Nvidia GeForceFX Ultra On Allocation  
By Mark Hachman    
 
Nvidia executives denied rumors that the GeForceFX Ultra series will be deliberately manufactured in a limited supply, but said the high-end part would be on allocation.  
Both the GeForceFX Ultra 5800 and the GeForceFX 5800 will be in stores at the end of the month, said Tony Tamasi, senior director of desktop product management at Nvidia.  
 
In addition, Tamasi acknowledged that the GeForceFX Ultra 5800 will be in short supply, but that the reason behind it was strong demand, which has outstripped the available supply of components. "The Ultra is in short supply; there's no denying that," Tamasi said. "That's not an uncommon situation. The (GeForce4) Ti 4200 was in short supply. The Ti 4600 was in short supply. We don't intend to make them in short supply, however."  
 
Earlier, an Nvidia spokesman said that the GeForceFX Ultra 5800 would be on allocation through April. Tamasi declined to confirm that date.  
 
However, Tamasi declined to comment on the number of GeForceFX components the company planned to make, saying that would violate Securities and Exchange Commission regulations concerning revenue projections. Rumors have surfaced that Nvidia only plans to make 100,000 units of the GeForceFX Ultra series.  
 
Tamasi also declined to comment on rumors that Nvidia is shifting its emphasis to the NV35 and subsequent parts. "There's always a next-generation product waiting in the wings, as soon as you buy something," he said. "But that's always the guarantee of any product in the tech industry."

New Features in DirectX 8.1
 
Pixel Shaders

• Added pixel shader versions 1.2, 1.3, and 1.4. These new versions expand existing functionality through a more powerful set of instructions, registers, and modifiers for programming pixel shaders.
  

Good evening everyone.  Please bear with us and take some time to read through this.  It gets pretty juicy.  THANKS!  
 
Well, it's only Wednesday and it certainly already has been a interesting week!  The 3DMark 2003 / NVIDIA buzz definitely has stirred up a bit of discussion and our feeling is that it is all for the better.  Typically, we don't like to mix it up with this kind of stuff. That "stuff" being one company's slant on another company's product, or Marketing Spin offered, that perhaps may have many motives behind it, that we are not fully up to speed on.  We try our very best to stick to the "facts".  How does a product perform?  What are its features and benefits versus competitive products etc.  It's relatively easy to evaluate hardware.  Fire it up, run some benchmarks, make sure your testing methodology is sound, and then analyze your data.  
 
That "testing methodology" part is the kicker though.  Your analysis is only as good as your test setup and test metrics, which is where this whole 3DMark 2003 debate comes into play obviously.  Frankly, there is definitely a case to be made for synthetic benchmarks.  Often times these tools can measure aspects of performance that are not fully exploited by current application software.  Whether it be games, business apps or professional design and engineering apps, if the software platforms aren't out there to exploit what is coming down the pipe for capabilities and features (DX9 for example), how is the end user supposed to know how a hardware product will perform in the future, when these new software technologies become mainstream?  This is where synthetic benchmarks should come in, in my opinion.  They should fill the void or lag that occurs between current generation hardware technology, to when a software developer can now make use of new hardware tools and performance levels, to deliver better games and other applications.  Frankly, a synthetic benchmark on hardware, for current generation gaming or application software metrics, has no utility at all.  We've always used 3DMark 2001 around HotHardware.com for it's DX8 emphasis, especially in the early days of DX8 gaming when there were few titles.  Now that benchmark is far less relevant, because you can just fire up Unreal Tournament 2003 and take some real frame rates.  However, 3DMark 2003 should have some value relative to DX 9 performance, by all rights, correct?
 
Well, we're trying to sort this ourselves at this point in time.  I hate to say it but I have to give Kyle Bennett at HardOCP a lot of credit for hitting the nail on the head today.  His article called "Benchmarking Right", hits the nail on the head in a big way.  Game Developers need to step up and provide hooks to benchmarking their game engines on the latest graphics, motherboard and CPU technologies.  It's the only way to get true accurate metrics on how your system will perform in "real-world" or should I say "real end user" experience.  Kyle and I have been at odds as of late, but on this topic, we are definitely on the same page.  However, I'll take this topic one step further....
 
If FutureMark wants to really drive their new benchmark utility as a metric, they need to focus on ADVANCED graphics technologies that are not exploited by current game engines and obviously won't be for some time.   If 3DMark 2003 was a DX9 only benchmark, they would be a lot closer to what a "synthetic" benchmark's real value proposition is.  I'll also further that by saying they need to cut out this "subscription" or "beta program" tactic, as they like to call it, because when you ask someone to pay in order to submit their input on something, you are setting up conflict of interest.  NVIDIA has flat out told us that "FutureMark wasn't taking their inputs on it's benchmark's programming methodology", so they didn't renew their subscription, which means NVIDIA is theoretically shut out from  discussion.  This scenario is a REAL problem.  Kyle is right on here again.  There should be a level playing field, with all the major graphics engine players contributing, regardless of "subscription"... otherwise, the benchmark could easily be affected with "bias".  Again, benchmarks are only as good as the methodology.  This current situation is just plain bad methodology.
 
And so, in the final analysis, what does this all mean to you, our loyal readers?  Well, as I've noted here, my opinion is that, at first glance, 3DMark 2003 does have some usefulness as a tool to measure DX9 performance, although it is even "light" in that department, as I understand it at this juncture.  So, my initial thought is that we'll include it in our round of benchmark testing but only use its detail scores for the DX9 portion of it's testing suite, since we can use gaming engines to test and evaluate every other DX8 metric.  However, we'll reserve the right to comment again here, once we have given 3DMark 2003 a full work-over.  
 
Was NVIDIA putting up spin?  Maybe a little but their legacy product performance, shown over at Tech Report, shows they have nothing to hide.  So, why have the major Graphics OEMs spinning up their engineering resources on synthetic DX8 benchmarks, when they should be focusing on performance on DX8 games?  They shouldn't but DX9 is another story and it should be done right, with everyone's input, from NVIDIA to ATi to Matrox, 3DLabs, SiS etc... with no paid "subscriptions", please.