AMD's Radeon HD 5870: Bringing About the Next Generation Of GPUs
by Ryan Smith on September 23, 2009 9:00 AM EST- Posted in
- GPUs
More GDDR5 Technologies: Memory Error Detection & Temperature Compensation
As we previously mentioned, for Cypress AMD’s memory controllers have implemented a greater part of the GDDR5 specification. Beyond gaining the ability to use GDDR5’s power saving abilities, AMD has also been working on implementing features to allow their cards to reach higher memory clock speeds. Chief among these is support for GDDR5’s error detection capabilities.
One of the biggest problems in using a high-speed memory device like GDDR5 is that it requires a bus that’s both fast and fairly wide - properties that generally run counter to each other in designing a device bus. A single GDDR5 memory chip on the 5870 needs to connect to a bus that’s 32 bits wide and runs at base speed of 1.2GHz, which requires a bus that can meeting exceedingly precise tolerances. Adding to the challenge is that for a card like the 5870 with a 256-bit total memory bus, eight of these buses will be required, leading to more noise from adjoining buses and less room to work in.
Because of the difficulty in building such a bus, the memory bus has become the weak point for video cards using GDDR5. The GPU’s memory controller can do more and the memory chips themselves can do more, but the bus can’t keep up.
To combat this, GDDR5 memory controllers can perform basic error detection on both reads and writes by implementing a CRC-8 hash function. With this feature enabled, for each 64-bit data burst an 8-bit cyclic redundancy check hash (CRC-8) is transmitted via a set of four dedicated EDC pins. This CRC is then used to check the contents of the data burst, to determine whether any errors were introduced into the data burst during transmission.
The specific CRC function used in GDDR5 can detect 1-bit and 2-bit errors with 100% accuracy, with that accuracy falling with additional erroneous bits. This is due to the fact that the CRC function used can generate collisions, which means that the CRC of an erroneous data burst could match the proper CRC in an unlikely situation. But as the odds decrease for additional errors, the vast majority of errors should be limited to 1-bit and 2-bit errors.
Should an error be found, the GDDR5 controller will request a retransmission of the faulty data burst, and it will keep doing this until the data burst finally goes through correctly. A retransmission request is also used to re-train the GDDR5 link (once again taking advantage of fast link re-training) to correct any potential link problems brought about by changing environmental conditions. Note that this does not involve changing the clock speed of the GDDR5 (i.e. it does not step down in speed); rather it’s merely reinitializing the link. If the errors are due the bus being outright unable to perfectly handle the requested clock speed, errors will continue to happen and be caught. Keep this in mind as it will be important when we get to overclocking.
Finally, we should also note that this error detection scheme is only for detecting bus errors. Errors in the GDDR5 memory modules or errors in the memory controller will not be detected, so it’s still possible to end up with bad data should either of those two devices malfunction. By the same token this is solely a detection scheme, so there are no error correction abilities. The only way to correct a transmission error is to keep trying until the bus gets it right.
Now in spite of the difficulties in building and operating such a high speed bus, error detection is not necessary for its operation. As AMD was quick to point out to us, cards still need to ship defect-free and not produce any errors. Or in other words, the error detection mechanism is a failsafe mechanism rather than a tool specifically to attain higher memory speeds. Memory supplier Qimonda’s own whitepaper on GDDR5 pitches error correction as a necessary precaution due to the increasing amount of code stored in graphics memory, where a failure can lead to a crash rather than just a bad pixel.
In any case, for normal use the ramifications of using GDDR5’s error detection capabilities should be non-existent. In practice, this is going to lead to more stable cards since memory bus errors have been eliminated, but we don’t know to what degree. The full use of the system to retransmit a data burst would itself be a catch-22 after all – it means an error has occurred when it shouldn’t have.
Like the changes to VRM monitoring, the significant ramifications of this will be felt with overclocking. Overclocking attempts that previously would push the bus too hard and lead to errors now will no longer do so, making higher overclocks possible. However this is a bit of an illusion as retransmissions reduce performance. The scenario laid out to us by AMD is that overclockers who have reached the limits of their card’s memory bus will now see the impact of this as a drop in performance due to retransmissions, rather than crashing or graphical corruption. This means assessing an overclock will require monitoring the performance of a card, along with continuing to look for traditional signs as those will still indicate problems in memory chips and the memory controller itself.
