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?

[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?

[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).

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

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).

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)

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.

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 ; (

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.