3DLabs Launching New GPU

h0tblack writes "...or VPU as they've seen fit to call it. The Register is reporting that 3DLabs will be releasing the P10 later this year. It's targeted at workstation and gaming markets with OpenGl2.0 and DX9 drivers having been seeded to developers already. Could be interesting as 3DLabs have been one of the key players in the development of OpenGL2.0. The P10 has over 200 SIMD processors throughout its geometry, texture and pixel processing pipeline stages to deliver over 170Gflops and one TeraOp of programmable graphics performance together with a full 256-bit DDR memory interface for up to 20GBytes/sec of memory bandwidth. More info can be found in the press release." There are also examinations of the new chip on Anandtech, Tom's Hardware, and no doubt many other hardware sites too.

Comments

Could be interesting, but

[dreamchaser]

Press releases can't render anything, to the best of my knowledge. I'll reserve judgement until I get my hands on a review unit. However, this can only be a good thing. Competition drives prices down and features up.

Re:High-End Video Cards

[Quasar1999]

This is an article about two unreleased graphics cards, one $600, one $900. No, that's not a typo, these graphics cards cost as much as a nice Athlon system. These aren't targetted at gamers, they're targeted at workstation users, and fuckwits.

Speak for yourself, I'm a gamer, and I'm more than willing to fork out $900 for a good video card. Hell if I spent $700 on the Geforce1 DDR when it first came out, why the hell not spend $900 on a fully opengl accelerated card? I've seen the current generation of High end cards from 3DLabs, and if this new generation is anything like the current, it's worth the $900 for gamers.

How long til we see THIS Slashdot article?

[Matey-O]

Extra! Extra! Linux ported to GPU!

Really, these things are getting massively more complicated than your ordinary P4 or Athlon.

And think; There's one less layer between the OS and the framebuffer!

Re:How long til we see THIS Slashdot article?

[Junks Jerzey]

Really, these things are getting massively more complicated than your ordinary P4 or Athlon.

Not really, though. They have simple units, then they put a whole bunch of them on there. They don't need nonsense like branch prediction and register renaming and all that. But they certainly are complicated in their own way.

Beyond3d

[linzeal]

Beyond3d, home to many respected (and notorious ) workers at various 3d companies such as nvidia, ati, and bitboys are discussing this right now.

A question...

[levik]

I mean I understand that the graphics market right now is hotter than the 1980s arms race, with companies trying to one-up each other constantly... But can somebody tell me if there are products currently on the market that take full advantage of the *current* crop of video cards?

When Geforce3 came out it didn't have much of a clock speed increase, but boasted features that if taken advantage of by the developers would make the games look *MUCH* better. And yet, the only trend in the gaming industry that I've spotted is cranking up the poly counts.

Re:A question...

[Junks Jerzey]

I mean I understand that the graphics market right now is hotter than the 1980s arms race, with companies trying to one-up each other constantly...

That describes the market a few years ago, but no more. These days, with GeForce 2 MXs being dirt cheap and no one having performance issues with them, no one--except neurotic geeks--gives any thought to updating their video cards.

But can somebody tell me if there are products currently on the market that take full advantage of the *current* crop of video cards?

The answer is an emphatic "no." I'm a game developer, and we were focusing on the Voodoo 2 as the low end until very recently. And the Voodoo 2 is still a much more powerful card than people realize, providing you work *with* it and don't just ask it to render 50,000 polygons per frame. I don't think we ever got to the bottom of the performance available in that card, and we certainly, certainly, never got anywhere near what you can really do with later cards, like the original GeForce. All of the fancy stuff you can do with the GeForce 3--mostly based around vertex shaders--is not backward compatible with 90% of the market, so we never touched it.

Fanboys don't want to hear that their cards aren't being pushed anywhere near the limits. The are much happier to have poorly written games that have high polygon counts and bad art, because then they can justify the money they spent on a new computer and/or video card.

Re:A question...

[Geek In Training]

When Geforce3 came out it didn't have much of a clock speed increase

Not over the GF2 Ultra series, but it was a pretty big jump from the MX and GTS cards most people had. In addition to the HUGE FPS jump in games like Quake III, it had all those eye-candy programmable things that are going into things like Aquanox and The New Doom (tm). Also, the memory increase to 64 then 128 megs of DDR graphics RAM allows for insanely better Anti-Aliasing at "normal" gaming resolutions like 1024x768. The NV25 core (GF4 Ti series) increases this further, where you can turn on full-scene anti-aliasing and still get killer performance in your old games.

I only play Quake 3 and RTCWolfenstein on a regular basis, but my GF2 GTS (on an Athlon XP 1600+) pushes a masochistic 0.3 FPS in Quake 3 demos with 4xFSAA. Testing with the new card (128 megs of 600MHz graphics RAM, I never could have imagined in 1999) shows that I'll turn on 8 way Aniso, 4xFSAA and STILL gank 60fps on my 1024x768 LCD. Starting at $199, which is my limit for a graphics card.

And trust me, there is a TON of difference in visual quality with 4xFSAA on using a 15" LCD.

So yes, the programmable pixel shading pales against the power of prettier pictures in your "old stand-by" games, like Q3A. (Alliteration is your friend.)

Re:A question...

[Rolker]

It's not a less decent monitor, it's a monitor with built-in natural anti-aliasing!

