ATI's 1GB Video Card

Signify writes "ATI recently released pics and info about it's upcoming FireGL V7350 graphics card. The card features 1GB of GDDR3 Memory and a workstation graphics accelerator. From the article: 'The high clock rates of these new graphics cards, combined with full 128-bit precision and extremely high levels of parallel processing, result in floating point processing power that exceeds a 3GHz Pentium processor by a staggering seven times, claims the company.'"

Comments

use as a cpu?

[Toba82]

Why doesn't ATi (or nVidia for that matter) make CPUs?

They obviously could make some very powerful chips.

Re:use as a cpu?

[Tyler Eaves]

Because for most general computing tasks, floating point doesn't matter.

Re:use as a cpu?

[Kenshin]

I'm thinking:

a) Tough market to crack. AMD's been around for years, and they're still trying to gain significant ground on Intel. (As in mindshare.) May as well spend the effort battling each other to remain at the top of their field, rather than risk losing focus and faltering.

b) These chips are specialised for graphics processing. Just because you can make a kick-ass sports car, doesn't mean you can make a decent minivan.

Re:use as a cpu?

[TheRaven64]

Building a GPU is trivially easy relative to building a CPU. Here are a few reasons why:

• You have an OpenGL driver for the GPU and a JIT for shader language programs. This means you can completely throw out the instruction set between minor revisions if you want to. An x86 CPU must mimic bugs in the 486 to be compatible with software that relied on them.
• You have an easy problem. Graphics processing is embarrassingly parallel. You can pretty much render every pixel in your scene independently[1]. This means that you can almost double the performance simply by doubling the number of execution units. To see how well this works for general purpose code, see Itanium.
• The code you are running is fairly deterministic and unbranching. Up until a year or two ago, GPUs didn't even support branch instructions. If you needed a branch, you executed both code paths and threw the result you didn't need away. Now, branches exist, but they are very expensive. This doesn't matter, since they are only used every few thousand cycles. In contrast general purpose code has (on average) one branch every 7 cycles.

GPUs and CPUs are very different animals. If all you want is floating point performance, then you can get a large FPGA and program it as an enormous array of FPUs. This will give you many times the maximum theoretical floating point throughput of a 3GHz P4, but will be almost completely useful for over 99% of tasks.

[1] True of ray tracing. Almost true of current graphics techniques.

Re:use as a cpu?

[pchan-]

However, how difficult would it be to write an operating system that offloaded floating point operations to the GPU, and everything else to the CPU.

Funny you should mention that. The Intel 386 (and up) architecture has built in support for a floating point coprocessor, so it can offload floating point operations. In the early days, you could buy a 387 math coprocessor to accelerate floating point performance. Then Intel integrated the 387 coprocessor onto the 486 series cpus, and today we just know it as "the floating point unit" (although it's been much revised, parallelized, and super-scaled).

As for offloading to a GPU, well... that's what we do today. It's called Direct3D, or Mesa, or Glide, or your favorite 3D acceleration library. The problem with this approach is that it requires very specialized code. It's not something that can be automatically done for just any code, as the overhead of loading the GPU, setting up the data, and retrieving the results would far exceed the performance gains. In only extereme cases does it pay off: the workload has to be extremely parallelizable, with almost no branching and predictable calculations. Basically what it ends up is that the algorithm has to be extensively tailered to the GPU. Even IBM has had major issues offloading general purpose operations to their special processing units, and those are much more closely coupled to the CPU.

Re:use as a cpu?

[fbjon]

No, it's a 404: real linky

Re:use as a cpu?

[ajs318]

There doesn't seem to be a market for a new CPU architecture.

One could make a blisteringly fast, ultra-RISC processor that completes every R-to-M instruction in one clock cycle {read the instruction from memory on the rising edge ["tick"], let the logic matrix outputs stabilise during the high period and write the result to memory on the falling edge ["tock"]} and every M-to-R or M-to-M in two {an extra tick is needed to read the operand from memory and the intervening tock is wasted; unless you can read on both edges, then M-to-R can be done in one cycle, with the logic matrix stabilising during the low period. M-to-M still takes two and might be better "faked" using two instructions; M-to-R followed by R-to-M, which would keep the instruction set simpler}. You could even do several R-to-R instructions in parallel, if you had more than one logic matrix {or a partitionable logic matrix} and a register bank designed so as to support multiple independent reads and writes. But such a machine wouldn't be at all architecture-compatible with anything else, and you would need effectively to run a kind of interpreter or emulator all the time to deal with the instruction sets of existing processors.

