Memory speed vs Clock speed... which is more important?

[HFBallad]

I've noticed that R280x comes in various factory overclocks depending on the manufacturer. What I found a bit interesting is that, instead of simply upping the clockspeed, the manufacturers also upped the memory speed.

Now, I understand that "the bigger number is better", but which (memory vs clock) overclocking is more important for a video card? I want to say "clockspeed" off the top of my head...

Memory speed only comes into play when there is a ****load of texture files (or using multiple screens / tons of AA), correct?

 

[LordEC911]

I've noticed that R280x comes in various factory overclocks depending on the manufacturer. What I found a bit interesting is that, instead of simply upping the clockspeed, the manufacturers also upped the memory speed.

Now, I understand that "the bigger number is better", but which (memory vs clock) overclocking is more important for a video card? I want to say "clockspeed" off the top of my head...

Memory speed only comes into play when there is a ****load of texture files (or using multiple screens / tons of AA), correct?

Only works when comparing the same cards but clockspeed > memory speed in the majority of the cases. As long as there is enough bandwidth to not completely bottleneck the GPU, clockspeed wins out since it is what does the work when rendering scenes/games.

When overclocking, if you increase the core clock you will see a near linear increase in performance.
Memory overclocking really depends on the card's specs, the game settings and the scene.

 

[Blkout]

Both important, no correct answer to this question. If one or the other wasn't important you would see video card with completely out of balance clock speeds like 1GHz core and 100MHz memory. They both play vital roles in maximizing performance.

 

[HFBallad]

Ah.
Well, in the R9 280x example that I used,

Gigabyte 1100 clock, 6000 mem
Asus 1070 clock, 6400 mem

As you can see, Gigabyte version has a slightly higher clock speed (~3%), but Asus has a a bit higher mem speed (~7%). My guess was that Gigabyte would be faster at lower resolutions, while Asus would be faster at higher resolutions.

 

[Blkout]

You're looking at a 1-2fps difference between those cards at the MOST. It would be completely unnoticeable except in a benchmark.

 

[Rizen]

Depends on where the bottleneck is on the card. I would imagine that with their 384-bit bus, most 7970/280X cards would benefit from core much more than memory.

 

[ccityinstaller]

Depends on where the bottleneck is on the card. I would imagine that with their 384-bit bus, most 7970/280X cards would benefit from core much more than memory.

Actually you would be quite surprised...If you look in the 7950 O/C thread, you will see a post a guy did on the second to last/last page that shows the difference between Core/Memory scaling in Crysis 3..While the game loves a super fast core, actually pushing the memory up and keeping the core down low produced some very interesting results...I will see if I can find the two graphs..Here they are...Once again, Props to Face2Face:

If anyone is interested to see how Crysis 3 scales with the 7950 here are some benches I did.

1080P - VHQ- Textures - HQ - system spec - 2XSMAA - 4.8Ghz 3570K + 16GB Sammy 32nm

As you can see, the super lovely memory we were both blessed with (both run MSI TF3 cards) scaled VERY, VERY, well in Crysis 3..I know this is just a single game, but still pretty cool.

 

[Kor]

Actually you would be quite surprised...If you look in the 7950 O/C thread, you will see a post a guy did on the second to last/last page that shows the difference between Core/Memory scaling in Crysis 3..While the game loves a super fast core, actually pushing the memory up and keeping the core down low produced some very interesting results...I will see if I can find the two graphs..Here they are...Once again, Props to Face2Face:



As you can see, the super lovely memory we were both blessed with (both run MSI TF3 cards) scaled VERY, VERY, well in Crysis 3..I know this is just a single game, but still pretty cool.

That's some monsterous mem o/cing right there, wish I could squeeze that kind of speed out of mine lol.

 

[Blkout]

That's some monsterous mem o/cing right there, wish I could squeeze that kind of speed out of mine lol.

It has been my experience from benching that 500MHz worth of memory speed is equal to about 100MHz worth of core speed increase in most benches. So 1000/6500MHz would perform very close to 1100/6000MHz in many game benches.

 

[LordEC911]

As you can see, the super lovely memory we were both blessed with (both run MSI TF3 cards) scaled VERY, VERY, well in Crysis 3..I know this is just a single game, but still pretty cool.

Yes because when they tested the GDDR5 clocks they had already overclocked the 7950 by 50% on the core.

Also take note of the 1200/1900 score vs the 1200/1650 score, a 4.5% performance increase for a 15% increase in memory clocks.
1300/1650 vs 1150/1650, 13% increase in core clock and a 11% increase in performance.