<%BANNER[top_768x90]%>
<%BANNER[banner_468x60_h]%>
<%BANNER[article]%>

News

ATI Technologies COO and President David Orton said last week internal Pixel Shader precision in ATI VPUs will be 32-bit extremely soon. Currently ATI Technologies DirectX 9.0-supporting graphics processors provide only 24-bit internal precision for Pixel Shaders – enough for current shaders and fully corresponding to Microsoft’s requirements, but still lower than generally-recognized 32-bit processing precision.

Answering questions of PC Watch web-site in Tokyo, Japan, Orton told a number of conceptions regarding future visual graphics processing units. Keeping in mind extreme pace of innovation in the graphics industry, no one expected the President of the world’s leading VPU company to explain anything concrete about the future-generations of VPUs, such as ATI R420, ATI R500 or even ATI R600, but stick to some general conceptions – this is what Mr. Orton managed to do quite well.

David Orton said that his company does not have plans to keep the internal precision of its Pixel Shader processing unit at 24-bit. He said that soon ATI’s hardware will get 32-bit internal precision, though, he denied to comment in which VPU this new feature will be implemented. Provided that Microsoft requires internal 32-bit precision for its Pixel Shaders 3.0 to be available next year, it is logical to anticipate 32-bit internal precision in ATI’s next-generation ATI R420 graphics processor.

Currently ATI’s R300, R350, R360, RV350 and RV360 graphics processors used in the RADEON 9500-, RADEON 9600-, RADEON 9700- and RADEON 9800-series of graphics cards support 24-bit internal precision for Pixel Shaders – the minimum precision required by Microsoft Corporation for DirectX 9.0. NVIDIA’s GPUs support 16-bit and 32-bit internal precision for Pixel Shaders.

ATI Technologies also does not say “no” to 64-bit internal precision, but notes some substantial difficulties with its implementation.

Talking about longer-term future, Orton indicated Pixel and Vertex Shaders 4.x – a part of DirectX 10 and Longhorn – to be extremely flexible and resemble each other pretty well in terms of formats. This means that graphics processing units will be yet another step closer to central processing units. Moreover, executing units or unit for Pixel Shaders and Vertex Shaders are anticipated to be the same. This may probably concern ATI’s future generations of VPUs – R500 or even R600.

Orton warned about too rapid process of innovation – according to the President of ATI, this may harm industry more than it may benefit. A very serious problem with technology improvement is ability of software makers to keep the same pace and not to be a couple of steps behind the hardware. Another issue is the necessity to shrink the size of graphics chips to keep their cost at an affordable level. One more concern is the cost of memory – 256-bit bus will stay here for long and graphics companies will have to develop and utilize high-speed, but cost effective types of DRAM, including GDDR-II and GDDR3.

Basically, all the issues mentioned above mean that graphics technology cycles are very likely to last about 18 to 24 months, not 6 to 12 month, as before, just as Mr. Orton indicated earlier this year. Furthermore, it means that revolutions in hardware are not very likely to happen frequently than once every 18 to 24 months.

<%BANNER[banner_468x60_f]%>

Discussion

Comments currently: 5
Discussion started: 12/03/03 03:06:20 AM
Latest comment: 12/03/03 05:36:59 PM
Expand all threads | Collapse all threads

[1-4]

1. 
24-bit internal precision for Pixel Shaders – the minimum precision required by Microsoft Corporation for DirectX 9.0?
This means that the Nvidia 16bit is not valid?
[Posted by: I | Date: 12/03/03 03:06:20 AM]
+ expand thread (1 answer)

2. 
#2 I agree.
Geforce4 uses 8 bit, why not use also 12,16,24,32, or variable bit. Unless it's easier to use only one mode. What will Ati have to say when they launch de 32 bit version? 24 wasn't enought?

Well at least my 9500@9700 will not be surpassed very quick if it can convert 32->24. Is there a performance hit while converting from 32 to 24? Any one know's?
[Posted by: I | Date: 12/03/03 09:23:32 AM]

3. 
I didnt mean #2 i agree but Matt I agree.
[Posted by: I | Date: 12/03/03 09:25:36 AM]

4. 

Whilst its a definite possibility, I imagine its about one of five ways to decide where your transistor budget gets allocated too, more pipelines, deeper pipelines, fatter pipe lines, number of stages and executions units in your pipelines are amongst the things you have to trade off.

If VS 3.0 / PS 3.0 require it - it might be given more priority from a marketing perspective - but remember no games using these features will come out until R420 is heading towards obsolence - as is usual for new generations of cards and teh features they introduce.

I'd say more and faster pipelines are for sure 12 * 1 maybe, maybe more execution units per pipelines - and am wondering in it will be R420 or R500 that delivers true fp32 in pixel shaders that can actually perform to decent levels.
[Posted by: g__day | Date: 12/03/03 05:36:59 PM]

[1-4]

You must log in to add comments.

Forgot password? Registration

remember me



Related news

Latest News

Wednesday, December 3, 2008

4:47 pm | IN BRIEF: Windows 7 Beta 1 Scheduled to Emerge in a Month. Microsoft Preps to Release Windows 7 Beta 1 in January, 2009

12:58 pm | Logitech Ships Billionth Mouse amid PC Mouse’s 40th Anniversary. Logitech Reiterates Leadership on PC Control Devices Market

12:40 pm | ATI Expects Graphics Chip Market to Be Impacted by Economy Crisis. Economic Slump May Affect Sales of Graphics Cards - Head of AMD's Graphics Product Group

11:56 am | Intel’s Anti-Theft Protection Tech for Notebooks Hits the Market. Intel Anti-Theft Protection Technology Available on Select Lenovo ThinkPad Laptops

8:46 am | Personal Computer Market Growth to Slowdown in 2009, Market Value to Decline. PC Prices Set to Fall Amid Shipments Growth in 2009 – IDC