9600GSO questions. [96 shader]

[kelliegang]

Ok, so I've got my 9600GSO: 786MB 96 shader version, I'm running the CUDA driver recommended by the FAH website [181.20].
All settings are at manufacturer standards no overclocking and I'm on Win XP with a Pentium D 3.2ghz [which is dual cpu]
I'm currently working on project 5903 with core_14 and am getting about 2400-2800ppd which is fine with me

Running GPU-Z I often see 0% fan speed utilisation and GPU temp 60... so apparantly idle?
Is this usual or should I be setting my priority higher [currently set at idle and I only get the option for idle/low] do I have to get the systray version in order to do this or is the console client adequate?

[pmsfh2008]

I have three of these OC'ed to 690/1725/800.
They are running 60 +/- 5 degrees.
Use rivatuner to OC and set the fan.
On 5903's I am getting +/- 3100ppd.
I have the systry client set slighty higher.
Hope this helps.
Good Luck and keep on folding

Code: Select all

Project : 5903
 Core    : Unknown
 Frames  : 100
 Credit  : 1680


 -- 9600GSO1_FH1 --

 Min. Time / Frame : 7mn 09s  - 3383.50 ppd
 Avg. Time / Frame : 7mn 29s  - 3232.78 ppd
 Cur. Time / Frame : 7mn 55s  - 3055.83 ppd
 R3F. Time / Frame : 7mn 52s  - 3075.25 ppd
 Eff. Time / Frame : 7mn 46s  - 3114.85 ppd


 -- 9600GSO2_FH1 --

 Min. Time / Frame : 6mn 30s  - 3721.85 ppd
 Avg. Time / Frame : 6mn 48s  - 3557.65 ppd
 Cur. Time / Frame : 7mn 37s  - 3176.19 ppd
 R3F. Time / Frame : 7mn 37s  - 3176.19 ppd
 Eff. Time / Frame : 7mn 33s  - 3204.24 ppd


 -- 9600GSO3_FH1 --

 Min. Time / Frame : 6mn 48s  - 3557.65 ppd
 Avg. Time / Frame : 7mn 01s  - 3447.79 ppd
 Cur. Time / Frame : 6mn 56s  - 3489.23 ppd
 R3F. Time / Frame : 6mn 52s  - 3523.11 ppd
 Eff. Time / Frame : 7mn 16s  - 3329.17 ppd


 -- GTX260-216 --

 Min. Time / Frame : 3mn 34s  - 6782.80 ppd
 Avg. Time / Frame : 3mn 36s  - 6720.00 ppd
 Cur. Time / Frame : 3mn 41s  - 6567.96 ppd
 R3F. Time / Frame : 3mn 41s  - 6567.96 ppd
 Eff. Time / Frame : 3mn 47s  - 6394.36 ppd

[chungenhung]

2400-2800ppd is the norm for 9600gso at stock speeds.
GPUZ might not read the fan sensor correctly.
a 0% would mean the fan is completely not running, which i doubt would happen.
You can check your fan and see if it is turning.
and, if GPU is idle your ppd would have been 0.

[kelliegang]

The idle question was related to information I Was seeing about other cards/drivers, something along the lines of the card being idle 2-3 seconds out of every 5, doing sporadic computations...

I did however run into an occasion where I wasnt getting any work done for about 15 minutes [whilst running an mp3 on windows media player]. The 1% took 25 minutes to complete when it normally takes 10 and from the point i turned off the media player it took exactly 10 minutes to complete the 1%.

I have three of these OC'ed to 690/1725/800.

My reason for choosing the 9600GSO was the lower wattage draw, what kind of extra power consumption do you experience with your cards overclocked to the above settings?
I wouldn't mind overclocking a little if it results in a worthwhile increase in terms of points per watt.

[FEAR6655]

So it looks like the benefit of the 9600GSO over 9600GT is pretty much gone now? I get 3204PPD on p5903 with my shaders at 1850MHz.

[kelliegang]

I thought it had always been the case with the higher work units that the field evened out dramatically.

[Gormar]

So it looks like the benefit of the 9600GSO over 9600GT is pretty much gone now? I get 3204PPD on p5903 with my shaders at 1850MHz.

Not so quite. You probably use core 1.24.
With core 1.24 9600GSO OC do about 3800 PPD. Core 1.25 make it down to 3100 PPD.
Check your PPD now.

[MtM]

So it looks like the benefit of the 9600GSO over 9600GT is pretty much gone now? I get 3204PPD on p5903 with my shaders at 1850MHz.

Not so quite. You probably use core 1.24.
With core 1.24 9600GSO OC do about 3800 PPD. Core 1.25 make it down to 3100 PPD.
Check your PPD now.

Only if you 'cheated' with the drivers. I always ran cuda drivers, and with 1.25 I got the same ppd as with 1.24, give or take a very light diffrence.

