLI| |N560GTX-Ti Hawk| Waiting on Sandy Bridge :) 
Hard DriveOSMonitorPower
Waiting on Sandy Bridge :) |Windows 7 Ulitmate 64-Bit| |Hanns-G HZ281HPB 1920x1200| |Silverstone Strider Plus 850W| 
Case
|SilverStone RV03| ETA: 04/05/11 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
|i7 2600K| |EVGA P67 SLI| |N560GTX-Ti Hawk| Waiting on Sandy Bridge :) 
Hard DriveOSMonitorPower
Waiting on Sandy Bridge :) |Windows 7 Ulitmate 64-Bit| |Hanns-G HZ281HPB 1920x1200| |Silverstone Strider Plus 850W| 
Case
|SilverStone RV03| ETA: 04/05/11 
  hide details  
Reply
post #4 of 10
Thread Starter 
Quote:
Originally Posted by ranerX3 View Post
very nice and helpful

test 3 is 570 alone ?...
Yes.
Test 3 under 263.09 has only a GTX570 installed on the mobo.
Test 3 under 266.35 has both cards installed but the Physx was assigned (via NVCP) to GTX570.


Quote:
Originally Posted by BKsMassive View Post
Defiantly getting a Physx card.

it would have to be higher than a GTX8800 though. GTX9800+ or a GTX260.
I will be able to get hold of a GTX285 sometime in March/April. It will likely to substitute the 8800GTX as Physx card.
Edited by windfire - 1/10/11 at 10:54pm
post #5 of 10
So in conclusion a single 570 is fine for Mafia II @ 1920x1200, the extra Physx card is a waste of time.

I love poorly optimised code to make CPU Physics look bad while GPU Physx look good. Nothing in Mafia 2 I haven't seen elsewhere without Physx.

I Wonder how much money Nvidia paid 2K Czech to include Physx in Mafia ii anybody know?
Edited by Battleneter - 1/10/11 at 7:56pm
Borris
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel I7 920@4GHz Gigabyte UD3 Zotac GTX580 @900 core 24/7 (950 afterburner only) 6Gig 
Hard Drive
80GB Intel SSD +2TB 
  hide details  
Reply
Borris
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel I7 920@4GHz Gigabyte UD3 Zotac GTX580 @900 core 24/7 (950 afterburner only) 6Gig 
Hard Drive
80GB Intel SSD +2TB 
  hide details  
Reply
post #6 of 10

Just to add to my above rant


Look at these amazing Physics, this guy must be running like 10 dedicated physics cards!!





Ow wait thats a very old 2001 game Red faction running on what is now primitive hardware using only a old processor and some efficient code.
Borris
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel I7 920@4GHz Gigabyte UD3 Zotac GTX580 @900 core 24/7 (950 afterburner only) 6Gig 
Hard Drive
80GB Intel SSD +2TB 
  hide details  
Reply
Borris
(13 items)
 
  
CPUMotherboardGraphicsRAM
Intel I7 920@4GHz Gigabyte UD3 Zotac GTX580 @900 core 24/7 (950 afterburner only) 6Gig 
Hard Drive
80GB Intel SSD +2TB 
  hide details  
Reply
post #7 of 10
I will try to find the link where I read this, but I believe if you do not use a dedicated PPU then the cloth simulations outside of the main character are calculated on the CPU. I wonder how that compares to if the single GPU was in charge of ALL the Physx. I guess the world will never know.
    
CPUMotherboardGraphicsRAM
Phenom II X6 1055T Asus M5A97 EVO 2 x Radeon 5870 2x4GB, 2x8GB 
Hard DriveHard DriveOSMonitor
120GB SSD 500GB WD 7300RPM x 2 Windows 7 Pro x64 2 x 23" Dell IPS 
PowerCase
750w OCZ A Desk 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
Phenom II X6 1055T Asus M5A97 EVO 2 x Radeon 5870 2x4GB, 2x8GB 
Hard DriveHard DriveOSMonitor
120GB SSD 500GB WD 7300RPM x 2 Windows 7 Pro x64 2 x 23" Dell IPS 
PowerCase
750w OCZ A Desk 
  hide details  
Reply
post #8 of 10
Thread Starter 
Quote:
Originally Posted by fball922 View Post
I will try to find the link where I read this, but I believe if you do not use a dedicated PPU then the cloth simulations outside of the main character are calculated on the CPU. I wonder how that compares to if the single GPU was in charge of ALL the Physx. I guess the world will never know.
Yes, you are correct. Apex Physx effects within Mafia II are: Clothing and Particles (smoke, debris).

If a dedicated Physx card is not used, the Clothing effect is done by the CPU.
See this (perhaps it is what you saw):
http://physxinfo.com/news/3892/mafia...th-simulation/

The 2nd bottom graph shows that framerate jumps from 31-32 to 38 when a GTS450 is used as a Dedicated Physx card, under the ''Only Clothing'' part.
This is also my experience when mine jumped from 38 to 43.

Other links that is worthy to read about Physx in Mafia II:
http://segmentnext.com/2010/08/24/ma...e-tweak-guide/
http://physxinfo.com/news/3628/mafia...x-performance/
post #9 of 10
mafia 2 is a terrible game to benchmark physx
albeit being the only recent one to feature it

there is a reason it's the only recent one with it

physx coding is terrible and outdated

especially when it's not optimized correctly and rushed together in mafia 2 , when you have to have a thread and multiple websites about how to "improve physx" by deleting cloth texture files
white phantom
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 950 GA X58 UD3R rev 2.0 zotac GTX 580 sli 6GB Corsair xms cl9 1600hz 
Hard DriveOptical DriveMonitorPower
Corsair ssd 32GB none asus 1080p 120hz, acer 1080p 120hz Thermaltake TR2 RX 850w 
Case
nzxt phantom 
  hide details  
Reply
white phantom
(13 items)
 
  
CPUMotherboardGraphicsRAM
i7 950 GA X58 UD3R rev 2.0 zotac GTX 580 sli 6GB Corsair xms cl9 1600hz 
Hard DriveOptical DriveMonitorPower
Corsair ssd 32GB none asus 1080p 120hz, acer 1080p 120hz Thermaltake TR2 RX 850w 
Case
nzxt phantom 
  hide details  
Reply
post #10 of 10
So based on your tests that's a $50-$80 card for 5fps?
I personally think this dedicated physx card thing is a joke
If you have the card and extra slot then sure go for it
But I question your sanity if you buy a $100 card just for physx
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › GTX570 vs GTX570 + 8800GTX Dedicated Physx: brief study