Slashdot Mirror
Trident XP4 Reviewed
ceebABC writes "In a new review, the Trident XP4 got a nasty reception. Based on the tests, it sounds like Trident has got some work to do on the thing. Looks like this GPU is dead on arrival." Our last story on Trident mentioned them coming back from the dead. Maybe not.
It makes me wonder why an AnandTech article gave such a different opinion. Which one is right?
From page two...
Then he proceeds to run this card at 1600x1200 with beta drivers against cards with excellent debugged drivers. Any sort of numerical or empirical evidence he could get at this point is about .05% useful to me.
Sure the thing may only get 4.9 FPS on a new demanding game at 1600x1200 with beta drivers. I bet you that same card will belt out over 60fps at 1024x768 when the real drivers are released.
People seem to forget that a video card driver's quality can be the difference between horrible performance and class leading performance. If the driver is not debugged and performance optimzed, there is nothing a hardware designer could do to make that card perform well.
I say that this is an excellent card that will allow users who do not want to spend $500 on a video card to play the latest and greatest games on the market. It is a Dx9 card, with full support. To me, this is an excellent card.
I bet they sell a whole boatload of these things to OEM manufacturers and those who do not really want to spend an entire car payment on moving some pixels around. -TinyManCan
They show the results of very different tests. 800x600 vs 1600x1200 resolution. That's a difference of 4x in required fill rate. Since neither of these tests were run at a relevant resolution (most laptops run around 1024x768), neither can be called conclusive. My guess it that Anand is waiting for more stable drivers to test "real" resolutions.
The extremetech.com review is pretty unfair, it's like testing a new Ferrari by seeing how much cargo it can carry and then declaring it a bad car because it doesn't haul as much as a Ford Explorer. This card is aimed at lower resolution (lower fill rate) applications that require low power and cost. Having a DX9 entrant into this arena to me is welcome.
We'll just have to wait for a real review to see if this card is any good.
-Ryan C.
-Ryan C.
While most of the /.ers have noted that the review of the XP4 on extremetech was unfair, they may not realize just how unreasonable extremetech was in putting this sub-$100 card against the best cards on the market.
Because the XP4 deviates from the long-established, direct-mode rendering (which is a brute force method) for tile-based rendering, they are going to need a lot of time to get their drivers in order before they will be able to compete properly with the familiar video cards. The only other card mainstream card that attempted this rendering approach was the Kyro series, which demonstrated that tile-based rendering does have huge potential and that drivers will make or break the card's performance.
Interesting enough, because video cards using the tile-based rendering method are more efficient by 200-300%* when compared to cards using the traditional method, they should see a much lower performance decrease as the screen resolution is increased when compared against direct-mode renderers (e.g. NVidia NV9 cards and ATI Radeon 9500s). While it's true that fill rates do increase substantially with increased resolution, direct-mode renderers simply will experience that much more overfill.
*Direct-mode renderers have an overfill rate of about 2 or 3; this means that for every pixel visible two or three more have been rendered and then disgarded. Tile-based renderers, on the other hand, disgard everything that won't be visible first and only render what's left, giving them an overfill of 0. Figuring out what to cull first before rendering has begun is more complicated than culling excess pixels after they are rendered; this complexity is what makes writing the drivers for a tile-based renderer such a difficult task.
Trident has set for themselves an incredibly difficult challenge: 1) Make a card that uses a tile-based rendering method, which means throwing out nearly everything the graphics card industry has learned the past couple decades. 2) In addition to the first task, they have added the complexity of sharing graphical resources, thus adding all the timing problems associated with such a configuration. If they achieve only 50% of the performance of Nvidia's Geforce4 TI4600, that alone would be a considerable achievement. If Trident meets the 80% performance target they set for themselves, it will be all the more impressive.