The RV770 Story: Documenting ATI's Road to Success
by Anand Lal Shimpi on December 2, 2008 12:00 AM EST- Posted in
- GPUs
Depression Sets in but the Team Goes On
The entire RV770 design took around three years, which means that while we were beating ATI up over the failure that was R600, those very engineers had to go into work and be positive about RV770. And it was tough to, after all ATI had just completely lost the crown with R600 and Carrell, Rick Bergman and others were asking the team to ignore what happened with R600, ignore the fact that they lost the halo, and try to build a GPU that aimed at a lower market segment.
Through all of my interviews, the one thing that kept coming up was how impressed ATI was with the 770 team - never once did the team fall apart, despite disagreements, despite a shaky direction, the team powered through.
The decision not to go for the king of the hill part was a decision that made a lot of sense with ATI, but there was so much history about what would happen if you didn’t get the halo part; it took a very strong discipline to cast history aside and do what the leads felt was right, but the team did it without question.
The discipline required wasn’t just to ignore history, but to also fight the natural tendency for chips to grow without limits during their design phase. What ATI achieved with RV770 reminded me a lot of Intel’s Atom design team, each member of that team had strict limits on how big their blocks could be and those limits didn’t waver.
Adversity tends to bring the best out of people. The best stories I’ve been told in this industry, the Intel folks who made Banias and the ATIers that were responsible for RV770 put their hearts and souls into their work, despite being beat down. Passion has a funny way of being a person’s strongest ally.
The Power Paradigm
We were all guilty for partaking in the free lunch. Intel designed nearly five years of processors without any concern for power consumption and the GPU guys were no different.
In the R300 and R420 days ATI was almost entirely ignoring power, since estimating how much power the parts would use was so off from the final product that they just didn’t care. It was such a non-issue in those days that ATI didn’t even have a good way to estimate power even if it wanted to, it was impossible to design for a specific TDP. Today ATI’s tools are a lot better, now targeting a specific TDP is no different than aiming for a specific clock speed or die size, it’s another variable that can now be controlled.
These days power doesn’t change much, the thermal envelopes that were carved out over the past couple of years are pretty much stationary (ever wonder why the high end CPUs always fall around 130W?). Everyone designs up to their power envelope and stays there. What matters now is every year or two increasing performance while staying within the same power budget. Our processors, both CPUs and GPUs, are getting more athletic, rather than just putting on pounds to be able to lift more weight.
One of the more interesting things about architecting for power is that simply moving data around these ~1 billion transistor chips takes up a lot of power. Carrell told me that by the time ATI is at 45nm and 32nm, it will take as much power to move the data to the FPU as it does to do the multiply.
Given that data movement is an increasingly power hungry task a big focus going forward is going to be keeping data local when possible, minimizing moving to registers and on-chip caches. We may see more local register files and more multi-tiered memory hierarchies. As chips get more complex, keeping the register file in one central location becomes a problem.
ATI admitted to making a key manufacturing mistake with R600. The transistor technology selected for R600 was performance focused, designed to reach high clock speeds and yielded a part that didn’t have good performance per watt - something we noticed in our review. ATI has since refocused somewhat away from the bleeding edge and now opts for more power efficiency within a given transistor node. With leakage a growing problem as you go to smaller transistors it’s not worth it to be super leaky to gain a few picoseconds. If you’ve got a 100W GPU, do you want to waste 40W of that budget on leakage? Or would you rather do 80W of real work and only waste 20W? It’s the same realization that Intel recognized during the Pentium 4’s term and it’s the mentality that gave us the Core microarchitecture. It’s an approach that just makes sense.
Facebook
Twitter
RSS
116 Comments
View All Comments
nezuko - Thursday, December 4, 2008 - link
I think those phrase is describe what Graphic-field is. Another year win, and another year lose. But from those situations, only hardworking and tough guy would be able to turn all upside down. And ATi team do make it. Now I relieved I make a decision to buy 4670, though not performance, it still does big bang for the buck. And with those Catalyst 8.12, I would be more grateful that I bought this video card. Has been downloaded it and now testing it.Would Anand make another article about those GP-GPU programming language to make a data parralel computing possible.
Well, I considering to build my Leo Platform in the H2 of 2009 when the AM3 Deneb is out, Sata 3, and RD890.
JimiP - Thursday, December 4, 2008 - link
Like many before me have said, this has to be one of the best articles I've ever read here at AT. It really puts things into perspective. We (the consumer) are always criticizing or praising everything that comes out and don't take into account the amount of hard work and time put into the release. I'm 4850 owner, and I couldn't be happier with the performance I've received. I would like to personally thank ATI/AMD and the entire team that put RV770 into play. Absolutely brilliant.I would also like to thank Anand for sharing this awesome experience with us.
zshift - Thursday, December 4, 2008 - link
I have to say this was a great article. Great idea to write about the story behind these guys and the rv770. musta been a helluva relief when they realized how great the gpus were in the market, especially after taking such huge risks. For these guys to pull through the way they did, with the whole gddr5 issue and the die-shrink/physical limitations is amazing. I thought I was stressed in college. I can't imagine what its like to design something like this for 3 years not being even sure it'll work in the end. That's one hell of a resolve, makes me like ATI a bit more than I already do.Keep writing great articles here, this is my favorite site to read reviews on, and this is another reason why.
