48xx performance - Higher core or memory overclock

[BlueWeasel]

Got 2 4830's sitting here in front of me: Sapphire 4830 and MSI 4830.

I currently use the Sapphire and have modded the BIOS to increase the core voltage to 1.15v. The RAM chips are Qimonda and won't OC past 1000Mhz. Currently, I run the card at a maximum OC of 760/995

I've not tested the MSI yet, but everything I've found online indicates it can hit 700+ core easy and RAM around 1050-1100Mhz. The VRAM is Samsung and higher rated than the crappy Qimonda chips on the Sapphire.

In theory, with the 4800 series of cards, does a higher core trump a higher memory OC?

Just assume that the Radeon 4830 can hit 725/1075-1100. How would that compare to the Sappire running at 760/995? Granted, I know the performance difference is minimal, but which is the better (albeit slightly) card.

 

[JBT]

Atleast for the 4890's the core is MUCH more important.
http://www.anandtech.com/video/showdoc.aspx?i=3555

 

[fffblackmage]

Originally posted by: JBT
Atleast for the 4890's the core is MUCH more important.
http://www.anandtech.com/video/showdoc.aspx?i=3555

the thing about the 4890 is that it has gddr5, giving the 4890 insane amount of memory bandwidth. I wouldn't expect any memory overclocking to help much because it's already sufficient.

Anyways, I've heard overclocking the memory typically helps more since many cards are usually memory bandwidth bottlenecked. However, performance gains may depend most on what resolution you play at.

 

[Rifter]

With my 4830 my massive core OC was very noticable in games. I was able to keep the core clock stock and raise mem to 1100 and saw no diff whatsoever in any game but did notice a diff in benchmarks. So the answer is if you are one of those guys who likes to run 3 or 4 way CF/SLI just to have the best 3dmark vantage score out there but hardly games anyways its worth OCing the mem. Otherwise i would go for the most core possible, its worth noting that i can only obtain my 800 core with the memory at stock clock. I would assume this is a power delivery problem as im sure my 800 clock takes more power and without raising the voltage its taking some from the memory. Im looking into bios hacking but its a moot point as im getting one of those factory OC 1Ghz 4890s as soon as i see one for sale in canada.

 

[fffblackmage]

Originally posted by: Rifterut
With my 4830 my massive core OC was very noticable in games. I was able to keep the core clock stock and raise mem to 1100 and saw no diff whatsoever in any game but did notice a diff in benchmarks. So the answer is if you are one of those guys who likes to run 3 or 4 way CF/SLI just to have the best 3dmark vantage score out there but hardly games anyways its worth OCing the mem. Otherwise i would go for the most core possible, its worth noting that i can only obtain my 800 core with the memory at stock clock. I would assume this is a power delivery problem as im sure my 800 clock takes more power and without raising the voltage its taking some from the memory. Im looking into bios hacking but its a moot point as im getting one of those factory OC 1Ghz 4890s as soon as i see one for sale in canada.

wtf. copy paste?

lol

 

[Modular]

I've heard that the 4850 is held back by the memory speeds from someone on this forum. I'm not sure how much I believe that though. I guess the best way to figure it out is to do some testing of my own. I do know that I gain a decent amount of FPS in pretty much every game I play by increasing the core from 625 to 775 on my 4850. I'm at a relatively low voltage too. I think I'll have to shoot for 800+ in a few weekends when I have more time.

 

[Rifter]

fffblackmage

i was decided on that over linking to the other post on the same topic lol

 

[SickBeast]

If the card has a 128-bit memory bus (4850, 4830, 4770) then it will be held back by the memory speed.

The 4850 may in fact be 256-bit but it's still held back because it uses DDR3 vram.

 

[Fox5]

Core typically matters more unless you're playing around with anti-aliasing.

 

[BFG10K]

Upping the core speed over the memory speed is almost a universal constant these days.

As an example, from my testing the 4850 often beats a 260+ when running 8xMSAA in recent OpenGL applications, despite the 4850 having about half the memory bandwidth of the 260+.

 

[error8]

Originally posted by: JBT
Atleast for the 4890's the core is MUCH more important.
http://www.anandtech.com/video/showdoc.aspx?i=3555

And yet, when you crank up the core too much, memory bandwidth starts to matter a lot.

"Note that in both cases we see a much bigger boost in performance. This means that while applications tend to be very heavily compute limited, at higher core clock speeds on AMD hardware memory bandwidth increasinly becomes a bottleneck."

 

[SickBeast]

The priorities don't matter anyway. Typically the core will run at a certain speed regardless of how high you clock the memory. It might make a 0.05% difference. Just run them both as fast as you can.

 

[Tempered81]

Originally posted by: SickBeast
Just run them both as fast as you can.

+1.

However,I'd try to get the most out of the core rather than the mem.

If you can choose between 850/1000 and 760/1025 (both stable), then I'd take the 850/1000.