ATI Patent: Selectible multi-performance configuration

rwolf

Rock Star
Regular
http://patft.uspto.gov/netacgi/nph-...7,075,542.PN.&OS=PN/7,075,542&RS=PN/7,075,542

The present invention relates to a selectable multi-performance configuration. Instead of the traditional methods of producing separate high-end and low-end graphics chips, the present invention produces processing systems in a single unit. The single unit is readily and functionally partitionable. Each partition is capable of independent operation. By using all of the partitions a high-end graphics processing system may be simulated and tested. By using a subset of the partitions, a low-end graphics processing system may be simulated on the same system without the added cost of re-design of either hardware or software.

This seems like a patent that makes sense. Hard to believe nobody has done this before.
 
  • Like
Reactions: Geo
Thanks. :) Always like a new patent. We've had discussions already about collecting these as part of our resource/reference role going forward as we expand the site. Edit: Mmmmm. . .maybe with a link, where applicable, to the thread discussing it. Yeah. Yeah, I like that. :)

It's. . . a quad tester? I'm wondering tho expressed more as a simulator it might be a yield thing? It seems to me I've heard stories in the past that ATI's redundancy testing was not as granular/robust as NV's, with the implication that it was happening later in the production process from fab to ready-to-go-to-retail.

It talks about simulating from "high-end" to "low-end", but historically ATI has taken out more than just quads in going from a high-end design to a true low-end design. This is what makes me suspicious that there might be other fish to fry here than just simulation.

Btw, what are we calling those things in the new design? An "array" rather than a "quad"?
 
Sounds to me like the low end chip is encapsulated in the high end chip and they can test the low end chip by partitioning the high end chip.
 
While I'm on my link crusade:

For patents the linking style is explained:

http://www.uspto.gov/patft/help/helpfaq.htm#single

e.g.

http://patft1.uspto.gov/netacgi/nph-Parser?patentnumber=7075542

For applications the linking style is rather more awkward, based upon:

http://appft1.uspto.gov/netahtml/PTO/help/helpfaq.html#patent_number

e.g.

http://appft1.uspto.gov/netacgi/nph-Parser?Sect1=PTO1&Sect2=HITOFF&d=PG01&p=1&u=%2Fnetahtml%2FPTO%2Fsrchnum.html&r=1&f=G&l=50&s1=%2220060103658%22.PGNR.&OS=DN/20060103658&RS=DN/20060103658

which I obtained by going to the "search for application number" page

http://appft1.uspto.gov/netahtml/PTO/srchnum.html

and putting in the number and using the URL that comes up in the search result. If I can find a way of refining it, I'll post.

Jawed
 
I've not read the full patent text (those things bore me) but two things come to mind:

1. Software testing. As a graphics programmer I'd love to buy an ATI X1900 and have some setting where I can simulate lower end 1800/1600/1300 parts properly in hardware. Although, I suppose this is probably possible via the drivers.

2. Power consumption. Maybe some sort of method where your laptop runs on more/less units according to what its doing (and what power source is available). Even more simple - Windows Vista's AERO Glass thingy requires full SM2 all the time, but probably not much horse-power. Could disable 75% of the units when doing "simple" desktop composition work so as to reduce power consumption and heat and then ramp up the power when a proper 3D intensive application is executed.

hth
Jack
 
nelg said:
Software controllable upgrades = money.
Not only money, but easy money!


Buy a x1300 now, and click upgrade, type in a credit card, and you've got a brand new x1800!

Instant gratification.


And I'd love to start hacking away at such a system to unlock the full potential, right away :D
 
Software controllable upgrades? Like selling X1900 hardware in an X1300 configuration? The silicon gods laugh :LOL:

I haven't read the patent text as yet - but is this related to design/testing or commercial implementation?
 
nelg said:
Software controllable upgrades = money.

I don't think ATI is planning that, look at what ATI's line will soon be :
RV505, RV516, RV535, RV560, RV570, R580+

six chips to cover the entire range, that's quite a lot!


that idea is great, though it sucks to see it patented (you can disable quads and vertex units on geforce 6 with rivatuner, no? and turn a X1800XL into a X1800GTO with a BIOS flash). I guess it would be accessible through more mundane software.
I like the idea of using one quad on the desktop (lower frequency as well), even on a desktop PC, much like you can disable cylinders on a V8 or V12 nowadays (but you have to be a fat rich bastard to buy a V8 or V12 :))
 
I think the point is that you come out and test the high end chip then partition the high end chip and you have the low end chips and then you can test them as well before you make them silicon. When the silicon is made you just make the die include the partitions that make that model of GPU.
 
Back
Top