That Darn Memory Bus

Among the entire GTX 600 family, the GTX 660 Ti’s one unique feature is its memory controller layout. NVIDIA built GK104 with 4 memory controllers, each 64 bits wide, giving the entire GPU a combined memory bus width of 256 bits. These memory controllers are tied into the ROPs and L2 cache, with each controller forming part of a ROP partition containing 8 ROPs (or rather 1 ROP unit capable of processing 8 operations), 128KB of L2 cache, and the memory controller. To disable any of those things means taking out a whole ROP partition, which is exactly what NVIDIA has done.

The impact on the ROPs and the L2 cache is rather straightforward – render operation throughput is reduced by 25% and there’s 25% less L2 cache to store data in – but the loss of the memory controller is a much tougher concept to deal with. This goes for both NVIDIA on the design end and for consumers on the usage end.

256 is a nice power-of-two number. For video cards with power-of-two memory bus widths, it’s very easy to equip them with a similarly power-of-two memory capacity such as 1GB, 2GB, or 4GB of memory. For various minor technical reasons (mostly the sanity of the engineers), GPU manufacturers like sticking to power-of-two memory busses. And while this is by no means a true design constraint in video card manufacturing, there are ramifications for skipping from it.

The biggest consequence of deviating from a power-of-two memory bus is that under normal circumstances this leads to a card’s memory capacity not lining up with the bulk of the cards on the market. To use the GTX 500 series as an example, NVIDIA had 1.5GB of memory on the GTX 580 at a time when the common Radeon HD 5870 had 1GB, giving NVIDIA a 512MB advantage. Later on however the common Radeon HD 6970 had 2GB of memory, leaving NVIDIA behind by 512MB. This also had one additional consequence for NVIDIA: they needed 12 memory chips where AMD needed 8, which generally inflates the bill of materials more than the price of higher speed memory in a narrower design does. This ended up not being a problem for the GTX 580 since 1.5GB was still plenty of memory for 2010/2011 and the high pricetag could easily absorb the BoM hit, but this is not always the case.

Because NVIDIA has disabled a ROP partition on GK104 in order to make the GTX 660 Ti, they’re dropping from a power-of-two 256bit bus to an off-size 192bit bus. Under normal circumstances this means that they’d need to either reduce the amount of memory on the card from 2GB to 1.5GB, or double it to 3GB. The former is undesirable for competitive reasons (AMD has 2GB cards below the 660 Ti and 3GB cards above) not to mention the fact that 1.5GB is too small for a $300 card in 2012. The latter on the other hand incurs the BoM hit as NVIDIA moves from 8 memory chips to 12 memory chips, a scenario that the lower margin GTX 660 Ti can’t as easily absorb, not to mention how silly it would be for a GTX 680 to have less memory than a GTX 660 Ti.

Rather than take the usual route NVIDIA is going to take their own 3rd route: put 2GB of memory on the GTX 660 Ti anyhow. By putting more memory on one controller than the other two – in effect breaking the symmetry of the memory banks – NVIDIA can have 2GB of memory attached to a 192bit memory bus. This is a technique that NVIDIA has had available to them for quite some time, but it’s also something they rarely pull out and only use it when necessary.

We were first introduced to this technique with the GTX 550 Ti in 2011, which had a similarly large 192bit memory bus. By using a mix of 2Gb and 1Gb modules, NVIDIA could outfit the card with 1GB of memory rather than the 1.5GB/768MB that a 192bit memory bus would typically dictate.

For the GTX 660 Ti in 2012 NVIDIA is once again going to use their asymmetrical memory technique in order to outfit the GTX 660 Ti with 2GB of memory on a 192bit bus, but they’re going to be implementing it slightly differently. Whereas the GTX 550 Ti mixed memory chip density in order to get 1GB out of 6 chips, the GTX 660 Ti will mix up the number of chips attached to each controller in order to get 2GB out of 8 chips. Specifically, there will be 4 chips instead of 2 attached to one of the memory controllers, while the other controllers will continue to have 2 chips. By doing it in this manner, this allows NVIDIA to use the same Hynix 2Gb chips they already use in the rest of the GTX 600 series, with the only high-level difference being the width of the bus connecting them.

Of course at a low-level it’s more complex than that. In a symmetrical design with an equal amount of RAM on each controller it’s rather easy to interleave memory operations across all of the controllers, which maximizes performance of the memory subsystem as a whole. However complete interleaving requires that kind of a symmetrical design, which means it’s not quite suitable for use on NVIDIA’s asymmetrical memory designs. Instead NVIDIA must start playing tricks. And when tricks are involved, there’s always a downside.

The best case scenario is always going to be that the entire 192bit bus is in use by interleaving a memory operation across all 3 controllers, giving the card 144GB/sec of memory bandwidth (192bit * 6GHz / 8). But that can only be done at up to 1.5GB of memory; the final 512MB of memory is attached to a single memory controller. This invokes the worst case scenario, where only 1 64-bit memory controller is in use and thereby reducing memory bandwidth to a much more modest 48GB/sec.

