Slashdot Mirror
AMD Releases 900+ Pages Of GPU Specs
An anonymous reader writes "Ending off the X Developer Summit this year, Matthew Tippett handed off ATI's GPU specifications to David Airlie on a CD. However, the specifications are also now available on the X.org site. Right now there is the RV630 Register Reference Guide and M56 Register Reference Guide. Expect more documentation (and 3D specifications) to arrive shortly. The new open-source R500/600 driver will be released early next week."
over NINE HUNDRED!!
The only way to get nVidia to release their specs is to show them that there is a real market.
I'll do my part and replace my AGP nVidia card with an ATI one as soon as there is a good review of an available card with this driver on Ubuntu.
no, wait, the other thing - tedious.
They've actually done it. It's time to buy an ATI card.
Ewige Blumenkraft.
Closed-source drivers can be OK, except they tend to discontinue support after a while. Eventually the binary driver won't load into a current kernel and you are high and dry. With open-source drivers, the prospects for long-term support are better.
AMD ie recently making more moves toward the open source community than either it or ATI did prior to the merger.It seems to me that AMD has realized that there is value in not only having the right products rolling off the lines, but also having a greater mindshare.
Google realized this early, and bought off a great amount of geek awe by using Linux as the basis for its computing grid. This popularity among geeks turned into word of mouth advertising which turned into huge market share (having a great product didn't hurt either). Google still tries to maintain the "we're just a benign bunch of geeks" image (an image which is eroding, as it becomes more apparent that they are more akin to a lovechild of M$ and the NSA than a giant sushi eating LAN party). This appeal to mindshare by making steps toward the community, genuine or not, may be part of what AMD is trying to do, at least to an extent.
There are other genuine benefits to being more open about its specs, most clearly highlighted by the use of ATI GPUs to process Folding@Home. Therefore it is conceivable that AMD GPUs and GPU/CPU combo chips in the future may, if more openly specced, be used in a wider variety of HPC applications.
Disclaimer: I am an AMD fanboi.
I hate printers.
...will that GPU run Linux?
Just imagine an SLI'd Beowulf cluster of these!
np: Masha Qrella - Insecure (Luck)
"I'm not anti-anything, I'm anti-everything, it fits better." - Sole
Come off it... that's not even enough for an Office document standard.
Worthless!
The specs are for the brand-spanking-new RV630 series, the mid to low range chip in the r600 line. And the specs are only for 2d modesetting at the moment, so they likely apply to the whole r600 series. When the 3d specs are released they'll likely be a separate spec sheet for each specific chip. So to answer your question, they are for the newest cards ATi currently makes not their old, outmoded ones.
thisnukes4u.net
The RV630 chip is anything but "old", and what runs the Radeon HD 2600 PRO and Radeon HD 2600 XT. These belong to the very latest Radeon R600 line -- AMD's Direct3D 10 / Shader Model 4.0 supporting GPU's. These are high end chips from 2007, currently only beat by the Radeon HD 2900 XT in performance, if only speaking of AMD/ATI. The M56 chip is the core of the ATI Radeon Mobility X1600 (released December 2005), which is still a very decent mobile chip, roughly corresponding to an NVIDIA Geforce Go 7600 in performance.
Hmm, I now also see the ATI FireGL V7600 runs the RV630 too. Maybe that could work out for something too...
Beware: In C++, your friends can see your privates!
They've released the specs, this doesn't mean anything yet. People forget just how complex graphics cards are. Writing a driver for something like a network card or SCSI controller is fairly easy, and that's also evident from how small the drivers are. There's just little to do. 3D cards are extremely complex, hence the massive amount of documentation. It isn't like there was just some magic number that needed releasing and the OSS drivers would be perfect with full support. There's now a ton of work to be done, since it sounds like it is just specs, not code, they are releasing.