And that is what the modern Athlon-type processors do anyway: a RISC processor is emulating an old CISC processor. I personally have a sort of gut feeling that this is unnecessarily slow: code that was optimised for the "outward" CISC instruction set may not be truly optimal on the underlying RISC processor, since it might occasionally have to take a diversion to do something which the CISC instruction specification required, but which was not necessary in this particular instance {e.g. setting a flag which will never be read}. Saving even apparently just a few clock cycles in this way might well result in a performance gain, especially if it can be done in a heavily-used loop -- particularly somewhere like the kernel, libc or the shell.

However, every new processor architecture -- or, for that matter, every new way of using an existing processor architecture -- requires recompilation of existing software. In the Open Source world, that's somewhere between trivial and a non-issue. But in the Closed Source world, it's a showstopper.

Re:use as a cpu?

[clydemaxwell]

because OP's 'obviously' statement seemed to imply he thought the current GPUs were a good marker of CPU quality -- i.e. he was impressed by the GPU specs, and so wondered why they don't make CPU's (7 times better than current) so GP was right in pointing out that it's not a very good indicator of CPU strength, and thus we have little reason to believe nvidia would make a good CPU. tada!

Re:use as a cpu?

[somersault]

the problem is that the graphics accelerators work in a very limited domain, so the gfx cards engineers can concentrate on the things that it needs to be fast at (floating point calculations, tranferring memory contents efficiently). Normal CPUs have a much wider scope, and while I'm sure the engineers that design/upgrade x86 processors do try their best to make the chips fast, they have to spread their work around, and make sure that every area is decent, rather than one area spectacular. Also graphics cards are fairly self contained, while processors have their motherboard/chipset (which for intel would include the memory controller) to dictate which type of RAM they can use, and how much bandwidth the whole system has etc.

The thing is that you couuuuld make an x86 that runs using GDDR3 etc, but it would be rather expensive, and nobody (well, no majority market anyway) is going to pay to produce that, if only a few thousand people can actually afford it. In time the costs will come down, but until then we common folk just have to stick with whatever AMD/Intel/Whoever are producing.

But anyway, the main point I made, maybe not in a very technically accurate way, was that it's easier to build something that performs well in one area, than to build something that does everything amazingly well (without costing the earth to buy it).

Re:use as a cpu?

[default luser]

Because the Pentium 4 was never made for powerful floating-point thoroughput using plain old x87 FP instructions. To give it better floating-point performance, a 128-bit vector unit (SIMD) was strapped on. With it, the Pentium 4 can perform 4 32-bit floating point operations every 2 cycles.

This is the one part of the Pentium 4 chip that is very much like the parallel execution units of a video card. If you want to maximize performance of the SIMD unit, you want a simple program with packed SIMD instructions. The key point is, the video card can do that better.

My wife gave me two thumbs up...

[Anonymous Coward]

...when I told her that I would buy an ATI card that would allow us to decrease the gas bill for our furnace next winter. Guys, you just have to give your better half a good argument and this graphics card is installed in your computer in no time. Just don't mention that you need to buy a better air conditioner to the summer... she'll discover that one. ;)

Hot hot hot!

[Anthony Boyd]

It's called the FireGL because it puts out heat at levels equivalent to a large fire. -T

Awesome!

[Rank_Tyro]

Now I can upgrade to Windows "Vista."

Re:Awesome!

[Jozer99]

Appreciate the joke, but for folks out there who think he is serious, Microsoft has said that the Intel GMA 900 and ATI Radeon X200 are the minimum graphics cards for using the "new" DirectX GUI. Vista will work on computers with less graphics systems, but in a compatability mode similar to Windows XP's GUI.

So?

[Tebriel]

Other than high-end graphics work, what the hell will this mean? Are you seriously saying that we will be seeing games needing that must video memory anytime soon? Hell, they have a hard enough time getting people to buy cards with 256 MB of RAM.

Re:So?

[Bitter and Cynical]

Other than high-end graphics work, what the hell will this mean?

Nothing. These cards are not meant for gaming, in fact if you did try and use it for gaming you'd be very upset. The FireGL line is a workstation card meant for things like CAD or Render farms that are very memory intensive and require a high level of precision. Its not meant for delivering a high frame rate and no gamer would stick this card in his machine

follow Nvidia into Physics?

[arjovenzia]

With all that beef behind them, i sure hope they will follow Nvidia (i actually have no doubt that they will) in offloading physics to the GPU. http://www.rojakpot.com/showarticle.aspx?artno=303 &pgno=0

it would be nice not having to purchase a top-notch CPU, GPU, and PPU (Physics Processing Unit) in the future, rolling the PPU and GPU together

Too bad its still an ATI...

