How Does Vram Speed Effect Performance at a set Gpu core Frequency

[dmferrett]

Does anyone know what the relationship is between say a 1250, 1500 and 1750Mhz Vram frequency and performance in terms of FPS at predetermined gpu core frequency.

I am trying to decide how hard to push vram. (Without spending hours running bench marks)

I've tried 1250,1500,1650 and 1700Mhz.

 

[pgi947]

I would say a few fps here and there, more likely to have a larger effect at higher resolutions though as the need for memory bandwidth becomes greater.

 

[Silent_Scone]

As above, it's beneficial when memory bandwidth is required mainly at higher resolutions or large amounts of AA. A lot of games if you're playing at 1080 will have plenty of fuel left in it on say a 780GTX without the need for overclocking.

 

[Orangey]

If it's a 290 then 1500 is a fine 24/7 if you can get it.

But each game is different. There is a huge difference with even just 200MHz in some games.

 

[Marine-RX179]

I think the biggest advantage of high memory bandwidth is that you lose much lesser frame rate when using SuperSampling.

The way I understand it, bumping core clock is for increasing raw grunt/power for pushing frame rate, whereas bumping memory clock/memory bandwidth is for reducing the amount of performance hit when using fancy post-processing graphic features- AA, SuperSampling, high-res texture etc.

 

[Silent_Scone]

I think the biggest advantage of high memory bandwidth is that you lose much lesser frame rate when using SuperSampling.

The way I understand it, bumping core clock is for increasing raw grunt/power for pushing frame rate, whereas bumping memory clock/memory bandwidth is for reducing the amount of performance hit when using fancy post-processing graphic features- AA, SuperSampling, high-res texture etc.

Interestingly on that point at 1080P with a 780Ti there is no difference in Tomb Raider benchmark with 2X SSAA minimum frame rate for me between 3600 and 3930mhz.

 

[Marine-RX179]

Interestingly on that point at 1080P with a 780Ti there is no difference in Tomb Raider benchmark with 2X SSAA minimum frame rate for me between 3600 and 3930mhz.

You have to bare in mind that the 780Ti already has extremely high memory bandwidth to begin with, so clock it just 300MHz higher is may be not gonna make much different on those settings. Also, I was more referring to the frame rate of no SuperSampling vs using SuperSampling- cards with high memory bandwidth such as the GTX780Ti/R9 290(s) (with 336GB/s, 320GB/s out of the box) enabling SuperSampling lose less frame rate in %...in contrast to cards such as GTX670/GTX680 with just around 192GB/s out of the box will lose up to around 50-55% of their frame rate just by enabling SuperSampling alone.

 

[Silent_Scone]

Sorry should have been clearer, wasn't insinuating you were wrong mate, just something I noticed myself. Most high end GPUs will have the bandwidth for there not to be a bottleneck as you say.

 

[dmferrett]

It would be great if someone could do a few benchies to see what the real world difference is say at 1200Mhz core, 5000Mhz, 6000Mhz and 7000Mhz VRam clock. Just so we have some numbers to gawk at.

 

[Meaker]

Lol, depends entirely on the card, which you have not even told us yet.

 

[Rusty0611]

It would be great if someone could do a few benchies to see what the real world difference is say at 1200Mhz core, 5000Mhz, 6000Mhz and 7000Mhz VRam clock. Just so we have some numbers to gawk at.

The clock speed doesn't mean anything independently. It's the actual bandwidth which affects performance.

So you'd be better off testing a 256 bit bus card in a memory bandwidth intensive game along side a 384 bit bus card. I'd expect the 384 bit bus card to not gain as much and plateau at a lower frequency.

Overclocking the memory on GK104 produced sizable gains but that was due to the increase in bandwidth as opposed to anything fancy about the design.

 

[teppic]

So you can't be bothered to do the benches but you want someone else to?

Performance difference will vary hugely depending on the GPU, bit width, resolution, game and settings.

 

[murah]

There is a relationship between clock and memory speed but it entirely depends on YOUR GPU, you see i could test my card and produce different results to you.

For instance my card with Kepler boost gives me 1202mhz on the core, the memory i got a 1ghz increase but it was trial and error and time that got me there. If you can't be bothered then just do your core clocks you will still get a sizeable gain.

But if you want to overclock your GPU properly THEN YOU HAVE TO SPEND THE TIME
Step 1: Core to max stable run heaven benchmarks check your score each and every time, when you hit the point your score decreases then you have hit your wall and its time to dial down the Core a bit once you have your sweet spot put core back to what it was before clocking.

Step 2: Memory to max stable, repeat heaven runs as above.

Step 3: Combining clock + memory, try a run with both of your max stable activated if you pass then you can fine tune. Run a game at this point for about an hour if you see weird graphical glitches or things are generally unstable or if you fail then decrease the core OR memory (Or both but try core first) my rule of thumb is at that point memory actually gives a bigger improvement (Game depending but mostly true) than core something in the region of 5 to 1 so if you reduce both memory and core take the core back by 5mhz jumps and the memory by 1mhz jumps if possible on your card.
If your score decreases but you pass, then you have to reduce core/mem or both.

estimated time roughly 3 hours. End result totally worth it.