So you'll probably want to wait and watch until the driver is ready to go and up to whatever performance and stability standards you need for your application. Switch now and you are likely to find yourself in essentially the same situation as before: ATi's binary driver, or an OSS driver that doesn't do what you want.
It'll be some time before this information can be transformed in to a fully functional, stable, fast driver. After all, if it were so easy, ATi and nVidia would have perfect drivers out on the launch of a new card and never need to do anything but minor updates.
Actually, (IIRC) the driver has been sitting in somebody's desk drawer for months, waiting for AMDTI to bless it. The developer got the specs under NDA or something quite a while ago.
Specifically, google "over nine thousand" since the OP was paraphrasing. On a related note, my karma is over NINE THOUSAND!
For the record, Nvidia says otherwise.
You all should be grateful instead of pissing in their Cheerios."Thank you, oh benevolent masters, for supplying the software required to use the hardware that you gave me in exchange for money." Was that suitably deferential, or should I bend my knee more?
Dewey, what part of this looks like authorities should be involved?
Heh - I think you're missing the implication of his statement.
Because Linux is less resource intensive, he's able to upgrade his distro several times on the same hardware, putting himself in the situation of having a new kernel with old hardware and old drivers that don't load in the new kernel.
If you want to upgrade Windows, you usually wind up needing a new machine, so: new machine, new video card, new drivers, new Windows -- not a problem. Well, at least not the same problem.
So it's not an issue of what's *wrong* with Linux, it's what's *right* with it. The problem is that this presents circumstances the hardware world isn't used to dealing with.
Where's the -1 delusional mod?
NVidia have been stalwart protectors of their hardware designs, mostly due to historical accident. A few of the principals at NVidia used to work at Sun, where they designed the GX graphics chip. As it turns out, a version of SunOS was released with a header file describing the chip's registers. Using that -- and a logic analyzer -- a company called Weitek successfully created a functional clone of the chip that was good enough such that Sun's own drivers worked on it. This stuck in the craw of the Sun guys, and evidently vowed no such thing would happen again.
Another historical accident was that NVidia did, in fact, have a few source code releases way back. And every time they did, so it seemed, they got hit a few weeks later with a patent infringement lawsuit, usually from SGI. NVidia solved this latter problem largely through the expedient of buying SGI.
So, no, I don't think they're going to do it, and certainly not within six months. And yes, I would be perfectly tickled to be wrong about that.
Schwab
Editor, A1-AAA AmeriCaptions
Specs for r500(the X1k series) are supposedly in the pipleline. For matching codenames->marketing names, I recommend http://en.wikipedia.org/wiki/Comparison_of_ATI_Graphics_Processing_Units And the reason why they are so "deficient" is they have a team of a few engineers helping the X.org folks write the drivers. They have also said that they will be providing code snippets in the future to help clear up unclear parts of the spec. This is just a teaser release, not all we're getting.
thisnukes4u.net
And the Windows Vista driver sucks, somewhat hilariously.
It wasn't too long ago that I was at Microsoft's Philadelphia offices for an Exchange 2007 presentation. The first thing that they wanted to show was a short video on a projection screen -- what they actually showed the audience was a Vista laptop with ATI graphics choking half way through a two-minute video and then puking an error message saying that the video driver crashed and was being restarted. And some guy behind me said "Twelve years later and they still can't get the presentation right."
In Soviet Russia, Jesus asks: "What Would You Do?"
Well, if I can't get specs, my next video card will be an nVidia. Why should I suffer because my HW vendor wants to hide something from me? Do they really believe that non-functional hardware gains them any marketshare?
Now how does this make a lick of sense? nVidia haven't released ANY specs.
Also, I'd imagine that 2d rendering is reasonably similar across chipsets, but I admit I'm just guessing there.
Bzzzz - wrong.
Read David's blog - http://airlied.livejournal.com/ - there are a whole pile of potential problems about that driver. David accepts that it was on questionable ground, and so it will probably never see the light of day.