Slashdot Mirror
Matrox Parhelia 512 Preview
SpinnerBait writes "Finally, you don't have to sift through all the unreleased and unauthorized
bogus information around the net about Matrox's upcoming 3D Graphics chip,
called the Parhelia 512. Matrox has taken the wraps off their next
generation GPU and
this Preview over at HotHardware goes through its feature set with a fine
toothed comb.
They also give you a very rare glimpse inside Matrox's Montreal Headquarters,
as well as a look at some very impressive technology demos, rendered on their
new chip. Looks like impressive stuff for sure."
...to release decent drivers. Tested and stable would be nice...
Here
The summary mentions quite a few interesting notes regarding the effect this card would have on current games.
- In "simple" games like Quake III Arena, the Parhelia-512 will definitely lose out to the GeForce4 Ti 4600. By simple we mean games that generally use no more than two textures and are currently bound by fill rate. NVIDIA's drivers are highly optimized (much more so than Matrox's) and in situations where the majority of the Parhelia's execution power is going unused, it will lose out to the Ti 4600. This can change by turning on anisotropic filtering and antialiasing however, where the balance will begin to tilt in favor of the Parhelia.
- In stressful DX8 games, Matrox expects the Parhelia-512 to take the gold - either performing on par or outperforming the GeForce4 Ti 4600. Once again, as soon as you enable better texture filtering algorithms and antialiasing the Parhelia-512 should begin to seriously separate itself from the Ti 4600. The quad-texturing capabilities of the core as well as the 5-stage pixel shaders will be very handy in games coming out over the next several months.
So from the look of it, Parhelia does not wipe out Nvidia (though I would like them to), but is a worthy competitor to nvidia in current games. It would be interesting to see how ATI and Nvidia match up to this new competitor in the coming months.
Be afraid. Be vewy vewy afraid.
Rapid Nirvana
I wonder if their new Parhelia can deliver on its promises? Have Matrox's openGL drivers improved significantly over the past few years? Poor openGL was what killed G200's promising future, and I would hate to see a repeat performance.
The human eye can distinguish about 10 million different colours. But it's more sensitive to some frequencies than others, so sometimes 24 bits (16 million colours) may not be enough.
For example, most people can distinguish between two very similar 24-bit medium greens but not between three or four similar 24-bit dark blues.
That said, no monitor can accurately represent 16 million colours, let alone several billions. Even if they could, the dynamic range of monitors is very limited compared to the range our eyes can see (ie, monitors have very limited brightness compared to the normal sunlit world), so most of those colours would be wasted.
Higher colour precision is good because it minimises round-off errors, but this applies mainly to internal calculations (some operations are done directly on the final framebuffer, but very few). For display, 24 bits (and a good monitor) are more than enough.
RMN
~~~
For those of you who don't already know, professional TV standards (specifically, D1, also known as SDI, though SDI is technically different) use 10-bit YCrCb video.
This means that any particular pixel may have up to 30 bits of color (even though the maximum difference between colors of pixels is less than that.
Obviously, this is not something that is easily accomplished with standard 24 bit/32 bit rendering. If you convert the SDI into something that can be represented in the frame buffer of the video card, then you've lost precision. This is unacceptable for broadcast! (And no, overlay isn't generally good enough since you want to capture the pixels for output though SDI)
Admittedly, this card isn't perfect- It would be nice to have 8 bits of destination alpha (for a key channel). 4 shades of keying just isn't enough...
In any case, having a card (finally!) support 10 bit rendering (especially the 10 bit rendering in openGL) in hardware will be wonderful!
Take a look at this explanation which explains what a parhelia is =)
interesting stuff
09 F9 11 02 9D 74 E3 5B D8 41 56 C5 63 56 88 C0
- First off, Q3A is used as THE single standard metric to see how a card will perform under a common load. It's a very good way to judge the raw speed of a card overall, and often provides good pointers as to overall performance in fancier modes or other games, but it certainly doesn't mean every game you play will be 100+ fps.
- Second, that figure is an AVERAGE. When actually gaming, the average framerate is not the issue - the MINIMUM framerate is the killer. 60 fps average is fine, but when the framerate drops to 10-15 fps in a heavy firefight, you're in trouble. A higher average framerate usually translates to a higher minimum as well. In fact, many sites have taken to quoting minimums as well, or even showing a complete framerate graph.
- Third, the ability to manage 100 fps at e.g. 1024x768 means only around 40 fps at 1600x1200, if your monitor extends that far, or perhaps only 30 fps at 1024x768 with 4x AA if it doesn't. Your card will need to score 200 fps if you want to improve your resolution/AA, or maybe even 300 fps if you want to do that and still keep your minimum fps above 60.
- Fourth, the same argument applies to other quality improvements like trilinear and anisotropic filtering. Taking 32 texture samples instead of 4 can really kill your framerate, so you better hope you're getting enormous framerates with non-anisotropic filtering if you hope to get acceptable speed with anisotropic filtering enabled.
- Fifth, Q3A is not the only game out there. There are a lot of more demanding games available today, even those based on the Q3A engine like RtCW, that will give you much lower framerates.
Combining two or more of the above factors can bring the fastest graphics card to its knees, even if it scores 200 fps in Q3A. We'll have to wait until we see scores of 300 or 400 before we can expect to play Jedi Knight II at 1600x1200 with 9x AA and 16-sample anisotropic filtering, while never dropping below at least 30 fps. But boy, will it look good when we can :-)
Ideally, a review will give individual scores for all the above - high resolution, AA, anisotropic filtering, a range of modern games, and all combinations of the above. But since this would entail a vast amount of testing and a huge array of numbers, most reviews settle for a few known tests that are indicative of performance in other tests. And the most popular of those is good old Q3A.
Why would anyone engrave "Elbereth"?
Matrox doesn't actually have a good history of getting cards out in a decent time frame. Figure that by the time this card is actually available (anyone remember the g400? how many months did it take to get one after it supposedly became available?) it will be irrelevant.
The next problem is that Matrox ruined their reputation in my eyes with the G200 by lieing about OpenGL. Lieing about how they were going to have it in November, then December, and so on... they kept this up until they announced the G400 and then suddenly the g200 was a no-go.
Ever since the G400 series it seems Matrox has been coming up with feature laden cards... trouble was no one asked for the features they chose to offer. Now they added even more features and a buttload of performance to boot. Yet as before, GF5 will be announced about the time this card is supposed to ship, and most likely be in stores at the same time.
* Winners compare their achievements to their goals, losers compare theirs to that of others.
As someone else pointed out, 10 bits of RGB does not equate to 10 bits of YUV. The Parhelia will give great 10 bit RGB previews (completely independant of output quality), and will even output a 10 bit YUV video signal - but only via S-Video, where the two colour signals get encoded together anyway. You need 10 bit component output, or 10 bit SDI, neither of which can be done by the Parhelia. It's more aimed at the 10 bit DVD market than a professional output solution.
The two-bit alpha limitation is largely irrelevant. For display on a monitor, RGB is all you need. Processing of deep-colour images should be done with at least 16 bits per component (including alpha) in memory for best results, then dithered down to 10 bit RGB for display. Key channel output requires a second video connector, so it won't do that at all.
Why would anyone engrave "Elbereth"?