Rough Benchmarks
Thresh:
Of course, no preview would be complete without a few benchmarks showing off what said SLI is capable of. Understandably, at this point in time Metabyte only had a controlled demonstration setup to show off their fledgling technology. We sat through a demonstration of Forsaken, Incoming, and several select tests of 3DMark suite.
| Benchmarks @ 1024x768x16 |
| Benchmark |
Percentage Gain over single card |
| Forsaken |
80-90% |
| Incoming |
70-80% |
| 3DMark |
25-30% |
Kenn:
First of all, we're sorry we can't print actual numbers in this review. The drivers and the technology are still in development, so Metabyte doesn't want anything quantitative quite yet. This is understandable, and we'll be pushing hard to see when we can run our own benchmarks on a near-release PGP setup.
Of course, these benchmarks are designed to show off the technology in the best possible light. This means fill-rate and vertex-limited applications. A game such as Forsaken, which runs at up to 100fps at lower resolutions on standard equipment, is certainly not CPU-limited, and execution time is not held up in the geometry process, as in id's upcoming Quake III arena.
You can see this more readily in the two 3DMark tests. Remember, this shouldn't be taken as a complete 3Dmark score, as the only tests run were Game 1 (Race), Game 2 (FPS), and Fill Rate 1 (Single Texture). However, you can see that the benefits of PGP are readily apparent. Again, remember that even Voodoo2 SLI doesn't deliver the ideal 2x performance. Expect a 30-40% increase in framerate for a COU-limited game such as Quake II. As with 3DFX SLI, PGP performs its best at high resolutions, which explains the 1024x768 setup.
Thresh:
Yadda yadda yadda. I wanted to also note that we didn't have anything to do with the driver or testing setup. So while we don't believe Metabyte was trying any kind of funny business, you really have to take any results presented here with a grain of salt, at least until we get some solid product to run real tests with. The "controlled" situation is understandable however. Both the hardware and software aren't yet finalized nor optimized, and a ton of things may change before it's stabilized. This preview is meant more as a confirmation that the technology really exists rather than a benchmarking session (we'll have that for ya a little later).