Ideally there would be a more absolute and expedient way to check for errors than looking at overall performance, but at this time AMD doesn’t have a way to deliver error notices. Maybe in the future they will?
Wrapping things up, we have previously discussed fast link re-training as a tool to allow AMD to clock down GDDR5 during idle periods, and as part of a failsafe method to be used with error detection. However it also serves as a tool to enable higher memory speeds through its use in temperature compensation.
Once again due to the high speeds of GDDR5, it’s more sensitive to memory chip temperatures than previous memory technologies were. Under normal circumstances this sensitivity would limit memory speeds, as temperature swings would change the performance of the memory chips enough to make it difficult to maintain a stable link with the memory controller. By monitoring the temperature of the chips and re-training the link when there are significant shifts in temperature, higher memory speeds are made possible by preventing link failures.
And while temperature compensation may not sound complex, that doesn’t mean it’s not important. As we have mentioned a few times now, the biggest bottleneck in memory performance is the bus. The memory chips can go faster; it’s the bus that can’t. So anything that can help maintain a link along these fragile buses becomes an important tool in achieving higher memory speeds.
Facebook
Twitter
RSS
327 Comments
View All Comments
SiliconDoc - Monday, September 28, 2009 - link
When the GTX295 still beats the latest ati card, your wish probably won't come true. Not only that, ati's own 4870x2 just recently here promoted as the best value, is a slap in it's face.It's rather difficult to believe all those crossfire promoting red ravers suddenly getting a different religion...
Then we have the no DX11 released yet, and the big, big problem...
NO 5870'S IN THE CHANNELS, reports are it's runs hot and the drivers are beta problematic.
---
So, celebrating a red revolution of market share - is only your smart aleck fantasy for now.
LOL - Awwww...
silverblue - Monday, September 28, 2009 - link
It's nearly as fast as a dual GPU solution. I'd say that was impressive.DirectX 11 comes out in less than a month... hardly a wait. It's not as if the card won't do DX9/10.
Hot card? Designed to be that way. If it was a real issue they'd have made the exhaust larger.
Beta problematic drivers? Most ATI launches seem to go that way. They'll be fixed soon enough.
SiliconDoc - Monday, September 28, 2009 - link
Gee, I thought the red rooster said nvidia sales will be low for a while, and I pointed out why they won't be, and you, well you just couldn'r handle that.I'd say a 60.96% increase in a nex gen gpu is "impressive", and that's what Nvidia did just this last time with GT200.
http://www.anandtech.com/video/showdoc.aspx?i=3334...">http://www.anandtech.com/video/showdoc.aspx?i=3334...
--
BTW - the 4870 to 4890 move had an additional 3M core transistors, and we were told by you and yours that was not a "rebrand".
BUT - the G80 move to G92 added 73M core transistors, and you couldn't stop shrieking "rebrand".
---
nearly as fast= second best
DX11 in a month = not now and too early
hot card -= IT'S OK JUST CLAIM ATI PLANNED ON IT BEING HOT !ROFL, IT'S OK TO LIE ABOUT IT IN REVIEWS, TOO ! COCKA DOODLE DOOO!
beta drivers = ALL ATI LAUNCHES GO THAT WAY, NOT "MOST"
----
Now, you can tell smart aleck this is a paper launch like the 4870, the 4770, and now this 5870 and no 5850, becuase....
"YOU'LL PUT YOUR HEAD IN THE SAND AND SCREAM IN CAPS BECAUSE THAT'S HOW YOU ROLL IN RED ROOSERVILLE ! "
(thanks for the composition Jared, it looks just as good here as when you add it to my posts, for "convenience" of course)
ClownPuncher - Monday, September 28, 2009 - link
It would be awesome if you were to stop posting altogether.SiliconDoc - Monday, September 28, 2009 - link
It would be awesome if this 5870 was 60.96% better than the last ati card, but it isn't.JarredWalton - Monday, September 28, 2009 - link
But the 5870 *is* up to 65% faster than the 4890 in the tested games. If you were to compare the GTX 280 to the 9800 GX2, it also wasn't 60% faster. In fact, 9800 GX2 beat the GTX 280 in four out of seven tested games, tied it in one, and only trailed in two games: Enemy Territory (by 13%) and Oblivion (by 3%), making ETQW the only substantial win for the GT200.So we're biased while you're the beacon of impartiality, I suppose, since you didn't intentionally make a comparison similar to comparing apples with cantaloupes. Comparing ATI's new card to their last dual-GPU solution is the way to go, but NVIDIA gets special treatment and we only compare it with their single GPU solution.