[Jackie_Chan_Fan]

ATI's opengl drivers are flakey on their non firegl line of cards. Some suspect thats by design.

Graphic card makers should get with the program and stop releasing firegl's and quadros. Just release really kick ass 3d accelerators for all.

That way we can all have full opengl support and not the lame opengl game drivers by ATI. Nvidia's gaming card opengl drivers are better than ATIs

Re:Too bad its still an ATI...

[somersault]

Wouldnt you rather have a card/drives that have been designed for higher framerates, but with lower rendering precision? You dont need 128 bit floating point precision to play Quake. You're not going to care if a vertex on your health pack is 0.000005whatevers off to the left of where it should be, and you're not even going to see that onscreen unless you go in realllly close. I'm maybe oversimplifying things, but I tried a few years ago running Counter-Strike Source with a Quadro graphics card, then some Radeon that was lying around; the Radeon had much better framerates. It may have been a generation up from the Quadro, but it would still have been way cheaper to buy initially, and people are *not* going to pay for that 'kick ass 3D accelerator' when a more budget version does the same thing, but with less memory, and drivers that are optimised to make games fast, rather than very precisely rendered.

Re:what... teh.....fuk

[TheRaven64]

This is a workstation card, not a games card. The people buying this are likely to be either CAD/CAM people with models that are over 512MB (the workstation it plugs into will probably have a minimum of 8GB of RAM), or researches doing GPUPU things. To people in the second category, it's not a graphics card it's a very fast vector co-processor (think SSE/AltiVec, only a lot more so).

Re:ATI Sucks for driver support with Linux

[kitejumping]

thats why I buy nvidia... if the performance is the same, may as well have driver support.

Re:Not bad...

[KajiCo]

3D Artists, Game Developers, Scientists who do three dimensional simulations, Weather forcasting models, Engineers who do real time simulations.

They're not really for gaming as much as they are for developing stuff.

Re:Obligatory...

[atari2600]

You mean, it will run Duke Nukem Forever, forever

Re:Whoa.

[Jozer99]

Explained in detail above. Suffice it to say that CPUs and GPUs are radically different. With GPU's, ATI can throw out old architectures and create new ones whenever they want (quite often). Since the hardware is accessed by a driver, the user isn't limited in what programs they can use. With CPU's, everyone is stuck with x86, which was invented in the 1980s. You can't break compatability with x86. GPUs do mostly simple floating point calcuations. Therefore, they are basically massively parallel FPUs. If they need to do a non-floating point calculation, they are quite slow. CPUs can do floating point calculations, but also many other types of calculations, and are about equally good at everything. For the sake of heat, power consumption, size, and cost, the FPU on a CPU is not nearly as large as a GPU. If each processing unit on a CPU was the size/power of a specialized processor (GPU, ect...), the chip would be gigantic, and so would be hard to make, expensive to buy, consume massive amounts of power, and emit unimaginable heat.

Not likely OS level

[HornWumpus]

But you can write very specific GPU code to solve some parallel FP problems.

For more general purpose FPOPs you will have a hell of a time getting enough gain in floating point performance to overcome the overhead chatter between the CPU and GPU that would be required to keep the states in synch.

I'd go so far as to say if the process can't be near complely moved (the CPU will need to feed it data and suck up results) onto the GPU then don't bother.

But I'm talking out of my ass, it could work. I'm just skeptical.

Re:what... teh.....fuk

[temojen]

Stream processor, not vector processor. The programming models are different.

ATI works great for me with Linux

[deek]

Flogging generic statements like "ATI sucks for Linux", is not very accurate. A better way of putting it is "ATI sucks for some cards under Linux".

I can certainly say that my laptop, with its ATI Radeon Xpress 200M chip, works wonderfully under Linux. Yes, I'm talking about their binary driver distribution. Using the latest version of their drivers. I'm also using the Xorg 6.9 xserver. It's fully 3D accelerated, as shown in the following command:

$ glxinfo | grep OpenGL
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: RADEON XPRESS 200M Series SW TCL Generic
OpenGL version string: 2.0.5695 (8.23.7)

I'm aware that the binary driver doesn't work with some ATI cards, especially some of the top range ones. But for what I use, it's brilliant. The installer is a little easier than the Nvidia one too. Thanks ATI, you've done a great job, from my perspective.

