core or shader speed more important?

[JeteroH]

Whats more important on a 8800gtx core or shader overclock?

 

[Riccochet]

I would say core. At least in my own experiments shader overclocking had little to no real world effect on performance.

 

[sumofatguy]

I would say core as well. just wait someone will inevitably post some 3dmark scores...

 

[Wolf-R1]

I would say core as well. just wait someone will inevitably post some 3dmark scores...

Aside from the obvious what's wrong with posting 3DMark scores to illustrate this point? You've offered no empirical evidence to support your assertion. I don't see why 3Dmark wouldn't be a decent benchmark when you talking about speed differences within one card. When you're comparing different cards and different systems then I agree, 3DMark fails.

 

[Rambo]

someone run some tests on some real world things. i would do it, but i dont have time.

 

[sumofatguy]

Aside from the obvious what's wrong with posting 3DMark scores to illustrate this point? You've offered no empirical evidence to support your assertion. I don't see why 3Dmark wouldn't be a decent benchmark when you talking about speed differences within one card. When you're comparing different cards and different systems then I agree, 3DMark fails.

yes, but I just don't agree with 3dmark at all. Whats the point in paying $20 for a piece of software that supposedly tells you how good your computer will be at games. why don't you just PLAY THE GAME and see for yourself? as well, 3dmark has been shown to give waaay too much of an emphasis on CPU speed over graphics processing power, doesn't this suggest that it may favour core frequency over shader speed or vice versa? I just don't see the point when you could simply play the game that you are hoping to play and judge for yourself.

 

[Nenu]

OK, here are some Crysis timedemo figures :

625/1566/900 - Stock clocks
36.71 fps

625/1890/900 - Shaders overclocked 20%
39.30 fps

750/1566/900 - Core overclocked 20%
37.96 fps

750/1890/900 - Core and Shaders overclocked 20%
40.95


% benefits are
Shader up 20% = 7% extra performance
Core up 20% = 3.4% extra performance
Both up 20% = 11.5% better performance = +1.1% faster than the sum of both, an extra 10.5% more performance difference

So it seems you are best off clocking them both up together as far as they will both go and then clock whatever can go higher a bit more.



In case you want to know, this is the PC spec:
BFG 8800GT OC 512MB
E8400@ 4.13GHz
4GB Ram running 1:1, 918MHz
X38 mobo
Crysis 1.2 running CCC mod v1.31 at level 3, 1920x1080 res.
XP Pro 32

 

[Supertag]

It depends on the game. Crysis likes shaders.

 

[Ogilvy]

OK, here are some Crysis timedemo figures :

625/1566/900 - Stock clocks
36.71 fps

625/1890/900 - Shaders overclocked 20%
39.30 fps

750/1566/900 - Core overclocked 20%
37.96 fps

750/1890/900 - Core and Shaders overclocked 20%
40.95


% benefits are
Shader up 20% = 7% extra performance
Core up 20% = 3.4% extra performance
Both up 20% = 11.5% better performance = +1.1% faster than the sum of both, an extra 10.5% more performance difference

So it seems you are best off clocking them both up together as far as they will both go and then clock whatever can go higher a bit more.



In case you want to know, this is the PC spec:
BFG 8800GT OC 512MB
E8400@ 4.13GHz
4GB Ram running 1:1, 918MHz
X38 mobo
Crysis 1.2 running CCC mod v1.31 at level 3, 1920x1080 res.
XP Pro 32

I'll give you a shiny new nickle if you run the standard built-in ISLAND GPU benchmark @ 1680 x 1050 on HIGH with no mods and report here with the result. I want to see how much I have to gain by going to 4 GHz. Oh...Keep your 8800 GT OC'd, also.

You wouldn't have to uninstall that CCC mod to do that, would you?


*edit* Haha... I didn't realize it was you Nenu. I'm asking a lot from you tonight, eh?

 

[Nenu]

As requested, mod removed, all settings changed to High, 1680x1050 res.

I simply renamed the Autoexec.cfg file the CCC mod created to remove the mod.
The TOD (Time Of Day) part of the mod is still active but that would only slow it down so it wont give us inflated figures.
This is using the 174.53 beta driver.

TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 45.55s, Average FPS: 43.91
Min FPS: 25.59 at frame 151, Max FPS: 54.65 at frame 1003
Average Tri/Sec: 34328040, Tri/Frame: 781848
Recorded/Played Tris ratio: 1.17

!TimeDemo Run 1 Finished.
Play Time: 42.85s, Average FPS: 46.67
Min FPS: 25.59 at frame 151, Max FPS: 54.71 at frame 967
Average Tri/Sec: 36654004, Tri/Frame: 785395
Recorded/Played Tris ratio: 1.17

!TimeDemo Run 2 Finished.
Play Time: 42.82s, Average FPS: 46.71
Min FPS: 25.59 at frame 151, Max FPS: 54.71 at frame 967
Average Tri/Sec: 36682636, Tri/Frame: 785386
Recorded/Played Tris ratio: 1.17

TimeDemo Play Ended, (3 Runs Performed)

Pretty good

 

[Ogilvy]

As requested, mod removed, all settings changed to High, 1680x1050 res.

I simply renamed the Autoexec.cfg file the CCC mod created to remove the mod.
The TOD (Time Of Day) part of the mod is still active but that would only slow it down so it wont give us inflated figures.
This is using the 174.53 beta driver.

