If, like many of your enthusiast peers, you've been staring at the counter on Futuremark's website for the past couple of days, wondering when 3DMark05 would make its debut, wonder no longer. If you're reading this, it's past 1pm on Wednesday, 29th of September 2004, and we're allowed to talk about the biggest deal in PC hardware benchmarking.
With 3DMark 99 and 2001SE released, used and abused by just about anyone that wants to test a graphics card in the past 5 years, and more recently a full system in the case of 2001 SE, 3DMark03 was less well received. Too GPU dependant (almost entirely!) for most people's liking, especially after the partly system-limited nature of 3DMark 2001SE, people also complained that 3DMark03's 'game' tests weren't representative of real games at all. No physics code is really run, there's no sound and it doesn't use an engine from a shipping game title.
What people failed to spot was that almost all of the other good, and by good I mean easily adjustable and completely repeatable benchmarks, especially for graphics cards, were nearly identical in their usage. They might have used game engines from shipping games, but little physics code is run (witness Doom3) and people almost always benchmark them with the sound turned off (UT2004 benchmark results are a fair bit slower with sound on, so people turn it off because the result looks better).
That 3DMark03 accurately represented the relative performance of the target hardware at the time (and still does), implemented graphical rendering techniques that have since shown up in triple-A games titles (the stencil shadowing technique in 3DMark03 is largely similar to the shadowing method in Doom3), and did a good job of providing the rest of the infrastructure needed for enthusiasts to compare results (the ORB) reliably and with a sense of authority, is largely swept under the carpet.
Sure, Futuremark had some problems with a certain IHV optimising for 03 heavily in its drivers for a certain generation of GPU products, and the driver validation experiment still looms large, but on the whole, if you want a nice, repeatable way of measuring your relative graphics card performance, 3DMark03 is a decent tool for the job. Combined with other benchmarks, something that most people forget about, it's hugely useful, not only in the final score it spits out, but also in the peripheral tests you can also run.
Synthetics might not be everyone's favourite cup of tea - after all, you can't play them - but combined with good game-based benchmarks (hopefully including some that the tester devised his or her self), they're very much needed.
Think about the pseudo-word Futuremark for a second. They want to benchmark the kind of GPU loads that games will (hopefully) be generating, in the future. It's a tool for the here and now, based on graphical algorithms Futuremark think will be prevalent in upcoming games shipping in the next couple of big game cycles.
We turn to today, and the release of 3DMark05. Building on the same basic principles as 3DMark03, 05 aims to be more GPU dependant that not, testing out rendering techniques they forecast will be used in the future, giving you something pretty to look at, and battering your graphics card into submission, no matter what you run. Even NVIDIA's SLI should be afraid of this beast. It'll be a long time before a single board will drop a 10K score, so the race is on. Macci et al, start your engines.
For the reviewer, it's worth evaluating 05 to see if it fits your testing remit. For me, I need repeatability, reliability, relevance, ease of use, good peripheral tools and a download my readers won't baulk at. I'm a fan of synthetics, and real game tests that also act like synthetics, when I evaluate brand new GPUs (rather than retail boards, where we use more game tests), so 3DMark05 is an exciting tool for me to look at.
So lets take a look at 05 in a bit more detail, to see if it at least fits how we test at HEXUS, examining 05's new shader engine and game tests. I've purposely used the research of this article as a means to see if 05 is a test I can use, so hopefully it'll help you too, if graphics card evaluation and performance testing is something you're interested in.