Same GPU, same clocks... different performance?

This is just a curiosity of mine, something I've noticed lurking around these forums.

I have the laptop in my sig, with a HD5650 GPU. Sony have underclocked it in comparison with most other manufacturers using the same GPU, but even when it's set with the same clocks it seems to be performing at lower levels than them. Examples:

Envy14 450/800: ~6800
Sony 450/800: ~6300

Lenovo y460 550/800: ~7300
Acer 5740G 550/800: ~7300
Acer 3820TG 550/800: ~7300
Sony 550/800: ~6800

I'm just curious why, with the same GPU and the same clocks, it would be producing lesser scores. The Sony is equal or higher than the other machines in other relevant specs (RAM, CPU). Other users with the Sony are getting the same scores, and the scores for the other machines are generally consistent too, so there's obviously something fundamental with the Sony that's causing the lower scores.

 

just overclock that

 

perhaps a cheaply made mobo?

different bios?

there are many reasons. usually 200-300 points in 3dmark06 is marginal error.

 

undervolted.

getting docked for a bad cpu *notices it's a 520m*

ok not the cpu

*shrugs*

 

200-300 would be a marginal error where the scores are fluctuating, like on desktops where configs are very different from computer to computer.. but 500 where they're all very clear and consistent on spec machines suggests that something's different. It could be a weedy vbios, but not much way of telling if it is, hmm

It is undervolted, but voltage in itself doesn't directly affect performance. It restricts stability and how high you can clock it, but at the same clocks it makes no difference.

*shrugs too*

 

It could be due to the Drivers being used as well.

 

Because Sony is a mess.
I've seen one Sony Vaio where you had to install their proprietary software for Fn+brightness to work.

 

I don't think it's the drivers. The score's been the same with 10.2, 10.3, 10.5 and 10.6 - which is the latest and the same everyone else is using.

Uhm, they're all like that.

 

As with any type of hardware, the parts are only half the result. The design and implementation is the other half. This is probably it.

 

Most GPU specs allow the manufacturers a choice of memory types...could be one with DDR3 and one with GDDR3 (either type can be used in a 5650 according to NotebookCheck).

 

You can't use DDR3 on a discrete GPU. It's gotta be GDDR3.

 

Even if sony undervolted their gpu's the core speed should not be lower in contrast to other manufacturers.
It would reduce the power flow to the gpu and result in lower temperatures, but the gpu would still remain at it's standard clock speed and should not have been reduced (unless the lower voltage doesn't allow for standard clocks - in which case, they shouldn't have touched the clocks or undervolting to begin with).

 

the op mentioned that the sony is indeed underclocked, which he swu=tiched to standard but still got subpar results.
op your card could have been both underclocked and undervolted by a significant degree.
best to check your temps first to make sure its not throttling.
then checkvoltages to see how undervolted it is and determine whether you need to increase voltage to 'support' standard clocks.

 

And don't forget that there is alot more then a GPU in a laptop that can affect 3DMarks.

 

My bet is the memory. It's possible that Sony used cheaper memory (worse timings etc.) knowing that it won't affect performance as the core is already underclocked so slower responding memory will suffice.

@htwingnut- we had this discussion many times. DDR3 is routinely used because it's cheaper- proof that it can be is in ATI specs, proof that it actually is- can be found here (dismantled notebook with HD5650)

 

Stop saying "its undervolted" people. That has nothing to do with performance (as long as the card is stable at a given voltage, the performance will be the same).

Its a vBIOS/BIOS or timing issue. Thats it. You could try flashing the vBIOS to a difference manu, but you could very easily brick your card. I assume that either your laptops BIOS or your vBIOS (less likely) throttle down your card at certain temps/workloads.

 

@ trvelbug

Ummm if the clocks he's using weren't supported by that voltage, then machine would simply turn off...it wouldn't result in a lower score...

 

Thanks for the replies guys. To address some points brought up:

Memory: If GPU-Z is to believed (and I know it can be misleading sometimes), it has GDDR3 memory. Comparison of GPU-Z with the other laptops I mentioned showed identical specs. But I suppose the only way to know for sure would be to open it up and physically look, but I'm not all that bothered about the performance to do that. I'm just curious about why.

I'm leaning towards it not being the memory, though. Standard memory clocks are 790Mhz, and yet I can push it to 950Mhz with no issues - if the memory was compromised I don't think it'd overclock that well.



Undervolting/underclocking: As I noted, the Sony is both undervolted and underclocked. But as I and classic77 also noted, voltage makes no difference to performace at identical clocks. Clocks dictate the performance, voltage only dictates how high the clocks can be pushed while being stable.



Temps/throttling: Temps aren't an issue at all. The highest temp I've seen it hit is 72 degrees, and that was overclocked at 620/880 after a prolonged gaming session. At "stock" clocks it barely even tickles 68 degrees.

Throttling is more interesting though - something I found during stress testing with Prime95 and Furmark was that if all four cores of the CPU AND the GPU are loaded to 100% at the same time, the CPU automatically drops to 1200Mhz. Not the GPU, the CPU. GPU's clocks stay perfectly stable.

It's not temperature related - it makes no difference if they've been running for a second and the temps are 55 degrees, or 20 mins and they're 70 degrees. As soon as 4 cores of the CPU and the GPU are under 100% load, the CPU drops its clocks instantly.

But again, in 3DMark terms, I don't think this is the issue. It only happens when all 4 cores are under 100% load - if it's just 2 or 3 cores, or 4 cores are only loaded 80%, it doesn't throttle. And I don't think 3DMark is all that demanding to load up the CPU. (3DMark's CPU score is in most cases higher than the other laptops I'm comparing with. It's just the 3D scores that are lower).



So yeah.. still eyeing BIOS/vBIOIS