ATI video card runs in 8x PCI Express speed - want 16x?

[Naki]

I have an old system with nVidia nForce 4 SLI chipset, which has PCI Express 1.0 and not 2.0.

For some reason, the ATI 4870 videocard (made by Gigabyte) I have runs in PCI Express 1.x x8 mode. Being PCI Express 1 and not 2.0, this makes the card pretty slow.
It is supposed to run at 16x. But it doesn’t. Is there a way to fix this? I use GPU-Z to check the mode - it says PCI Express 1.1 x8, capable of 16x.

The motherboard has two PCI Express 16x video slots, but 16 lanes total - so when two videocards are used, they run in x8 + x8. If one videocard is used (and the second video slot is empty), it SHOULD be at 16x mode.

Motherboard BIOS is latest. Videocard is, for some reason, not having BIOS updates on the Gigabyte web site. There are other models with 4870, which do have BIOS updates, but not my model.

[kalniel]

Checked for jumpers or cards that you need to place in the second slot to make the first slot x16? I still doubt x8 would be limiting.

[CAT-THE-FIFTH]

Both the PCI-E slots on my 975X motherboard are limited to PCI-E 8X and TBH it does not seem a limiting factor for my HD4830.

Even an HD5870 only loses a small amount of performance when placed in a PCI-E 2.0 4X slot and AFAIK this should have the same amount of bandwidth as a PCI-E 8X slot:

http://www.techpowerup.com/reviews/A...xpress_Scaling

[Georgy291]

er i dont think x8 is limiting in any way possible with a single 4870, i was playing with mine by going in to bios and then reducing the link from x16 to x1

then benching it, down to x4 the difference was negligible but i was doing it with a PCI-E 2.0 and abit AX78 mobo

[scaryjim]

From this bit-tech review of your motherboard:

On the subject of the SLI option, you will see that there's no SLI selector card between the PCI-Express x16 slots. This is because ECS don't use one and so both slots are PCI-Express x8 speed, all the time, even when you use just one PCI-Express x16 graphics card.

So there's no way to fix it, because it isn't broken - that's just the way it works. Sorry

[Naki]

Thanks scaryjim, that sucks... Well, I am sure PCI Express 1.1 x8 is a limiting factor, because my 3D Mark scores are pretty low. Other people with the same card get much higher 3D Mark scores (using PCI Express 2.0 of course).

[kalniel]

Thanks scaryjim, that sucks... Well, I am sure PCI Express 1.1 x8 is a limiting factor, because my 3D Mark scores are pretty low. Other people with the same card get much higher 3D Mark scores (using PCI Express 2.0 of course).

And all the other more important factors are exactly the same between your test and theirs? (Like CPU speed etc.)

[scaryjim]

While your 2.4GHz Opteron can probably just about keep up with a 4870 (although it will be a partial bottleneck), I suspect your overall memory bandwidth - since you're running DDR1 - is going to be a more significant bottleneck. If you really want to get the best out of that 4870 you need to do a full system upgrade, I'm afraid....

[kalniel]

While your 2.4GHz Opteron can probably just about keep up with a 4870 (although it will be a partial bottleneck), I suspect your overall memory bandwidth - since you're running DDR1 - is going to be a more significant bottleneck. If you really want to get the best out of that 4870 you need to do a full system upgrade, I'm afraid....

I've not seen memory bandwidth cited as a bottleneck at gaming resolutions before, I'm pretty sure they actually showed the skt 939 chips were actually slightly faster in gaming than AM2 at the same clock speed, due to lower cache latency, despite the reduced bandwidth.

Another example would be when I tried OC'ing my computer - I had to drop the RAM divider because running 4 sticks was tough on the controller, so now I'm running at DDR 384 or something - clearly less bandwidth than DDR400. Yet the affect of increasing CPU speed a little is shows a positive benefit on fps, so the decrease in bandwidth hasn't had much effect. (With a 4870).

[DanceswithUnix]

I've not seen memory bandwidth cited as a bottleneck at gaming resolutions before, I'm pretty sure they actually showed the skt 939 chips were actually slightly faster in gaming than AM2 at the same clock speed, due to lower cache latency, despite the reduced bandwidth.

Well that was true in the early days of AM2, but DDR2 latencies came down a lot. Then the first shrink of AM2 CPUs improved things like SSE and the memory controller, and the core updates have come steadily in since then.

So DDR2 should now be worth the equivalent of one speed grade. Not a stunning difference, but one that you can probably feel.