Re:High-End Video Cards

[4of12]

No, that's not a typo, these graphics cards cost as much as a nice Athlon system.

I don't care. It's still a lot cheaper than a top of the line SGI workstation.

The ratio of costs for all the parts in a typical PC
(motherboard:CPU:disk:powersupply:OS:graphicscard)
have shifted some over the years. More accurately, though, as the performnce of certain keys pieces has increased to adequately fulfill the needs of the users, it's natural to start looking to satisfy unmet needs.

An OpenGL card like this would be wonderful for scientific visualization, CAD, CAM, etc.

While the price is an important point, in my market $600-$900 is not a big deal.

Creative has bought 3d labs

[gargle]

It's worth pointing out that Creative has bought 3d labs, and Creative's CEO Sim Wong Hoo has every intention of taking 3d Labs out on an aggressive push into the consumer 3d market. See article.

no fan?

[geektweaked.com]

how about we just give you a GF2 with the XF86_SVGA server and we'll just RIP THE FAN OFF FOR YOU!

there you go! you're welcome!

oh, you DIDN'T want it to catch on fire...

-c

I'm waiting for return to bus-based computing

[swb]

...where all you have are CPU cards with whatever specialized adapter is necessary to provide the apporpriate electrical connectivity to peripherals.

Each card is a basically a CPU board with its own memory. The common bus between cards is really a switch to limit card-card contention. One card is the bus master running the kernel. Processes can be shuttled between CPU boards as processing power is available.

The thing is we're getting to the point where just about every PCI device has a CPU on it (NICs with encryption/acceleration engines, RAID cards). Why not just put high-speed general purpose CPUs on the cards and use it as a highly integratable/segmentable cluster?

The actual kernel could do more scheduling and less work, since the "NIC" CPU card could theoretically run large parts of the IP stack in addition to the NIC driver, as an example.

TI 34010...

[BaronM]

...was the first PC-market, full programmable graphics chip, as far as I know.
Any website proclaiming full programmability as new or revolutionaly is simply demonstrating a lack of historical knowledge. 34010/34020 based boards competed with the first-gen fixed function graphics accelerators for Win 3.x, but couldn't compete on price/performance with the fixed function BitBLT engines from S3 et al, and the flexibility of being fully programmable meant nothing to PC users who were accustomed to dumb EGA/VGA cards.

Re:TI 34010...

[morcheeba]

That brings back memories!

The 34010 kicked butt! It was used by Atari's Hard-driving game. It had a lot of neat features, including hardware X/Y addressing (i.e. move x,y,pixel), bit-level addressing (you could twiddle any bit in memory, or write a word/byte on any boundry), and built-in simple graphics operations (copy a block of memory, xor source & destination, use larger of the two, subtract, union, difference, add but don't overflow, etc)

But what was *REALLY* cool was the math coprocessor, the 34020. It was blazingly fast (almost, but not quite as fast as the industry-crushing i860 IIRC), but it featured a programmable microcode so you could create your own instructions and get every ounce of performance out of the machine. I'm still looking for a processor that will allow that... we're getting those with modern NPUs (cradle, intel IXP1200), but these generally lack floating point functionality.

This is what OpenGL 2.0 is about

[marm]

OpenGL 2.0 addresses exactly your concerns - a vendor-neutral shader programming language, and this is precisely why you're seeing 3Dlabs pushing hard for it. It seems they will be first to market with a fully programmable graphics pipeline, and they need the software technology to go with it...

DirectX 9 also addresses the same issues and provides a standard shader language (actually DirectX 8.1 has a standard shader language already, but it lacks a certain amount of the programmability that will be present in DirectX 9), but there are a lot of reasons for the graphics card vendors to favour OpenGL over DirectX. For instance:

• There are a lot of users of high-end 3D hardware for whom Windows is anathema. Think about all the effects shops that traditionally have used IRIX and are now moving over to Linux... DirectX ties the cards to Windows, OpenGL does not. This is a growing, and more importantly, prestige market for high-end PC 3D vendors... Linux is bringing them to the PC from SGI/IRIX solutions, and is bringing them sales with it. I think NVIDIA understand this one, just a shame few of the other 3D vendors do yet...
• There are an awful lot of 3D apps that are heavily tied into OpenGL and rewriting them for DirectX would be a serious undertaking, whilst modifying them for OpenGL 2.0 to take advantage of the new shader features and extra programmability of the graphics pipeline will be a relatively simple task in comparison.
• What if Microsoft decided to get into the 3D market by buying one of the existing major players? Sure, Microsoft might be responsive to the 3D vendors now, but I suspect they wouldn't be if they had a vested interest in one of the players. Perhaps it seems unlikely, but it seems Microsoft has ambitions in the hardware business - witness the X-Box. It's a doomsday scenario from the point of view of the 3D vendors, sure, but no doubt it's something that a few vendors have thought about.
• Even if Microsoft doesn't do such a thing, OpenGL allows them 3D vendors room to breathe - they can implement new features as they please without Microsoft having to give them the nod.

Hopefully OpenGL 2.0 will see a resurgence in OpenGL use. I don't like the idea of the 3D market being controlled by Microsoft, and I don't think the 3D vendors do either. Kudos to 3DLabs for leading the way!

new kernel option?

[orangesquid]

And the kernel option of the future is...

Processor type and features
...
Floating point emulation? [y/N]
Floating point acceleration via 3dLabs VPU? [Y/n]