Widescreen Gaming Forum

Hello,

i want to share with you some bizarre benchmark results.

CPU: Intel Quad Q9650 @3.8 GHz
RAM: 8GB (4 x 2) PC2-6400 @ 840 MHz 5-5-5-16

Until recently, I've had 1GB Radeon 5870.
I was stuck in some games (GTA IV, Gothic 4) @ average 25-35 fps in 5040x1050 Eyefinity, no matter what internal game & catalyst 3D gfx settings I chose. Sometimes I suffered from fps drops to 15-20 fps.

Some old GTA IV benchmark @ 5396 x 1050, GFX settings @high,
Radeon 5870 OC @ GPU 900 MHz / Mem 1250 MHz

Statistics
Average FPS: 32.25
Duration: 32.90 sec
CPU Usage: 67%
System memory usage: 67%
Video memory usage: 95%

Seemed to me like I need more VRAM.
So I upgraded to 2GB Radeon 6970.
Now, surprise:

Statistics
Average FPS: 28.58
Duration: 33.20 sec
CPU Usage: 66%
System memory usage: 47%
Video memory usage: 42%

Even less fps !

How is it possible, that I have lot of unused CPU, GPU, DDRAM, VRAM, that game fails to use ?!

Some more tips:
- GTA IV Multithreading is OK, no Intel core is used @100%, they stay in 50-80 % range.
- When I disable shadows in GTA IV totally, there is 10-15 fps improvement.
- Gothic 4 shows exactly same low performance as before - stuck @ 30 fps.
- some other games run now in Eyefinity much better, some not.
- Radeon 5870 benchmark was made @ 4GB 1080 MHz DDRAM2. Had to modify timings after I installed another OCZ DDR2 PC2-6400 Reaper CL4 4GB. Old one overclocked much better.

There must be some bottleneck. Probably has to do with RAM, is there a way to test it ? Some software to log amount of RAM transfers?

Do you have any other ideas ?

PS. Sorry for my bad English, is not my native language.

Sounds about right that there no improvement between the two in most title, ATI messed with there naming conventions.

The fact that, by AMD’s own admission, its new single-GPU flagship goes up against Nvidia’s second-fastest board is probably not what AMD’s loyal fan base wanted to see. And it again puzzles me to see such unrepresentative naming in play here. Think I’m off my rocker? That anyone reading Tom’s Hardware should simply “know better” and not worry what these cards are called? Just try telling me this slide from AMD's press deck doesn’t look completely off:

http://www.tomshardware.co.uk/radeon-hd-6970-radeon-hd-6950-cayman,review-32073-22.html

Some of the info is in that article but if you want more info on ATIs messed up naming convention for the 6000 series cards you need the article on the 6850 and 6870.
http://www.tomshardware.co.uk/radeon-hd-6870-radeon-hd-6850-barts,review-32028.html

_________________
: ) Skid : )
My Multi-Monitor Focused YouTube Channel: http://www.youtube.com/c/SkidIncGaming
My Twitch channel which I've been known to stream on from time to time: http://www.twitch.tv/SkidInc
"Imagination is the only weapon in the war against reality."

Sounds about right that there no improvement between the two in most title, ATI messed with there naming conventions.

You're right, I'm switching to 5970, as there is no GTX with triple monitors support and I don't want to go into SLI / Crossfire yet.

Still, I don't really understand where is the bottleneck, since I have GPU, CPU and memory reserves.

These two cards are indeed very similar performance-wise and the benchmark run is too short for the additional VRAM to be used. GTA4 is a heavily CPU-bound game anyway, and even in Eyefinity only a massive GPU upgrade would make any difference - a jump from unplayable to ok, basically. The overclocked VRAM of your 5870 is the reason why you get better results with the older card, not to mention the slightly better CPU/RAM clocks.

The percentages only give you a basic idea of what type of resource each advanced setting requires; they don't mean there is no bottleneck. My percentages are very similar to yours despite our very different PC configs, simply because we use the same settings and resolution. And then again my average fps is roughly twice yours (~65fps).

So - yes, the DDR2 is part of the bottleneck. As far as GTA4 is concerned, I'm afraid your only option for improved framerates is a platform upgrade (CPU+RAM)...

Switched to Radeon 5970.

Game runs differently : I have fps drops ONLY when I turn around for the first time, or drive fast through city (probably when new textures are being transfered to VRAM).

When I stay in one place fps rate is much higher on 5970 then on 6970.

Seem logical - 2xGPU, 2 x 1 GB VRAM vs 1 GPU, 2 GB VRAM.

Basically, in 6970 there were less fps fluctuations.

Sadly, GTA IV benchmark shows about 28-30 fps average on both cards.

As there is massive fps gain when I disable shadows, without GPU usage change, I suspect that GTA IV has some own build in shadow processing system that need lot of data transfers, as GPU & CPU are not used in 100%.

I just ordered :
- I5-2500k
- GA-P67A-UD7
- OCZ DIMM 8 GB DDR3-2000 Kit

We'll see, how this will help.

Next step could be changing the 5970 to 4GB version & or going Crosfire.

I don't think more VRAM would help, especially with GTA IV since the framebuffer never gets big enough to require that much memory. To really know if more VRAM would help you'll need a profiler like VTune to tell you how big the framebuffer is getting. Also remember that multi-GPU implementations each have their own framebuffer, meaning a dual processor card like the 5970 with 4GB will have a maximum framebuffer size of 2GB. Like scavvenjahh said, this game is CPU bound, meaning no matter what kind of hardware you use, the game is coded in such a way that it will always be utilizing it more than the GPU.

I'd say your two biggest bottlenecks are the bus width in ATi cards and the FSB speed of your CPU, so upgrading your CPU/Mobo/RAM will definitely help. Unfortunately you still need one card/monitor with nVidia, so you'll be spending more money if you move to them. The larger bus width of nVidia cards will improve performance though, especially running at high resolutions. ATi cards are still using 256-bit memory as far as I know, while nVidia cards are typically 384-bit. ATi tries to make up for this deficiency by using faster clock speeds.