VIA to Launch Another "Paper" GPU This Year?

alexsok

Regular
Taken from x-bit labs:
VIA has already announced its Savage XP/AlphaChrome graphics processors in June, however, right after the announcement decided to redesign the chip as there were a lot of problems with it. Now it is September, almost three month after the announcement and we still have no graphics cards based on the GPUs from VIA/S3 Graphics, hence, here is just another “paper launchâ€, when the device is announced, but its shipments have not begun yet and are not going to in the nearest future. It is ridiculous, but according to this web-site, the company may launch another “paper†graphics chip this year. The highly anticipated code-named Columbia GPU is set to be sampled and possibly introduced in the fourth quarter 2002.

The Columbia graphics processor has 8 rendering pipelines with 1 TMU per each, supports DirectX 9.0 functions, including Pixel Shaders 2.0 and Vertex Shaders 2.0. The chip is targeted to run at 300 - 400MHz and is going to be manufactured using TSMC’s 0.13 micron technology. I believe the GPU is more likely to work at 300 or 340MHz rather than at 360 to 400MHz due to the fact that VIA has no experience in creating such complex chips. Unofficial sources claim that graphics cards based on the upcoming GPU will be able to achieve about 11000 – 15000 points in 3DMark 2001SE benchmark. Although it is impossible to predict real-world application’s performance based on 3DMark score only, but even knowing that ATI’s RADEON 9700 PRO reaches about 12000 marks, I doubt S3/VIA will be able to beat them not only in benchmarks, but in the real games as well.

Keeping in mind that even NVIDIA’s NV30 is delayed due to the problems with 0.13 micron manufacturing technologies, and TSMC apparently changed the design rules, it is very unlikely that Columbia will appear earlier than in February or even March 2003. Talking about the long-term plans of the company it should be pointed out that they plan to reveal their Columbia 2 graphics core by the year 2004.

The one thing I should stress is that graphics cores developed by VIA and SiS are proposed for integration in their upcoming North-Bridges. The discrete GPU market-share narrows and there is very little space there for only two or three companies. Integrated chipset market grows significantly and since VIA and SiS are already quite strong players there, they simply have to introduce more advanced integrated solutions. As a result, do not expect their discrete GPUs to beat the products from ATI or NVIDIA.
 
I just have to say that from the history of S3 from the Virge to the Savage 3D to the Savage4 to the Savage2000, every single one of their graphics chips has sucked majorly, in one way or another. I really do not expect S3 (or, rather, I suppose, the graphics division that formally worked at S3) to pull a rabbit out of their hat now.
 
Chalnoth:

Now now, I actually quite liked my 864 with 2MB of vram. It was quite a fast chip/board for the time.

But everything since, I pretty much agree with. ;)

Nite_Hawk
 
The one thing they delivered was Texture Compression, which every 3dmanufacturer has adobted. So not all things they have done are bad. Savage3D had some serious cool things, did anyone see the special Unreal version? But it was always too late and clock speeds were behind the rest of the 3dcards out there...
 
Still doesn't change the fact that every 3D product they put out was crap. Yes, they did deliver TC...but that's nothing compared to what other manufacturers have put forth, particularly nVidia.
 
I myself don't believe anything useful will come out from this new chip...

Savage2000 was probably their best chip (it was doing better than GF1SDR at some tests back then), but the drivers sucked.

I really don't see them offering any competition to nVidia or ATI.
 
Let's hope it uses PVR tech. It's been too long since we heard anything from Img... how about a DFR with Metagence DSPs for vertex and pixel processing...

/me daydreams on
 
marco said:
The one thing they delivered was Texture Compression, which every 3dmanufacturer has adobted. So not all things they have done are bad. Savage3D had some serious cool things, did anyone see the special Unreal version? But it was always too late and clock speeds were behind the rest of the 3dcards out there...

