<!—#set var="article_header" value="NVIDIA New Reference Driver 12.xx
More Power For Pentium 4?» —>
Introduction
Writing about new NVIDIA drivers feels somewhat strange. First of all, drivers in general happen to be under constant development. Particularly drivers of graphics cards however are almost changing every week. Only a major change in terms of performance or general behavior is making it worthwhile dedicating an article to a new driver release. We decided that NVIDIA’s latest 12xx-series of (of course unofficial) drivers is worth a closer look.
Congratulations! NVIDIA’s Latest Official Driver Is More Than 6 Months Old!
NVIDIA has a very special situation when it comes to its driver support. While NVIDIA’s rather huge software department under Dwight Diercks is constantly pumping out new driver revisions on an almost weekly basis, NVIDIA is rather unwilling to openly share those latest drivers with its broad customer base of hundreds of thousands of end users. Don’t just take my word for it! Pay a quick visit to the official driver section of NVIDIA’s website and you will find that the latest officially available NVIDIA graphics card driver carries the revision number 6.50!! You can even read the date when NVIDIA released this ancient driver version! It is no less than half a year old, from December 4, 2000!
Let’s quickly remember what happened in the last six months. Since December 2000 NVIDIA has released a lot of new graphics chips as well as chip versions. GeForce2 Go, GeForce3 and lately the ‘customer confusion products’ GeForce2 MX100, MX200 and MX400 sum up to altogether 5 different graphic chips or chip versions. Each of those is younger than NVIDIA’s latest official reference driver, creating a situation that is rather bizarre.
NVIDIA’s ‘Leakage Policy’
Many of you might now be asking, «What on Earth is Tom’s problem? I’ve just downloaded driver rev. 12.10!» Yes, indeed! We are not stuck with ‘old chum 6.50’, because it’s extremely simple to find NVIDIA’s latest ‘leaked’ driver in numerous places on the web. However, those drivers come without NVIDIA’s official blessing and carry the well known tag ‘Unsupported’ with the usual comment ‘use those drivers at your own risk!’ The result is that tech-savvy end users, who want their NVIDIA graphics card to perform on the cutting edge, with the latest features and highest performance are simply left without any support. The only alternative is to wait until your graphics card vendor, like e.g. Elsa, Creative Labs, Asus, MSI, Leadtek or Guillemot, incorporate NVIDIA’s latest reference driver version into its next official driver release. The rule of thumb is that those official card vendor drivers are typically some hefty 5-10 reference driver releases behind NVIDIA, which is hardly very satisfying to the performance freaks amongst us.
If you ask NVIDIA about ‘leaked’ drivers, officials are usually getting kind of tight. A typical example is the ‘prelude’ to this article. NVIDIA was aware of my work for this driver review and I was asked which ‘new driver’ I was actually testing. When I answered back that I am testing driver revision 12.10, the reply was «Why didn’t you ask us for the official driver to use instead of some «hack» from the Internet?» It seems crystal clear that NVIDIA does not appreciate the leakage of drivers, at least from an official point of view. Here are two official statements to this issue.
|
On the leaks, it is a major issue for us. We absolutely don’t do it on purpose. We have not figured out how to contain it. It is a big problem because the interim releases cause support problems for us. They have not gone through the QA process and usually have known bugs. Further, they are usually at interim stages of tuning so having them hit the websites doesn’t allow us to have that big bang of performance increase. Its much more exciting and easier to manage for us to have a large increase versus several small increases. We are putting in place several security measures to try and contain the problem and plan to put in place serious repercussions for those that violate our non-disclosures. |
Thanks Danno!
NVIDIA’s ‘Leakage Policy’, Continued
| Tony Tamasi, NVIDIA’s Director of Product Management:
Leaked drivers continue to be a major problem for us, and cause for a great deal of pain for everyone involved. We really do not want the drivers to leak, and we continue to try and prevent that from happening. We’ve gone as far as turning off a huge number of customer accesses to our prerelease drivers, as well as virtually eliminating all developers from access to prerelease drivers — but the leaks persist. Part of this problem is that our business model requires us to enable our manufacturing partners and OEM’s continuous access to our updated drivers in order for them to progress with bug closure and GPU/system qualification and WHQL certification. Because we are not vertically integrated, that means that top tier OEM’s, the contract equipment manufacturers that build products for those OEM’s, and key AIC manufacturers all need early access to prerelease drivers. Typically these drivers are known to be «buggy» in general, but get built to address a specific bug or bugs that a customer reported, and help to keep the QA pipeline flowing (typically QA cycles at OEM’s can take as long as 3 months). As we progress, the driver continues to improve in quality, eventually to the point where we have a single driver binary that can pass WHQL for all our OEM customers across our product line. We do our best to only post drivers that meet a high level of quality — at a minimum WHQL certification for our product line. The updated drivers we post will meet that mark. The current drivers posted on our web site (built from the rel6 tree) have been posted for about 4 months, so you’re right — we’re definitely due for a new release. The new drivers (12.xx) are based on an updated code tree and class interface (they support the NV20 architecture of course), but along the way we also changed versioning schemes. The «single-digit» prefix driver base (6.xx, 7.xx) are based on the NV10 class [GeForce2 type chips — comment of the editor] architecture, and the two-digit prefix drivers signify the transition to the new class architecture to support NV20 [GeForce3 type chips — comment of the editor]. We went to more digits primarily to provide more flexibility for versioning for specific customer issues. |
Thank you, Tony!
