September 1 - 4, 1998
Archive

Performance: Better than Pentium II? We were pleasantly surprised by the benchmark test scores achieved by the four 333-MHz Celeron PCs in this roundup: the Compaq Deskpro EP C333/6400/CDS, the Dell Dimension XPS V333c, the HP Brio 7100 Series, and the Micron Millennia C333. The four machines--each with Windows 98 and 64MB of RAM--posted ZD Business Winstone 98 scores that were, on average, 56 percent better than those of the average Celeron/266, 35 percent better than those of a Pentium II/266, and even 7 percent better than those of a Pentium II/300.

 

I was just back from COMDEX 98 (a exhibition of PC products in Singapore ), guess what, I saw Creative selling thier Creative Blaster Riva TNT Card. WOW, talk about being sneaky. Here's some info :

1) It doesn't have TV-Out.
2) Its 16MB AGP CT-6711. (Oh, there is a PCI Version CT6701 )
3) It comes with Incoming, Forsaken and 2 other utilities.
4) It comes with a MPEG-1 Player, no MPEG-2 player though.
5) Comes with 95, 98, NT4, NT5 drivers. (Thats what it says on the box)
6) OpenGL ICD for 95 and NT.
7) There is an expension adaptor on the card for "future expansion".

An interesting note is that the System Requirements says: >=P90 for PCI version, PII or compatible for AGP version.

 

New to the 620 is support for an 8-Mbyte frame buffer and a 230-MHz RAMDAC with a resolution of 1,600 x 1,200 x 32 bits per pixel at 85 Hz. Other new features are an improved 2-D acceleration engine with stylish pattern, NT fractional format and multiple-scan-line drawing capabilities, trapezoid fill and DirectX 6.0 and OpenGL ICD support.

The 620 features a proprietary Super AGP architecture for video data transfer. Super AGP is the internal bus architecture for improving the integrated VGA core performance. It features dual 64-bit data buses, one for the VGA and host interface and the other for the VGA and system DRAM interface. The frequencies on both buses can run at 100 MHz with an 800-Mbyte/second bandwidth.

 

NVIDIA's RIVA TNT is certainly the most exciting product in this test and it made many of NVIDIA's recent claims come true. The 6 million triangles/s claim however seems wishful thinking rather than actual truth, as you can see when looking at the results at 640x480 and compare them with the Voodoo2 results. At this low resolution the frame rate is not fill rate, but polygon rate limited and that's exactly where the Voodoo2 looks better. 3Dfx only claims 3.3 million triangles per second for Voodoo2 though. Nevertheless TNT's 3D performance is very good and the excellent image quality makes this chip a pleasure to play with. It was very impressive to see that TNT seems to have the best AGP interface to date, beating S3's Savage3D in their own demo and that although TNT doesn't use texture compression. There is a big downside for many of you however, since TNT does not like to be driven by a slow CPU at all. Owners of cacheless Celerons and all owners of non-3DNow! Socket7 CPUs should really get a different 3D chip, e.g. the Savage3D or upgrade to a 100 MHz Pentium II. Even the support for 3Dnow! is not yet existing and we can only hope that it will be fast and available soon.

 

It really looks like the K7 has the capability to be a killer chip.  Based on a .25 fabrication process, the K7 has to prove its worth in the market. The plan at AMD is to deliver the K7 processor in a module that is physically, but not electrically, interchangeable with Intel's Slot 1 Architecture. The K7 will run at clock speeds of 500MHz and 550MHz with a 200mhz bus when it is introduced sometime in next year's second quarter, according to AMD's roadmap. AMD has stated that the K7 will support Symmetric Multi-Processing (SMP). If this becomes a reality, the K7 will be the first non-Intel x86 chip that is able to use more than one processor in a system. 

 

If these guys play their cards right they could be the next Canopus of the 3D-accelerator market. There's no denying the Wicked 3D is a unique product that stands out ahead of the majority of the competition's Voodoo 2-driven products. Being a company of engineers who can create top-notch drivers has really helped them out a lot.

 

As far as image quality goes, this is a 2 sided street. On the 2D side of the issue, I don't think that any benchmark number in the worlds can explain 2D quality. I use a Viewsonic G series 17 inch monitor. Although the monitor I use is not capable of of reaching the proposed limit of 1600x1200 of the BEAST, I was able, for the first time I might add, to get my desktop running at 1280x1024 at 70Hz. Considering that I had never before looked at a 2D/3D part that allowed me to run the desktop at this resolution, I was amazed. On the 3D side of the coin, things were a little different. Obviously 32bit textures were much cleaner and clearer than 16bit ones. This board did enable some resolutions that I had never been able to achieve in my games before. As far as S3TC goes, I did notice a slightly "grainy" feel to the textures, but I assure you it was very hard to notice at 1027x768 and above. Still, the slight grainy look was present, and is a side effect of the S3 texture Compression scheme. I am unsure of whether this was an artifact of texture conversion or not. I can only assume that if that is the case, Direct X 6.0 games that build in support for the the S3TC scheme will not have this grainy feel.