Pity they didn't embed METAL API into S3TC. That way when MS licensed DXT1 S3 metal could've survived!
 
alexsok said:
Savage2000 was probably their best chip (it was doing better than GF1SDR at some tests back then), but the drivers sucked.
You need to turn that around :)

The driver teams did a pretty good job considering the limitations of that piece of hardware. If they had been allowed to concentrate on the chip's strengths then it would have been a pretty good part, management had other ideas. Then they all left to set up SonicBlue.
 
OVERLORD said:
Pity they didn't embed METAL API into S3TC. That way when MS licensed DXT1 S3 metal could've survived!
Metal wasn't anything special :)

It was very useful for prototyping and the Unreal / UT ports, but D3D did everything it did and apart from needing longer primitive batches wasn't that different in performance.
 
Dio said:
OVERLORD said:
Pity they didn't embed METAL API into S3TC. That way when MS licensed DXT1 S3 metal could've survived!
Metal wasn't anything special :)

It was very useful for prototyping and the Unreal / UT ports, but D3D did everything it did and apart from needing longer primitive batches wasn't that different in performance.

Their metal drivers looked and performed better than their dx drivers ever did.

Seeing that Tim Sweeney went to all that trouble creating extra textures CD for UT you'd think S3 would've had their native API ready for SAVAGE2000.

But instead some ignoramous seeked to drop this feature. Savage2000 was stillborn @ launch. No killer ap on release. S3 fumbled and never recovered, only releasing metal later. Their relationship with software developers just stunk!

UT on METAL still one of the best experiences I've seen on PC platform. Their 16bit metal still puts some 32bit displays to shame.

S3 driver team sure have good experience killing hardware development. Hope the same ain't true for COLUMBIA!
 
Savage2000 was very bug-ridden peace of silicon. If anything S3 driver team deserves credit for making it work at all.
 
Geeforcer said:
Savage2000 was very bug-ridden peace of silicon. If anything S3 driver team deserves credit for making it work at all.

Donno, maybe...

It's just that most reviews stated that problems lied mostly in the drivers and not in the hardware.

Perhaps they were wrong...
 
alexsok said:
Geeforcer said:
Savage2000 was very bug-ridden peace of silicon. If anything S3 driver team deserves credit for making it work at all.

Donno, maybe...

It's just that most reviews stated that problems lied mostly in the drivers and not in the hardware.

Perhaps they were wrong...

Since when do most reviewers have a clue, let alone an insight in the innerworking of hardware?

There people here who can tell you more then I... but S3s PR blamed many of the hardware problems on the software. Take the notorious T&L unit - it was broken in hardware, but S3 used drivers as an excuse. "Hardware is allright, just wait for the new drivers"...
 
Take the notorious T&L unit - it was broken in hardware, but S3 used drivers as an excuse. "Hardware is allright, just wait for the new drivers"...

yeah, i was suprised about that statement myself!
 
Geeforcer said:
Savage2000 was very bug-ridden peace of silicon. If anything S3 driver team deserves credit for making it work at all.

As far as I'm concerned, Geeforcer, it could have been either.

Still, the same company was in control of both the driver development and the hardware development, so the blame still rests solely in one place: S3. To place the blame any more specifically than that, you'd have to be within S3, and it's a job that S3's management should have been tasked with...
 
OVERLORD said:
Their metal drivers looked and performed better than their dx drivers ever did.
Mainly because the features that were supported were more limited. Only features that games were using had to be supported, not some obscure feature that just causes misery.
Seeing that Tim Sweeney went to all that trouble creating extra textures CD for UT you'd think S3 would've had their native API ready for SAVAGE2000.
No one ever accused S3's management of being too smart.
S3 driver team sure have good experience killing hardware development. Hope the same ain't true for COLUMBIA!
You're blaming S3's hardware problems on the driver team? That doesn't make any sense whatsoever.
 
alexsok said:
It's just that most reviews stated that problems lied mostly in the drivers and not in the hardware.

Perhaps they were wrong...

It depends on the scenario (feature set being used). Generally, problems existed in both areas, and finger pointing one or the other actually has little bearing on the end result.
 
Back
Top