I used to believe that NVIDIA does not really care much about the leaked driver issue. After all, it ensures that there are thousands of end users deliberately playing beta testers. I also don’t know if I would welcome «enforced security measures» against driver leakage, because it would keep the performance hungry and tech-savvy NVIDIA graphics card owners from remaining at the cutting edge.
Driver Revision 12.xx — Enhancing Pentium 4
Finally I am getting to ‘the beef’. What is so special about revision 12.xx that I am dedicating an article to it? In the last week I was told by numerous sources that NVIDIA’s latest leaked driver contains a significant amount of code that is specifically enhanced for Intel’s Pentium 4 processor. While this may be a completely normal development, it could have a significant impact on the current situation in the processor market. In other words, this technical issue could easily have political repercussions.
So far Intel’s Pentium 4 processor is having a difficult stand. Intel’s flagship processor is still too expensive for the performance it offers in today’s applications, especially since AMD’s latest price drop. AMD’s current flagship, the Athlon 1333 processor, can easily keep up with Intel’s Pentium 1700, even though Pentium 4 has a significant clock speed advantage over its competitor from Austin, TX. In the past Intel could justify its high-price policy with the superior performance of its high-end processors. This is where Pentium 4 has failed so far. The main reason is the lack of software support for Pentium 4’s new architecture. Once this situation changes, Intel will have a much easier life bringing Pentium 4 to the masses.
If you look at the configuration of OEM systems you will find out that the vast majority is equipped with NVIDIA graphics cards. If those graphics cards would all use drivers that are enhanced for Pentium 4, it could give Pentium 4 a very important performance boost.
NVIDIA stated that driver revision 12.xx has indeed Pentium 4 optimized code:
| As you know, the P4’s micro architecture is substantially different from P3, specifically in the areas of cache and front side bus (of course there are other difference as well — SIMD2, etc.). We’ve rearchitected much of our memory management code to make better utilization of the higher front side bus bandwidth from the P4, host memory, and to the graphics processor. Additionally, we’ve optimized our code for the increased cache line size, and implemented better prefetching throughout the driver to help compensate for the P4’s deeper pipeline. |
Now you see why we consider it very important to have a closer look at this new NVIDIA driver. It could well have an impact on future Pentium 4 sales!
Expectations
A graphics card driver is of course only able to influence programs that access the display buffer. NVIDIA’s driver 12.10 won’t be speeding up disk transfers, but it could obviously improve Pentium 4’s scores in 3D-games, professional OpenGL-software and maybe even in 2D (office applications).
We compare driver revision 12.10 to driver revision 6.67. The revision 6.67 may sound pretty old, but we should not forget that it used to be pretty up to date until rather recently. The revisions 7.xx never worked properly and as far as I know there weren’t any revisions 8.xx — 10.xx. The first semi-official GeForce3 driver revision was 11.01. Each driver will be run on a Pentium 4 1.7 GHz system as well as on an Athlon 1333 MHz DDR system.
Test Setup
I decided to use a GeForce2 Ultra graphics card in this test, because it is based on the widely used NV10(GeForce2)-type architecture as found on graphics card equipped with GeForce2 GTS, GeForce2 Pro, GeForce2 Ultra, GeForce2 MX or GeForce2Go. Only a very small minority of people owns GeForce3 cards right now, which would make a test with this kind of card rather meaningless to most of you.