 

The Celeron 333A came in today... unfortunately though, I'm not very impressed.

Don't get me wrong - it's a great chip, but doesn't overclock as well as the Celeron 300A, and is definitely not worth the extra money. At 500MHz (5x100), it will boot into Windows 98 and everything will work fine, but there will be some random crashes (tried from 2.00v to 2.30v).

The few crashes I can live with - what I can't live with is that fact that to get that far, I have to disable the L2 cache. When running at 500MHz with L2 cache enabled, the system will boot into Windows 98 (only with 2.30v) but won't go for a minute or two without crashing. I'm guessing heat is the problem. I'll be ordering a ComputerNerd dual-fan cooling unit for it, then try 500MHz again. For now though, I'm running at the next logical step - 416MHz (83x5)... unfortunately, the BH6 doesn't support the 90 or 95MHz bus speeds, so that's the highest speed the Celeron 333A will run stable at on the BH6.

Intel Corp. is hoping to ignite sweeping changes in the PC audio and motherboard businesses with new hardware specifications to be studied in depth at the Intel Developer Forum here later this month.

Initially announced in July, the audio/modem riser (AMR) card and the mobile daughtercard (MDC) will combine audio and modem functions, and will remove from the motherboard the analog circuitry for both. Intel will announce completion of the specifications and begin its promotional push for AMR and MDC next week.

A number of early AMR-compliant products will be announced next week, including the RipTide chip set from Rockwell Semiconductor Systems (Newport Beach, Calif.). Volume shipments of AMR systems are not likely until the second half of 1999, said Russ Hampsten, Intel's audio marketing manager.

 

Will CL's TNT be a reference design? I want to be certain I can use reference drivers on the TNT that shall reside in my AGP slot.

To which Dylan replied:

No, the GB Riva TNT does not use the reference PCB design. We've completely redesigned the PCB for better heat dissapation, and to use CAS2 memory refresh instead of the default CAS3. This yields better performance.

However, none of these PCB changes should stop nVidia's reference display drivers from working just fine on our card.

 

If you're still using an old ISA card or the generic sound board that came with your PC, you owe yourself an upgrade, and this card is probably one of the best ways to do it. It supports the popular A3D API for 3D audio in games, as well as DirectSound3D. Its solid performance and relatively low price make it a great contender for king of the Vortex-based boards.

One note of advice, however: this card uses the Vortex chip for sound, but if you'd like to be on the cutting edge, you might prefer to wait just a few months for Aureal's Vortex 2 to hit the shelves, since it will provide everything here and a lot more.

 

Some of the feature that differ from the other RivaTNT

1)  Allow DIGITAL internal connection from a Pure3D II (LX) to the Riva TNT. 
     GLIDE image is displayed with the Spectra
2)  Allow Analog display through the Spectra 2500 with ANY 3Dfx Card.
3)  Non-Reference design.
4)  TV Out includes S-Video and Composite RCA
5)  Heat-Sink & Fan for extra cooling.

Click here for a picture of the card

 

Raven 16MB Voodoo Banshee AGP - $149.99
MANUFACTURER: QUANTUM 3D
AVAILABILITY: 09/30/1998
DELIVERY TIME: 2-4 Day Delivery from release.

Download the latest powerstrip; Whilst it's downloading, read this report from Doctor H. regarding his findings on G200 overclocking. The following frame rates are with regard to Quake2 using the wrapper.

Normal, after bootup: 47.9 fps (higher than 100mhz clock)
84mhz clock: 38.5 fps (this is the clock setting that seems to be the DEFAULT*, though 100mhz is the true DEFAULT*)
90mhz clock: 41.9 fps
95mhz clock: 44.5 fps
100mhz clock: 47.0 fps
105mhz clock: 49.4 fps
110mhz clock: 51.7 fps
115mhz clock: 54.0 fps
116mhz clock: 54.4 fps
117mhz clock: Quake2 locked up in 2nd timedemo trial (though Win98 runs fine).
118mhz clock: Quake2 locks up earlier (Win98 still fine).
119mhz clock: Quake2 locks up on first frame (Win98 still OK).
120mhz clock: NOT POSSIBLE due to program limitation (I believe).

Thanks to Alexey Yarmolenko for confirmation with some more diverse testing.