go anand! :p
strikeback03 - Thursday, December 4, 2008 - link
I agree with everyone else that the article is very well written. I am not sure if these would even be the right guys to ask, but did you bring up any of the driver issues your other recent articles have mentioned with them? As you have mentioned before, it is probably not the best business plan to assume nVidia will screw up again, and they should probably get their crossfire support in order for the good feelings about this strategy to continue.Dyno1979 - Thursday, December 4, 2008 - link
Definitely one of the best articles I've read lately. And I didn't even notice that typo, probably because I was reading the article instead of looking at it.5 stars
CarrellK - Thursday, December 4, 2008 - link
The "sweet spot" strategy would have amounted to *nothing* without the efforts of many very talented engineers (and a little luck as Anand has noted). They made the 770 happen and deserve the lion's share of the credit.I didn't think Anand would use this for anything other than background here-and-there in future articles. I fully expected him to politely cut me off at some point and say "about those future architectures..." which would have lead to Eric, Mike, and Mark telling a different interesting story. Thanks to Anand et al for telling this part of the 770 story. Responding to a comment or two in the posts:
* Sorry to quench the speculation - the AMD purchase had no effect on the 770's execution. Dirk Meyer and the other AMD executives supported Rick, a guy that they really didn't know, during some pretty tough times at AMD. They did their jobs so that we could do ours.
* The price range for 770-based cards was determined back in 2005 - it was an essential factor limiting the GPU cost, one of the big gambles. We had no clue what nV's 2008 pricing would be, but we did know what the gamers wanted. At launch we were tempted oh so briefly to launch at a higher price given the competitor's product offerings. It took some will-power for the starving man (us) to pass up a banquet (profits). We had a sneaking suspicion there was a lot of unhappiness about the direction prices had gone, and didn't want to be a party to that for the sake of a few weeks better revenue. Greed never pays. Remembering your customers does.
P.S. We don't keep any dart-board pictures of Anand around the office. However I *do* recall seeing his picture somewhere and thinking at the time that it *would* make a good dart target. Just a thought... :-)
lyeoh - Sunday, December 7, 2008 - link
You guys got the sweet spot right as far as I'm concerned (I'm not sure if it's true for others - does it show up in the units sold?)Before the ATI 3800 (RV670), and Nvidia 8800GT, it seemed like after shelling out a few hundred US dollars, you'd only get low/medium quality at current games. And cheaper cards were pathetic to unusable for new games.
So I stuck to playing old games with on my old video card (Ti4200) - which was decent in its time.
After the beginning of the new "sweet spot" era, this year I bought a 9800GT (and a new PC). While the 9800GT is not as good as AMD/ATI's offerings in hardware performance terms, I was concerned about ATI's drivers/software. A colleague tried an ATI card on his office PC, but in the end he had to switch to Nvidia to get his multiscreen set up on Linux working the way he wanted, and I had seen a fair number of complaints from others. So far Nvidia's drivers have been OK for me whether in Windows or Linux.
On the other hand I've seen too many Nvidia cards failing in hardware terms (bad caps, bad whatever). So pick your poison ;).
But if the cards aren't totally crap, it often takes less time to just replace a faulty card, than to keep tinkering with drivers and software configs (sometimes to no avail).
Anyway, many thanks for helping to make stuff affordable, even though I picked Nvidia again ;).
In the end I'm still back to mostly playing old games though...
MrSpadge - Saturday, December 6, 2008 - link
Thanks Andantech, ATI & AMD for this amazing article!And I'd like to add a point which has not been raised yet, at least in this discussion: the "small and fast enough" strategy only works because GPUs hit the realm where they're power limited!
The point is, whenever you go multi-GPU you loose performance due to inefficiencies and communication delays and there are also some transistors lost to redundant logic. If you had the choice between one 100 Mio transistor chip or 2 50 Mio ones, then the 100 Mio one would certainly be faster; assuming both could run at the same clock speed, which previously was determined by chip design (basically identical in the example) and process (identical).
But GT200 is too big, it can not fully fledge its clock speed wings because its power limited. Imagine GT200 at 1.5 - 1.8 GHz shader clock - it would be much more in line with performance expectations. RV770 on the other hand can be pushed quite a bit and on the 4870 it chews up lots of power for such a small chip - but that's OK because this power envelope has been accepted and the performance is there to justify it. And the 2 GPU versions are succesful because the power envelope on such "freak"-cards is larger.
And another frequently overlooked aspect: not all of GT200s transistors contribute to game performance. The 30 shaders which are 64 bit capable must be large and don't help games at all (and probably won't for quite some time). This is a very forward looking feature for games and a feature of immediate benefit for GP-GPU.
MrS
Frallan - Friday, December 5, 2008 - link
Thank you m8!Not only for delivering good products but also for delivering good information and entertainment.
Please convey to the other "Fellows" the heartfelt thanks of this community.
JimmiG - Thursday, December 4, 2008 - link
Congratulations to Anandtech for one of the most interesting articles this year. Congratulations to ATI/AMD for putting out their best and most exciting product since R300/9700 Pro.The industry really needed something like RV770. When the 9700 Pro came out in 2002, it was at the cutting edge of technology and performance, far ahead of the previous champion, the Ti4600, yet it launched at only $399. Nvidia launched the 8800 Ultra and GTX280 at $800 and $600 respectively, even though neither GPU introduced any significant new features, only moderately higher framerates.
I currently have a 4850 512MB which I bought in July and I love it... It runs all my favorite games at great framerates and with fantastic image quality at 1680x1050. Still, I wouldn't considering myself an "ATI fan". When it's time for me to upgrade again, I will buy the best card in the $200 range and won't care whether the sticker on the GPU fan is green or red.