| Pentium 4 System | Athlon System | |
| Motherboard | Asus P4T, Bios 1005 beta 1 | MSI MS-6341, pre-release BIOS |
| Memory | 256 MB Samsung PC800 RDRAM | 256 MB Mircron PC2100 DDR-SDRAM, CL2, Setting 8-8-5-2-2-2-2 |
| Hard Drive | IBM DTLA 307030, 30 GB, 7200 RPM, ATA100, FAT32 (Win98) / NTFS (Win2k) |
|
| Network Card | NetGear FA310TX |
|
| Graphics Card | NVIDIA GeForce2 Ultra Reference Card, Driver 1210 / 667 |
|
| Power Supply | 400 W |
|
| Operating System | Windows 98 SE / Windows 2000 Professional Service Pack 1 |
|
| Desktop Screen Resolution | 1024x768x16x85 |
|
| Quake 3 Arena | Retail Version |
|
| Unreal Tournament | 4.28 |
|
| Evolva | Demo |
|
| Dronez | Demo |
|
| MDK2 | Demo |
|
| Mercedes Benz Truck Racing | Demo |
|
AquaNox AquaMark
The benchmark of this upcoming underwater action 3D game is hardly showing any performance change. In fact, the scores got a tiny bit worse with driver revision 12.10. In this benchmark Pentium 4 is not able to show any improvement.
Dronez Demo
The story is completely different with the upcoming OpenGL action 3D game ‘Dronez’. Pentium 4 used to perform better than Athlon with the old driver already. The new driver revision 12.10 makes both processors perform about 10% better than before, with Pentium 4 gaining a tiny bit more performance than Athlon.
Evolva
Evolva is a DirectX7 3D game that might be suited best for the GeForce2 Ultra card we used in our test. You can clearly see that Pentium 4 gains more performance than Athlon once driver revision 12.10 is used.
Unreal Tournament
The engine of UT is not exactly one of the freshest anymore and it is well known to be limited by the processor rather than the graphics card. Athlon used to beat Pentium 4 in this benchmark and this fact still remains with the new 12.10 driver as well. However, while Athlon shows a tiny performance drop, Pentium 4 has a little performance gain.
MDK2
MDK2 at 1024x768x32 is fully limited by the video memory bandwidth, so that new drivers are not able to show any performance difference. Everything stays exactly the same.
Quake 3 Arena
The last 3D game benchmark in this article is Quake 3 Arena, which is too old to be a significant test for 3D-card performance anymore. You can still see that Pentium 4 is gaining a tiny bit, while Athlon is losing a few frames per second with driver release 12.10.
Things are more obvious when using the NV15-demo. While Athlon is only gaining very little, Pentium 4 scores 3.2 fps higher with driver revision 12.10 than with driver revision 6.67. Pentium 4 is able to benefit from the new driver once more.
Professional OpenGL — SPECviewperf 6.1.2
The most impressive impact of the new driver revision 12.10 can be found in SPECviewperf. In the sub-benchmark DX-06 Pentium 4 gains more than 100%! The P4-scores in Advanced Visualizer (Awadvs-04) and Light-04 are also significantly increased. Only MedMCAD-01 doesn’t seem to like the new driver release 12.10. Pentium 4’s score drops by some 20%. At the same time, the Athlon scores remain almost identical. It doesn’t seem as if the OpenGL-part of the new 12.10 driver release was in any way blessed with new Athlon optimizations.
The overall result of SPECviewperf shows what a difference the Pentium 4 optimizations of the new driver are able to make. While Pentium 4 used to be slower than Athlon in this test before, it is now able to beat Athlon in all but one sub-benchmark.
2D Benchmarks — Sysmark 2001
The new driver release 12.10 did not show any impact on the Sysmark 2001 scores of either of the two processors, which is why we decided to omit the results.
Conclusion
The benchmark scores generated with the 12.10-driver may not display as much of a difference to the scores of driver 6.67 as some of you might have expected. However, we were clearly able to see what difference the Pentium 4 optimizations are able to make. NVIDIA’s implementation of P4-enhancements into their new graphics card drivers could be seen as a first step into the era of P4-optimized software. Intel can only hope that others will soon follow NVIDIA’s example. You can’t blame Intel of laziness. Intel’s excellent tools as well as documentations required for the generation of P4-enhanced code are available for quite some time. Once the majority of software has been tailored to Pentium 4, we will finally see the benefits of Pentium 4’s architecture. If Intel continues its aggressive pricing policy along with this development, Pentium 4 could become a lot more attractive. AMD needs to be aware of this situation. It has to ensure that Athlon enhancement tools will make their way into the offices of software developers as well. This will be particularly important for upcoming AMD-processors with a new architecture, such as e.g. the Hammer-line.
Let’s see what happens next. We all know «one swallow does not make spring.» It might well be that we will have to wait a long while until another important software will actually become P4-enhanced. Until then Pentium 4 will have a tough standing against AMD’s very attractively priced Athlon processor, which is performing extremely well with today’s applications.