TimeDemo Play Started , (Total Frames: 2000, Recorded Time: 111.86s)
!TimeDemo Run 0 Finished.
Play Time: 45.55s, Average FPS: 43.91
Min FPS: 25.59 at frame 151, Max FPS: 54.65 at frame 1003
Average Tri/Sec: 34328040, Tri/Frame: 781848
Recorded/Played Tris ratio: 1.17

!TimeDemo Run 1 Finished.
Play Time: 42.85s, Average FPS: 46.67
Min FPS: 25.59 at frame 151, Max FPS: 54.71 at frame 967
Average Tri/Sec: 36654004, Tri/Frame: 785395
Recorded/Played Tris ratio: 1.17

!TimeDemo Run 2 Finished.
Play Time: 42.82s, Average FPS: 46.71
Min FPS: 25.59 at frame 151, Max FPS: 54.71 at frame 967
Average Tri/Sec: 36682636, Tri/Frame: 785386
Recorded/Played Tris ratio: 1.17

TimeDemo Play Ended, (3 Runs Performed)

Pretty good

I gotta get me one of them E8400's. Thanks again, Nenu!

 

[Nenu]

Hehe thanks
Yeah they are great, especially with an 8800 card.

With mine at 4.1GHz and the CCC Ultra High mod - Level 3, I can run native 1080p around 30 to 40fps, wonderful cos I just got a new big telly
I'm replaying the game and it looks awesome.

Just remembered... I did some CPU scaling tests about a week ago and found no difference in Crysis going from 4GHz to 3GHz.
That was Zero performance difference.
Other games will make heavier use of the CPU but felt you might like to know that.

 

[Ogilvy]

Hehe thanks
Yeah they are great, especially with an 8800 card.

With mine at 4.1GHz and the CCC Ultra High mod - Level 3, I can run native 1080p around 30 to 40fps, wonderful cos I just got a new big telly
I'm replaying the game and it looks awesome.

Just remembered... I did some CPU scaling tests about a week ago and found no difference in Crysis going from 4GHz to 3GHz.
That was Zero performance difference.
Other games will make heavier use of the CPU but felt you might like to know that.

Well, I get about 37 fps with my overclocked 8800 GT at those settings..maybe there's a difference btw. 3 and 4 GHz..and it's only noticeable on the benchmark. I've done some research on various systems doing the Crysis GPU benchmark and 33-39 fps. is about average on HIGH @ 1680 x 1050. According to your score, you've beaten 8800 GT and GTS SLI systems. That's pretty damn good...

http://www.overclock.net/nvidia/303427-official-crysis-benchmark-results-thread.html

http://forums.extremeoverclocking.com/t286518.html

 

[RaDeonXtreme]

interesting thread.thnx

 

[owkia]

You should also remember that as long as you don't use Rivatuner to unlink your core and shaders, the shaders will automatically scale increase as you increase core speed.

 

[Nenu]

Well, I get about 37 fps with my overclocked 8800 GT at those settings..maybe there's a difference btw. 3 and 4 GHz..and it's only noticeable on the benchmark. I've done some research on various systems doing the Crysis GPU benchmark and 33-39 fps. is about average on HIGH @ 1680 x 1050. According to your score, you've beaten 8800 GT and GTS SLI systems. That's pretty damn good...

http://www.overclock.net/nvidia/303427-official-crysis-benchmark-results-thread.html

http://forums.extremeoverclocking.com/t286518.html

I'm not sure what could cause the difference other than perhaps slightly messed up Crysis game (Recent re-install of the game so doubtful), or perhaps the 4MB to 6MB CPU cache makes a difference. I wouldnt expect that big a difference though.

I wonder if running the game without CCC makes more use of the CPU, I'll do a 3GHz test again to see if it makes a difference from 4GHz.

 

[Nenu]

Well, the new results show how careful you need to be when benching as I get identical results to you now.
I found that even though I set everything to high, Object Detail was set to low and greyed out so couldnt be changed.
After resetting everything to defaults, they all were automatically set to high but this time all the settings stuck.


The new results show the correct framerates but still no difference between 3 and 4GHz CPU.

3GHz timedemo
Play Time: 52.84s, Average FPS: 37.85

4GHz timedemo
Play Time: 52.83s, Average FPS: 37.86

It looks like your current CPU is fine for Crysis and a single 8800GT

 

[Hamidxa]

It depends on the game. Crysis likes shaders.

Precisely.

Games that are really shader-intensive will benefit more so from a shader overclock than a core overclock, Crysis being one such game.

 

[Hamidxa]

You should also remember that as long as you don't use Rivatuner to unlink your core and shaders, the shaders will automatically scale increase as you increase core speed.

So you mean to say if I use NTune to overclock my core, then my shaders will automatically be overclocked as well, by the same percentage?

Not that I don't believe you, but could you provide a link please.

 

[Nenu]

So you mean to say if I use NTune to overclock my core, then my shaders will automatically be overclocked as well, by the same percentage?

Not that I don't believe you, but could you provide a link please.

Yep unless you use Rivatuner to unlink them, they will be linked and do clock together.

 

[owkia]

So you mean to say if I use NTune to overclock my core, then my shaders will automatically be overclocked as well, by the same percentage?

Not that I don't believe you, but could you provide a link please.

Grab GPU-Z and you can see the proof for yourself. Look at the shader speed GPU-Z shows with the core at stock speeds. Then OC the core and look at the shader speeds again.

Or you can use Rivatuner instead of NTune to OC. Keep the core and shader linked. As you raise the core speed you'll see the shader bar increase right along with it.

 

[Hamidxa]

Grab GPU-Z and you can see the proof for yourself. Look at the shader speed GPU-Z shows with the core at stock speeds. Then OC the core and look at the shader speeds again.

Or you can use Rivatuner instead of NTune to OC. Keep the core and shader linked. As you raise the core speed you'll see the shader bar increase right along with it.

Yep unless you use Rivatuner to unlink them, they will be linked and do clock together.

Cool.
Learn something new everyday.