There are some numbers: (89mhz /119mhz )
D3dtest,fillrate (50/68) mps
Quake1 (60/78)fps
Final reality (4.51/4.61)rmark
Incoming 640x480x16 (49.47/64.06)
Incoming 800x600x32 (29.74/36.27)

 

As long as you're not limiting yourself to the VIA MVP3 Chipset, the ASUS P5A easily carries its weight among the three most recommended ATX Super7 Motherboards.  How well will the sales of the PA-2013 and the Photon 100HC fare now that the P5A is a part of the show?  There is no way to predict, but be aware of the fact that ASUS has done a tremendous job with their newly released Super7 line of motherboards.  Let's just hope that the release of the P5A/-B isn't too late to help AMD's K6-2 sales.

 

Don't discount 3DNow! as just advertising (like MMX was).

New Win98/95 Matrox 3dnow! drivers (8/31/98 version - Vsync off):

(At default G200 timing)
Forsaken (640x480) : Avg: 80fps Peaks: 150fps
(800x600) : Avg: 70fps Peaks: 120fps
(1024x768): I didn't test

(At 110mhz G200 timing (set with a hacked PowerStrip))
Forsaken (640x480) : 131fps 178fps
(800x600) : 90fps 116fps
(1024x768): 65fps 82fps

This is just on a AMD K6-2/300 at 100mhz bus on an ASUS P5a motherboard. Only 8MB on the card, hopefully going to 16MB soon.

For complete system stats, go here.

Perhaps the high price tag will put many interested buyers on hold. But with all the high end features that rivals even professional home theater equipment all packed onto this tiny PCB, the value of the card is unmatched by any card in the market. Also, the fidelity of the card has absolutely no competition.

Creative's Environmental Audio Extensions will open new doors in 3D audio modeling. With the pledged support of many game developers keen on delivering new games with EAX support, and the possibilities of improvements like the inclusion of AC3 decoding, the Live! will not only be the ultimate choice for gamers but it will also be very appealing to audiophiles as well.

 

ASUSTeK will be releasing a RIVA TNT based board and will be shipping in volume next month, but samples will be available in about 2 weeks. Maybe this schedule will change if there is high demand for this card, so we'll see what happens....

There will be 4 different board configurations as follows:-

AGP-V3400TNT/8MB(SGRAM)
AGP-V3400TNT/TV/8MB(SGRAM)
AGP-V3400TNT/16MB(SDRAM)
AGP-V3400TNT/TV/16MB(SDRAM)

Here's a couple of quick Q&A's regarding this NEW TNT Based Card from ASUSTeK:-

1. Is there any performance down grade by using the SDRAM in your TNT board?

The answer is "No". Because there is no signal connected to the block-write pin in SGRAM, the performance is the same between these two versions.

2. Why ASUS have two versions 8MB/SGRAM, 16MB/SDRAM?

This is due to the data-width of RIVA TNT. Because the TNT interfaces with memory by 128-bit and the current available SDRAM is 1*16Mbit and SGRAM is 512*32Mbit, we have to make most use the 128-bit performance by use 8 pieces SDRAM(16*8=128) or 4 pieces SGRAM(32*4=128). These 128-bit designs make sure both versions get the same performance.

 

We asked:
Is it possible for the Voodoo2 chip-set to be slower using two texture units rather than one? Suppose a game required more than 4 megabytes of texture memory for the first texturing (detail rendering). If the V2 was using both it's TMU's it would be forced into loading some textures across the pci-bus during gameplay. Is there a possibility that in some circumstances this would be slower than using a single TMU with all textures loaded locally? Does this situation exist with Epic's 'Unreal'?

Jeff White from 3Dfx answered:
While it is possible, in most cases the second TMU is used for such things as light maps and fog maps.  These are very small textures that do not amount to large bandwidth issues.  Trying to do lightmaps with one TMU would be slower than doing it with two TMUs.  The light map could be stored in the second TMU's memory and would never need to be uploaded again, while just using a single TMU, it would have to be loaded every frame.  In an extreme case, if both memory pools were full, then using two TMUs would still be faster because only one set of triangles would need to be sent to the card.    Using a single TMU, two sets would need to be sent.  With the same features turned on in a game, then the two TMUs will win every time.

 

Not bad for a CPU that costs $150. It's plain to see that Intel has accomplished all of their goals with the Celeron 300 A, it's a fantastic CPU. We're talking super low cost for a super high return, as well as a super-extreme overclocking ability here. It's important to note that Cyrellis used an "OEM" version of the Celeron 300 A for testing, meaning that this is the type of CPU that will come equipped in pre-assembled PCs from various vendors ...

Otherwise, Intel has definitely hit one out of the park with the Celeron A, and it appears that even the Celeron brand name itself will be rescued from the "mediocre" definition it earned from the earlier Celerons, and will finally stand for what Intel intended it to: High performance, and Low cost.