Fire GL2, GL3, GL4 chipsets and games

[Kraid2xd]

okay, I was just browsing through newegg.com for vid cards and I come across these insanely priced cards with something called a fire GL chipset.... all priced at $1200+

I can't seem to find much info on the gaming capabilities of these cards....

Anyone know how well they would run games like QuakeIII or directx 8 games like Aquanox?

Heck, how well do they fare against say, a geforce3 ti500?

 

[Rand]

Such graphics cards as the FireGL and the like are not marketed for home consumers. The FireGL brand is a professional level OpenGL graphics card intended for 3D Modelling, CAD, Mechanical Engineering, etc.

Usually they tend to do quite poorly in gaming applications. Using the FireGL2 as an examnple it would be outperformed by a GF2 MX level card quite handily in games. Many such cards don't even provide any driver support for DirectX at all.

The FireGL is made for a very different audience and for very different uses then that of consumer graphics cards.
Things that would be unheard of and ridiculously expensive to include on a gaming card are common place on professional level graphics card like thge FireGL brand. As an example very few professional level graphics cards have a memory bus width of less then 256bit.... in compariosn no gaming card has a memory bus wider then 128bit and indeed to design one would be prohibitively expensive.
The FireGL brand supports 16 separate lighting sources in hardware, double that of the GF3. Sustained polygonal throughputs of the FireGL2 are vastly superior to the GF3, though peak rates are similar.
On the other hand, pixel and texturing fill rates of the GF3 are almost 10X that of the FireGL2 in many instances.

The architectural differences between pure professional OpenGL solutions like the Wildcat II, FireGL, Evans & Sutherland models etc and the gaming cards typified by the GeForce 3 is like night and day, both have very different focuses.

Drivers for professional level graphics cards are typically optimized for very specific uses and features, optimizations for individual applications are the common and accepted norm in professional level graphics cards.

The capabilities/design/drivers/architeture of the FireGL boards and the like, are vastly different then graphics cards intended for the home market and conventional gaming.



<< Heck, how well do they fare against say, a geforce3 ti500? >>


They are simply not comparable as they were never meant to be used for any of the same tasks.
In some areas the The GF3 Ti500 can realistically perform nearly 20X faster then the FireGL2. In other areas the GF3 would ground do a virtual halt, and put a tremendous load on the processor, while the FireGL would leave the processor nearly inactive and support features the GF3 will have never heard of and fly through the same tasks a GF3 would choke at.

FWIW, there are professional level variants of the GF1/2/3 in Elsa's Quadro lineup of graphics cards, but even the Quadro tends to share many of the GF's advantages and disadvantages relative to the FireGL which was born and bred from the beginning to as a professional OpenGL accelerator.

 

[merlocka]

Tom did a professional 3D review last summer. Check it here.

There's another review floating around which includes the Fire cards against Quadro2, Geforce3, and Geforce DCC. Can't find it now though...

In general, the Geforce cards do OK considering that they are consumer cards but the pro cards excel in most tests.

 

[Rand]

<< Tom did a professional 3D review last summer. Check it here.

>>



I really would not suggest putting much faith in Tom's review, generally I appreciate his efforts but that "review" was severely lacking. There were huge areas of CAD left totally uncovered and even the benchmarks he did run had some extremely dubious results to say the least.
Neither did he even touch upon rendering accuracy which is often of considerably more import then sheer speed in the intended market.
He ran no scaling or processor utilization tests etc.

 

[merlocka]

and even the benchmarks he did run had some extremely dubious results to say the least.

I know very little of CAD benchmarks and applications, which data was extremely dubious?

 

[Rand]

<< and even the benchmarks he did run had some extremely dubious results to say the least.

I know very little of CAD benchmarks and applications, which data was extremely dubious? >>



3D Studio Max R3.1 results for one were very unusual, 3D Studio Max often tends to be almost a best case scenario benchmark for the Quadro and yet it was handily beaten by by both the WildCat II 5110, and FireGL4. 3D Studio Max is extremely fond of the Quadro, and it renders Studio Max faster then a WildCat II 5000, Tom has it at almost half the speed of the WildCat II 5110 instead of the more realistic 15% difference one should see.
Tom didnt even mention the fact that the Quadro can maintain real time in the viewport while rendering in 3D Studio Max, while is an awfully big factor in it's favor.

Spec Viewperf is very much a synthetic benchmark, and it's results can tell a lot about potential... but it's seldom an accurate masurement of real world performance. Tom makes no mention of that and almost seems to pass it off as a real world benchmark.

Almost all of his benchmarks were very generic, he didnt even touch upon any specific performance benchmarks focusing on one feature like wireframe manipulation, gouraud shaded models, high polygon loads, structural elements, multiple light sources, phong and smooth blinn shading etc. etc.

Application specific profiles and optimizations in the drivers?
He didnt say a word about whether he used them in some or all of the tests.

All of the cards tested had very specific advantages... for example the FireGL lacks fillrate, but handles complex shading very well, and is incredible at calculating lighting. The FireGL is extremely processor dependent, and doesnt typically associate well with dual processor operation, it scales almost linearly with processor speed though. the FireGL driver team optimizes for anything and everything you can possibly imagine, and they consistently manage to eek out more performance then the hardware base would lead you to believe is possible.

Quadro's gaming roots making it a speed demon at manipulating heavily shaded objects, but is very weak under sustained high polygon loads and has the lowest hardware lighting sources support of any of the common CAD cards. The Quadro has the weakest driver support of any of the common CAD cards, and only has basic application specific optimizations.

The WildCat II's are pure speed in wireframes, and extremely good in CAD style shading, but chokes badly with the more complex shading used in 3D modelling. The WildCat II's put a very low load on the processor, and don't scale well with processor speed, the 3DLabs drivers are well suited for dual processor operation. The low processor utilization allow you to be more productive and keep the system responsive when rendering while other cards would keep the processor very busy.

Tom of course doesnt tel us any of this, nor gives us any benchmarks that really allow us to gauge performance in very specific tasks.
Image quality and rendering accuracy which is often more important then pure speed is not once mentioned.

Perhaps unfortunately he doesnt let on that the powerful T&L engines of many of today's gaming cards enable them to give decent performance in very light CAD scenes.
I'm not sure if the GF3 was available when he did his review but if so I find it unfortunate he doesnt mention that the GF3's flexibility in it's T&L engine often makes it slower for CAD tasks thn the pure brute force GeForce 2, and the GF3's drivers have severe performance discrepancies in many CAD apps. This holds true for the GF3's Quadro varient in the Quadro DCC which is often inferior to the GF2 Pro based Quadro 2 Pro.