Re:ATI works great for me with Linux

[yellowcord]

Try turning on the whiz bang hardware accelerated effects in KDE. You get a nice garbage screen if you turn on "Use translucency/shadows" in Window Behavior under desktop in the Control Center. Apparently it works great on the FOSS drivers for the Radeon 8500 and below but I get a screen full of garbage on my 9800. Probably not a big deal for most people but its very annoying to me.

Also frustrating is the lack of support for games played through Cedega, I just installed the Cedega timedemo tonight and Guild Wars almost but not quite works. I have characters missing heads and the frame rate is about 2 frames per second. I can get decent frame rates with a different setting but it crashes if I go to another district. This may be fixable, but so far my next computer is going to have an nVidia GPU.

so...

[Anubis350]

Are you saying it renders flames *very* realistically? :-P

Re:so...

[pcgabe]

Denholm: Nice screensaver... I love the way the smoke seems to be coming off the top of it. (The IT Crowd)

Can i reallocate that memory as system memory?

[mOOzilla]

Can i reallocate that memory as system memory?

Width of the floats?

[mosel-saar-ruwer]

Signify: full 128-bit precision

TheRaven64: or researches doing GPUPU things. To people in the second category, it's not a graphics card it's a very fast vector co-processor (think SSE/AltiVec, only a lot more so)

Traditionally, ATi floating point numbers were only 24-bits wide [i.e. only "three-quarters" of single precision, which is 32-bits].

nVidia, IBM Sony Cell, and Altivec support only 32-bit floats.

MMX supported no floats whatsoever. SSE supported 32-bit floats. SSE2 & SSE3 support 64-bit floats. x86 supports 80-bit floats.

So what is this 128-bit stuff all about?

I don't suppose there's a chance in hell that these could be quad-precision floats, could they?

Re:"workstation cards" what r they 4?

[Mithrandir]

I work a lot with the visualisation end of the market and recently have been working with NASA on the CEV project(s). Some models that we deal with are in the gigabyte file size just for the geometry for a single subassembly. This card would make viewing some of these things far easier as you can preprocess and schlepp almost all the geometry to the video card as a VBO and never have to pass it over the bus again. Makes for tremendous performance gains.

Re:what... teh.....fuk

[dbIII]

the workstation it plugs into will probably have a minimum of 8GB of RAM

Will it really need that much system memory? The first SGI workstation I saw had double the amount of video memory to system memory.

now if only they knew how to make drivers

[xamomike]

sounds great and all, but have they gotten around to paying their own programmers to make drivers that actually work, and install off the CD it comes with, instead of outsourcing it to a few guys in their basement?

Seriously, I've owned 6 different ATI cards of differing lines this year, and only 2 of them installed properly with the drivers that came on the CD. That just aint right.

Yes but

[Anonymous Coward]

Does it get you banned in World of Warcraft?

256 MB is small

[emarkp]

Try rendering medical image data as a 3D texture (well three textures actually, one for each primary image). With 300 images, 256KB per image, x3 textures, that comes out to 225MB just for the textures. I deal with datasets like these routinely, and more video memory is a welcome development.

Re:"workstation cards" what r they 4?

[LookoutforChris]

As a fellow Pro/ENGINEER user this is not my experience. What version are you using and how big are your models? The latest version is a hog (as always). I can't imagine using it on an old Dell with a FireGL and doing anything very complicated. I have to admit I'm not a fan of ATI cards, their OpenGL support seems to be very flaky. But I like the larger memory on these new cards and the price is good. Price wise this card would seem to compare favorably to a top model WildCat Realizm or a top model nVidia Quadro.

Not for gaming, for graphics workstations!!

[Anonymous Coward]

Ok, I cannot beleive the absurd number of posts I am seeing from lamers who think this thing is for video games. Hello People! Both ATI and NVidia have had seperate high-end workstation lines for years now! This is nothing new. Where have you people been?

This card is for people who need serious rendering of high detailed scenes and 3D objects, not serious frame rates for games. For applications where image quality, complexity, and accuracy are much more important than frame rate. The GPUs in these high end workstation cards are geared in a totaly different manner and actually suck for video games! These are great for CAD/CAM, medical imaging (like from CAT and EBT scanners), chemical modeling, and lots of other hard core scientific and 3D developement type stuff.

Finally!

[LookoutforChris]

That's 1GB of unified memory, so less than 1GB is available for textures ; (

It took them long enough; this is definitely the direction to go.

Almost 4 years ago Silicon Graphics gave a final revision hurrah to their best graphics product: InfiniteReality. A pipe sported 1GB dedicated texture memory, 10GB of frame buffer memory, 8 channels per pipe, and 192GB/s internal memory bandwidth.

