The RV770 Story: Documenting ATI's Road to Success
by Anand Lal Shimpi on December 2, 2008 12:00 AM EST- Posted in
- GPUs
Just One Small Problem: We Need a New Memory Technology
The R600 GPU had an incredibly wide 512-bit memory interface, the problem with such a large interface is that it artificially makes your die bigger as you’ve got to route those interface pads to the memory devices on the board. For RV770 to have the die size ATI wanted, it needed to have a 256-bit memory interface, but using (at the time) current memory technology that wouldn’t give the GPU enough memory bandwidth to hit the performance targets ATI wanted.
When the options were either make the chip too big or make the performance too low, ATI looked elsewhere: let’s use a new memory technology. Again, put yourself in ATI’s shoes, the time was 2005 and ATI had just decided to completely throw away the past few years of how-to-win-the-GPU-race and on top of that, even if the strategy were to succeed it would depend on a memory technology that hadn't even been prototyped yet.
The spec wasn’t finalized for GDDR5 at the time, there were no test devices, no interface design, nothing. Just an idea that at some point, there would be memory that could offer twice the bandwidth per pin of GDDR3, which would give ATI the bandwidth of a 512-bit bus, but with a physical 256-bit bus. It’s exactly what ATI needed, so it’s exactly what ATI decided to go with.
Unfortunately whether or not GDDR5 shipped by the summer of 2008 wasn’t all up to ATI, the memory manufacturers themselves had a lot of work to do. ATI committed a lot of resources both monetarily and engineering to working with its memory partners to make sure that not only was the spec ready, but that memory was ready, performing well and available by the summer of 2008. Note that the RV770 was going to be the only GPU that would use GDDR5, meaning that it was ATI and ATI alone driving the accelerated roadmap for this memory technology. It’s akin to you trying to single handedly bring 100Mbps internet to your city; it’ll happen eventually, but if you want it done on your time table you’re going to have to pickup a shovel and start burying a lot of your own cable.
ATI did much of the heavy lifting with the move to GDDR5, and it was risky because even if RV770 worked out perfectly but the memory wasn’t ready in time the GPU would get delayed. RV770 was married to GDDR5 memory, there was no other option, if in three years GDDR5 didn’t ship or had problems, then ATI would not only have no high end GPU, but it would have no performance GPU to sell into the market.
If GDDR5 did work out, then it meant that RV770 could succeed and that it would be another thing that NVIDIA didn’t have at launch. That is, of course, assuming that ATI’s smaller-die strategy would actually work...
Facebook
Twitter
RSS
116 Comments
View All Comments
MrSpadge - Saturday, December 6, 2008 - link
Exactly what I was thinking! That's why I got a 8500LE back then, when Geforce 4 was not in (public) sight yet.FireSnake - Wednesday, December 3, 2008 - link
... which one is Anand (on the picture at the beginning of the article)?I always wondered how he looks like ... I guess the one on the right.
3DoubleD - Wednesday, December 3, 2008 - link
I've had Anandtech as my home page for 5 years and I've read almost every article since (and even some of the older ones). This is by far one of your greatest works!Thanks
hellstrider - Wednesday, December 3, 2008 - link
Kudos to Anand for such a great article, extremely insightful. I may even go out and purchase AMD stock now :)I love AMD even when it’s on the bottom, I own 780G + X2 + hd4850, in hopes that Deneb (or AM3 processors for that matter) will come in time to repeat the success of rv770 launch, at which point I will upgrade my obsolete X2 and have a sweet midrange machine.
My only concern is that Nvidia is looking at all this smirking and planning an onslaught with the 55nm refresh. There is a very “disturbing” article at Xbitlabs that Nvidia is stock-piling the 55nm GT200 parts; seems like that’s something they would do – start selling those soon and undercut 4800 series badly.
I’m just a concerned hd4850 owner and I don’t want to see my card obsolete within couple of months. I don’t really see AMD’s answer to 55nm GT200 in such short period of time?!?!
Any thoughts?
Goty - Wednesday, December 3, 2008 - link
I don't think you'll have to worry too badly about the 55nm G200s. NVIDIA won't drop prices much, if at all; they're already smarting from the price drops enacted after the RV770 launch. There's also the fact that the 4850 isn't in the same market space as any of the G200 cards, so they're not really competitive anyhow.ltcommanderdata - Wednesday, December 3, 2008 - link
I always imagined designing GPUs would be very stressful given you're trying to guess things years in advance, but this inside look at how things are done was very informative.On GDDR5, it's interesting to read that ATI was pushing so hard for this technology and they felt it was their only hope for the RV770. What about GDDR4? I thought ATI was a big supporter of it too and was the first to implement it. I'm pretty sure Samsung announced GDDR4 that could run at 3.2GBit/s in 2006 which isn't far from the 3.6GBit/s GDDR5 used in the 4870, and 4GBit/s GDDR4 was available in 2007. I guess there are still power savings to be had from GDDR5, but performance-wise I don't think it would have been a huge loss if GDDR5 had been delayed and ATI had to stick with GDDR4.
And another interesting point in your article was definitely about the fate of the 4850. You report that ATI felt that the 4870 was perfectly specced and wasn't changed. I guess that meant they were always targeting the 750MHz core frequency that it launched with. Yet ATI was originally targeting the 4850 at 500MHz clock. With the 4870 being clocked 50% faster, I think it should be obvious to anyone just looking at the clock speed that there would be a huge performance gap between the 4850 and 4870. I believe the X1800XL and X1800XT had a similarly large performance gap. Thankfully Dave Baumann convinced them to clock the 4850 up to a more reasonable 625MHz core.
One thing that I feel was missing from the article was how the AMD acquisition effected the design of the RV770. Perhaps there wasn't much change or the design was already set so AMD couldn't have changed things even if they wanted to, but they must have had an opinion. AMD was probably nervous that they bought ATI at it's height when the R580 was out and top, but once acquired, the R600 came out and underperformed. Would be interesting to know what AMD's initial opinion of ATI's small die, non-top tier targetted strategy was although it now seems to be more consistent with AMD's CPU strategy since they aren't targeting the high-end there anymore either.
hooflung - Wednesday, December 3, 2008 - link
The final frontier market share wise is to steal a major vendor like eVGA. If they can get an eVGA, BFG or XFX to just sell boards with their warranties AMD would be really dominant.JonnyDough - Wednesday, December 3, 2008 - link
The best thing I've ever read on a tech site. This is why you're better than THG.Only one typo! It was a "to" when it should have been a "too."
Chalk one up for the red team. This makes my appreciation for AMD rise even more. Anyone willing to disclose internal perspectives about the market like this is a team with less secrecy that I will support with my hard earned cash. So many companies could stand up and take a lesson here from this (i.e. Apple, MS).
Keep articles like this coming, and I'll keep coming back for more.
Sincerely,
~Ryan
epyon96 - Wednesday, December 3, 2008 - link
I have been an avid reader of this site for close to 8 years. I used to read almost every CPU, GPU and novelty gadget articles page to page. But over the years, my patience is much lower and I realize I get just as much enjoyment and information from just reading the first page and last page and skimming a few benchmarks.However, this is the first article in a while that I spent reading all of it and I thoroughly enjoyed it. These little back stories with a human element in one of the most interesting recent launches provides a refreshing change from boring benchmark-oriented articles.
I hope to find an article based on Nehpalem of a similar nature and other Intel launches.
GFC - Wednesday, December 3, 2008 - link
Wow, all i can say is that i loved this review. It was realy enjoyable to read, and i must give my thanks to Anandtech and Carrell!