If you want the full numbers:
1) On average, the 5870 is 30% faster than the 4890 at 1680x1050, 35% faster at 1920x1200, and 45% faster at 2560x1600.
2) Note that the margin goes *up* as resolution increases, indicating the major bottleneck is not memory bandwidth at anything but 2560x1600 on the 5870.
3) Based on the old article you linked, GTX 280 was on average 5% slower than 9800X2 and 59% faster than the 9800 GTX - the 9800X2 was 6.4% faster than the GTX 280 in the tested titles.
4) Making the same comparisons, 5870 is only 3.4% faster than the 4870X2 in the tested games and 45% faster than the 4890HD.
Now, the games used for testing are completely different, so we have to throw that out. DoW2 is a huge bonus in favor of the 5870 and it scales relatively poorly with CF, hurting the X2. But you're still trying to paint a picture of the 5870 as a terrible disappointment when in fact we could say it essentially equals what NVIDIA did with the GTX 280.
On average, at 2560x1600, if NVIDIA's GT300 were to come out and be 60% faster than the GTX 285, it will beat ATI's 5870 by about 15%. If it's the same price, it's the clear choice... if you're willing to wait a month or two. That's several "ifs" for what amounts to splitting hairs. There is no current game that won't run well on the HD 5870 at 2560x1600, and I suspect that will hold true of the GT300 as well.
(FWIW, Crysis: Warhead is as bad as it gets, and dropping 4xAA will boost performance by at least 25%. It's an outlier, just like Crysis, since the higher settings are too much for anything but the fastest hardware. "High" settings are more than sufficient.)
SiliconDoc - Tuesday, September 29, 2009 - link
In other words, even with your best fudging and whining about games and all the rest, you can't even bring it with all the lies from the 15-30 percent people are claiming up to 60.96%--
Yes, as I thought.
zshift - Thursday, September 24, 2009 - link
My thoughts exactly ;)I knew the 5870 was gonna be great based on the design philosophy that AMD/ATi had with the 4870, but I never thought I'd see anything this impressive. LESS power, with MORE power! (pun intended), and DOUBLE the speed, at that!
Funny thing is, I was actually considering an Nvidia gpu when I saw how impressive PhysX was on Batman AA. But I think I would rather have near double the frame rates compared to seeing extra paper fluffing around here and there (though the scenes with the scarecrow are downright amazing). I'll just have to wait and see how the GT300 series does, seeing as I can't afford any of this right now (but boy, oh boy, is that upgrade bug itching like it never has before).
SiliconDoc - Thursday, September 24, 2009 - link
Fine, but performance per dollar is on the very low end, often the lowest of all the cards. That's why it was omitted here.http://www.techpowerup.com/reviews/ATI/Radeon_HD_5...">http://www.techpowerup.com/reviews/ATI/Radeon_HD_5...
THE LOWEST overall, or darn near it.
erple2 - Friday, September 25, 2009 - link
So what you're saying then is that everyone should buy the 9500 GT and ignore everything else? If that's the most important thing to you, then clearly, that's what you mean.I think that the performance per dollar metrics that are shown are misleading at best and terrible at worst. It does not take into account that any frame rates significantly above your monitor refresh are for all intents and purposes wasted, and any frame rates significantly below 30 should by heavily weighted negatively. I haven't seen how techpowerup does their "performance per dollar" or how (if at all) they weight the FPS numbers in the dollar category.
SLI/Crossfire has always been a lose-lose in the "performance per dollar" category. Curiously, I don't see any of the nvidia SLI cards listed (other than the 295).
That sounds like biased "reporting" on your part.