And an Onyx system could have up to 16 pipes! That's 8.3M pixels per pipe, or 133M pixels from a full system! And all in 48-bit RGBA. And those are just the raw numbers, there were a great many high end features only found on InfiniteReality. Don't ask what it costs ; )

Sorry for the passionate post. It seems that Slashdot is very PC-ish and narrow in its viewpoint (Imagine a Beouwolf of... Can it run Doom3 ... etc.) so I couldn't resist blabbing about high-end kit that's off topic.

I've had the pleasure of using a small Onyx system. Too bad SGI is dead dead dead. Still they provide a good target for everyone to shoot for. Some day the above power will be available for a few hundred dollars for the average person. Though I think it will be atleast 5 years before the quality and features of InfiniteReality4 are at a consumer level. And never will we have workstations like SGI's again ; (

Re:Whoa.

[Ozymand E. Us]

Would be hard to make, expensive to buy, consume massive amounts of power, and emit unimaginable heat.

So...Intel?

*Ducks*

ATI loves you in the ass

[ThoreauHD]

http://wiki.duskglow.com/tiki-index.php?page=Open- Graphics
http://www.osnews.com/story.php?news_id=13844

If you are talking about XGI, ATI just bought them and closed the code.

http://www.linuxgames.com/news/feedback.php?identi ferID=8255&action=flatview

Re:Not bad...

[RingDev]

This is a workstation card, not a gamer's card. 1 gig of memory on a graphics card will not help any video game on the market (even the 512 cards are over kill). Like the FireGL cards, they are not all that great for gaming, but are extremely impressive for rendering. Great for 3-d content creation, scientific modeling and other rendering intencive activities.

-Rick

This is anthrocentric

[Anonymous Coward]

I'm tired of hearing this anthrocentric nonsense about chips.

GPUs are not faster than CPUs because the engineers can "concentrate on one area" instead of "spreading their work around". It's not that the floating point performance of the x86 would be faster if only Intel had the time to pay attention to it. That's ridiculous.

GPU tasks are highly parallel. CPU tasks are not. nVidia can toss 24 pipelines onto a chip and realize a huge performance gain. Intel can't, because much of the time those pipelines will be empty waiting for the results of the other lines.

This fundamental difference is what separates the two domains, not it being "easier to build something that performs well in one area, than to build something that does everything amazingly well (without costing the earth to buy it)."

You need to keep your science and your homey folk wisdom separate.

Speaking from a purely

[KingBahamut]

Support standpoint, we at the ubuntuforums find the support of ATI cards to be very frustrating. Their drivers dont work , and when they do, its spotchy at best.

http://ubuntuforums.org/showthread.php?t=148531
http://ubuntuforums.org/showthread.php?t=122094
http://ubuntuforums.org/showthread.php?t=148415
http://ubuntuforums.org/showthread.php?t=141090
http://ubuntuforums.org/showthread.php?t=137343
http://ubuntuforums.org/showthread.php?t=76147
http://ubuntuforums.org/showthread.php?t=75001

This is probably the largest complaint we get on the Ubuntu Forums and the UDSF(http://doc.gwos.org/ in the way of graphics cards. I think I even remember being told at one point that ATI is so driven on DirectX development that they likely dont care much about developing Open Source Drivers, or even a decent working Proprietary driver.

There have been a few Petitions to do so
http://www.petitiononline.com/atipet/petition.html
http://www.petitiononline.com/ati3/petition.html

And countless others. The community asks, almost begs, and all ATI does it laugh.

Its sad, really sad.

http://wiki.cchtml.com/index.php/Main_Page

Please don't joke about this.

[mosel-saar-ruwer]

Yes, they are 128-bit floats. They're needed for doing HDR.

PLEASE don't joke about this.

Do you have any idea how many math/physics/chem/engineering geeks would just kill for 128-bits in hardware?

It would be very, very cruel to get their hopes up like that, only to find out that you were being sarcastic...

Re:Please don't joke about this.

[Rufus211]

Sorry, but it's almost certinaly talking about 4x32bit SP floats. It's simply marketing multiplying out numbers for a bigger one.

Re:Good thing this is a workstation card

[argent]

why the hell would anyone want 256mb of textures on an already stuttering GPU.

Because 90% of programming is an excersize in caching, and if you can just cache the textures you can let your GPU just get 'em instead of waiting for it to finish saying "g-g-g-give me a t-t-t-tex... t-t-text-t-t-... gimme a damn bitmap!"