The 9600gt I have is doing pretty close to the gso's on stock clocks, and if I oc it it will surpass them with ease. But, I get more errors from it then, and I think it's better to have 0 eue's then a couple as who knows for sure every wrong calculation will trigger an eue

[mikeb12]

kelliegang, you can get a little boost on your 9600gso by increasing the shader clocks.. in my experience 1728 was about the highest the my evga 9600gso would go and remain eue free. many people run 1782, but I found that the higher shader oc carried random eue's with it...

it's fairly simple. just dl precision tool, install it, click unlink and slide the shader over to 1728 and apply. you can leave the rest of the clocks (ram and core) at stock.
http://www.evga.com/PRecision/

this will give you a little better frame times and wu completion times. just watch your client closely for eue's, but all the 9600gso's I've seen will handle 1728 fine. some have problems at 1782 and some don't, all depends on the chip you lucked out with. I feel safer recommending 1728 as the safe shader oc. or even one shader notch lower at 1674. it will still be an improvement over the stock 1375 shader clocks.

don't mess with the core or ram, they'll have vrtually no effect on frame times, and ocing the ram in my experience is a good way to guarantee eue's.

[Gormar]

The 9600gt I have is doing pretty close to the gso's on stock clocks, and if I oc it it will surpass them with ease. But, I get more errors from it then, and I think it's better to have 0 eue's then a couple as who knows for sure every wrong calculation will trigger an eue

I didn't "cheated" with the driver but I use 128.08 not CUDA blessed 181.20. I've only heard about problems with 185.xx for Vista x86.
I talk from experience. I had 9600GT 512MB OC (750/900/1750) and when I change it to 9600GSO 384MB OC (650/850/1784) I've got quite nice PPD jump (800+).

[kelliegang]

Upped my shader clock to 1632 [EVGA precision tool said it wasnt stable above this increment of 96, probably due to my power supply being underpowered], am now seeing up to 3000ppd [averaging at 2750ppd] out of the 9600gso on project 5904

Just updated fahmon's project list with the re-benchmarking that occurred and I'm looking at 3150 ppd averages now, with low of 3020 and high of 3412...

Thanks for the info on how to overclock, no EUEs or anything so far either [except when I was testing the clock speeds which i stupidly did when I was already at 66% on a 5904 project].

[FEAR6655]

So it looks like the benefit of the 9600GSO over 9600GT is pretty much gone now? I get 3204PPD on p5903 with my shaders at 1850MHz.

Not so quite. You probably use core 1.24.
With core 1.24 9600GSO OC do about 3800 PPD. Core 1.25 make it down to 3100 PPD.
Check your PPD now.

Checked the log, and I'm using Core 1.25. So then my statement looks to be correct.

[MtM]

The 9600gt I have is doing pretty close to the gso's on stock clocks, and if I oc it it will surpass them with ease. But, I get more errors from it then, and I think it's better to have 0 eue's then a couple as who knows for sure every wrong calculation will trigger an eue

I didn't "cheated" with the driver but I use 128.08 not CUDA blessed 181.20. I've only heard about problems with 185.xx for Vista x86.
I talk from experience. I had 9600GT 512MB OC (750/900/1750) and when I change it to 9600GSO 384MB OC (650/850/1784) I've got quite nice PPD jump (800+).

Systems differ, easy as that. Not saying what you seen is not true, but what I seen is diffrent, maybe you're better then me with finding the right os/driver or something else, but I seen some wu's with no diffrence at all between them and some with about 500+ diffrence it just depended on wu's and drivers. The gt is now on xp32 the gso's on vista x64, and often they ran diffrent drivers. I seen reports of gso's doing much better then mine, even when I ran ~1800 shaders and oc'd core/mem, and I seen slower reports and all I can think is 'well I'll be darned, I'm either doing something wrong or they use magic'

[YashBudini]

I only OC the shaders, and I run 1800 all the time. Increasing the other stuff doesn't yield much.

[mikeb12]

I only OC the shaders, and I run 1800 all the time. Increasing the other stuff doesn't yield much.

1800 yields 1782 on the shader oc. the next stop is 1836... if you set it in between these, the actual shader clock just rounds up or down to the appropriate stop... read this and it will explain better http://forums.pureoverclock.com/showthread.php?t=4258 the user has a chart of his set vs actual shader clocks as an example...

it's misleading because appears like you can set the shader at 1800 with your slider, but if you view the actual clocks in the real time graphs in RT or prceision tool, then you'll see the rounding effect.. they round up/down in 54hz increments.

just for reference, what I've found the stops at..
1728-1782-1836-1890-1944

then just leave the core and ram at stock...

1728 was my safe 9600gso shader that never had any eue's on any of the wu types, it could run for weeks and weeks with no eue's in the log at 1728.
1782 was my max shader oc, that had an occasional eue on some wu's...not eue to pause type eue's, the only way I'd find them is with MtM's fahwatch that displays all your eue codes in the fahlog.txt file.. alot of people say they have no eue's, but don't actually search the entire log, that's just their opinion from keeping an eye on fahmon. which is just brief windows of time samples. fahmwatch will quickly scan your entire log files and reveal eue's you were not aware of.. http://www.mtm78.nl/fahwatch/publish.htm