What is my bottleneck?

[cloudzero]

Rig1:

AMD Phenom X4 945 @ 3.4 GHz
BIOSTAR TA790GXE 790GX
1 DIMM 2GB Patriot DDR2 800
3D Studio Max Render Time: 33 seconds

Rig2:

AMD Phenom X4 810 @ 3.0 GHz
ECS GF8100VM-M3 NVIDIA GeForce 8100
1 DIMM 2GB Patriot DDR2 800
3D Studio Max Render Time: 30 seconds

Why is my 810 rig rendering faster than the 945?
Even though the 945 is clocked at a higher speed?

 

[xaira]

sumthins fishy here

 

[Upendra09]

what is the gfx on the first rig?

 

[cloudzero]

sumthins fishy here

your telling me

what is the gfx on the first rig?

both are using onboard graphics because these rigs are built for raw CPU power, so only the ram/cpu matters

 

[Mousemonkey]

It's the Nvidia effect, obviously. :lol:

 

[Reynod]

http://products.amd.com/en-us/DesktopCPUSideBySide.aspx?id=526&id=534

less cache and speed ... sounds very weird.

I bet some apps in the background are chewing up cpu cycles on the 945 ... or your mobo drivers are not happy.

Try hyjackthis and a driver update / reinstall.

 

[cloudzero]

http://products.amd.com/en-us/DesktopCPUSideBySide.aspx?id=526&id=534

less cache and speed ... sounds very weird.

I bet some apps in the background are chewing up cpu cycles on the 945 ... or your mobo drivers are not happy.

Try hyjackthis and a driver update / reinstall.

my 945 is actually the 95W version. and the rigs are not connected to the internet. so spyware is out of the question. i monitor 3dsmax's cpu usuage when it is running, and it never drops under 99%.

 

[4745454b]

Some thoughts I have. The most obvious difference between the rigs is the chipset. I had thought that the 790 was better then the 8100, but perhaps the 8100 has a better subsystem. The only other idea I have is what did AMD change between the 8xx and 9xx series? The 8xx has 4MBs of L3 cache, while the 9xx has 6. Digging deep in my memory, I recall an article about AMD increasing the latency of the L3 cache to support more memory. If true, the difference in chipset and changes in L3 speed is probably the 10% difference your seeing.

 

[Reynod]

Swap the 810 and the 945 into the other rig and run it again.

Simple way to exclude all other things.

Make sure you reset the mobo settings (remove the OC) on the Biostar mobo first.

Post the results and we can then ponder some more ...

 

[ainarssems]

Reset 945 to stock clocks and try then, could be some bug @ overclocked speed or mismatching CPU/HT/RAM speeds.

 

[cloudzero]

Swap the 810 and the 945 into the other rig and run it again.

Simple way to exclude all other things.

Make sure you reset the mobo settings (remove the OC) on the Biostar mobo first.

Post the results and we can then ponder some more ...

Rig1:

AMD Phenom X4 810 @ 3.32 GHz = +0.30GHz
BIOSTAR TA790GXE 790GX
1 DIMM 2GB Patriot DDR2 800
3D Studio Max Render Time: 33.7 seconds

Rig2:

AMD Phenom X4 945 @ 3.35 GHz = -0.02GHz
ECS GF8100VM-M3 NVIDIA GeForce 8100
1 DIMM 2GB Patriot DDR2 800
3D Studio Max Render Time: 27.4 seconds

WOW i am able to push the 810 way more with the 790GX, but it seems the ECS is still significantly faster. No way does a difference of 0.02GHz make a 6 second difference!

 

[4745454b]

Looks like its the chipset then. The 8100 constantly has turned in the better performance. I have no idea why this would be.

It could also be that the ECS > Biostar. Perhaps the Biostar has "noisy" ram traces or something like that. Does the Biostar speed up if you set the clocks back to stock? Can you test with some other program to see if its something related only to 3D studio?

 

[cloudzero]

the 8100 isn't as good as you think. maybe its just ecs, but there is no VCORE adjustment, no dram latency adjustment, no multiplier option. the only thing you can change is the FSB.

Yes, the anomaly is only in 3dsmax, but the OS and 3dsmax installation is exactly the same.

 

[4745454b]

Sorry, I didn't mean to imply that the 8100 is a better overclocker. It does seem to be the better chipset based off of what your giving us.

If the 790 has all those adjustments, its possible that you need to adjust some of them so the 790 is running at its full potential. If the 790 needs a bump in vcore and a change in dram latency, it might provide better numbers. Just because the OC is stable doesn't mean there isn't room for improvement. Also, if its only 3Dmax that shows this, there might be something else entirely going on.

 

[cloudzero]

i found a clue, some what.
if you know 3dsmax renders an image using "buckets". a bucket is a square maybe 32x32 pixels on an image. each core is assigned a bucket, and when it's done, it gets assigned another bucket. there is 4 buckets on the image at all times, until the entire image is done.

what i found is that on simple parts of the image, let say a flat surface, buckets finish very fast. it is about a fraction of a second. on the biostar 790gx, it takes a bit longer for each core to receive a new bucket when it is done with the current one.

this time delay is not noticeable to the naked eye, but when a whole image is done, it adds up to a few seconds.

is there something on the biostar 790gx that is slowing down the core switching?

 

[4745454b]

As I tried to imply earlier, it might be because its a biostar, not because of the 790 chipset. If the biostar used cheap materials, or a 4 layer PCB and not a 6 layer, then you might have some signal problems at higher HTT speeds. You didn't list your RAM speeds, so I don't know if this is the problem. (with the memory controller on the CPU, you might be having the classic sync/unsync'd issue as well.) If the CPU has to issue the memory request more then once, or if one system is running sync'd while the other isn't, they could cause one system to finish first. There's a lot of variables at work here, including whether a better 790 board would have the same problem. I wonder if anyone else can look into this, you have me curious if the 790 isn't as good as we all think it is.

 

[cloudzero]

there is no latency option in the ecs 8100, it is at 400MHz 5-5-5-12

on the biostar it is 425MHz 4-4-4-6