How NVIDIA spreads out memory accesses will have a great deal of impact on when we hit these scenarios. In the past we’ve tried to divine how NVIDIA is accomplishing this, but even with the compute capability of CUDA memory appears to be too far abstracted for us to test any specific theories. And because NVIDIA is continuing to label the internal details of their memory bus a competitive advantage, they’re unwilling to share the details of its operation with us. Thus we’re largely dealing with a black box here, one where poking and prodding doesn’t produce much in the way of meaningful results.

As with the GTX 550 Ti, all we can really say at this time is that the performance we get in our benchmarks is the performance we get. Our best guess remains that NVIDIA is interleaving the lower 1.5GB of address while pushing the last 512MB of address space into the larger memory bank, but we don’t have any hard data to back it up. For most users this shouldn’t be a problem (especially since GK104 is so wishy-washy at compute), but it remains that there’s always a downside to an asymmetrical memory design. With any luck one day we’ll find that downside and be able to better understand the GTX 660 Ti’s performance in the process.

The GeForce GTX 660 Ti Review Meet The EVGA GeForce GTX 660 Ti Superclocked
Comments Locked

313 Comments

View All Comments

  • CeriseCogburn - Sunday, August 19, 2012 - link

    There are 8 OC 660Ti's on newegg right now and only 3 released at stock.
    By chance alone the reviewers will be reviewing an OC'ed 660Ti, as was pointed out in the article you did not read, there is NO "standard design" pushed by nVidia so the partners have a free reign to come out of the gate with OVERCLOCKS a rocking.
    They have done so.
    So now, OC is the standard and overwhelming production with the 660Ti
    Get use to it.
    Unfortunately AMD has been a severely restrictive control freak nazi master dom smacking down and hurting their partners and has not allowed freedom. Then, in the usual control monster hold the gamers back fashion, they finally okayed their GE crap to their broken and hurting partners so they could charge a lot more.
    Evil, greedy, tyranny control, amd
  • Galidou - Sunday, August 19, 2012 - link

    4 out of 18 cards for AMD are reference coolers on newegg, nothing different on the other side... 11 out of 18 are overclocked and the other 3 non-reference coolers that are not overclocked are begging to be boosted.

    So if OC is the standard, why not try to push it farther, factory overclocking versus aftermarket overclocking isn't much different if the video cards take it so easily. :)

    ''severely restrictive control freak nazi master dom smacking down and hurting'' even had to use the word Nazi... comon, be less of a fanboy, it's just childish, that was ridiculous. I'm not diminishing any of both companies, if you really want AMD to die, we'll all cry for no more competition will be alive. We'll be back to the days of geforce 2 gts at 800$. Praise the war and be a little more respectful please.
  • TheJian - Monday, August 20, 2012 - link

    I don't want AMD to die, but I wouldn't mind seeing them bought as they're already well on their way to death without our help. I can't justify buying cpu's that completely suck now (granted our crappy court system took forever to pay AMD for Intel's crap), and won't do it just to help them out.

    If the courts had seen fit to pay them what they truly got screwed out of (I'm reminded of buying white box ASUS boards because Asus was afraid to even put their name on the box!) when they were on top for 3 years at least, we wouldn't be having this discussion. They should have been given 15-20 billion from the ill-gotten 60+ billion earned from that time forward (as I'm sure market share would have gone up with more money to produce more stuff, keeping the fabs etc). It's not my wallets job to help now. They need to claim bankruptcy and get bought. Management has blown their ability to compete due to the financial burdens now facing them.
    http://investing.money.msn.com/investments/financi...
    Take a look at the last 10 years. Overall a loss of 6 billion or so. The previous 10 looks no better. In fact I think if you go back to inception, they haven't made a buck overall. That's not good. Shares outstanding in those last 10 years...DOUBLED. DILUTED! 344mil to 698 mil. You can't keep selling shares to cover old debts forever. Eventually nobody will loan you money and you can't operate at that point.

    The stock has been cut in half in the last 5 months. Intel will continue to crush them as they can't invest 4Bil like Intel is doing now over the next few years to stay ahead. You just can't win without R&D.

    OC is the standard but you're talking about doing it on your own, vs sanctioned by the manufacturer and coming as default like that on almost every 660 TI out there (which is what he pointed out), then you go off about fans? He's talking about SPEEDS already OC'd on the cards by default regardless of the fan on it. The makers of the card (msi, xfx, gigabyte etc) are SELLING THEM OC'd. You don't have to do anything buy buy it and stick it in. It's already overclocked, and overclocking itself to the highest clock it can without damage (that's built into the 600 series, NV is a step ahead here). That changes on a per gpu basis too...Very nice.

    Attack the man's data (if you can) not the nazi crap. Comparing the actions of one company to the actions of a well known person or group's actions (while I'd have chosen something other than nazi's) when said company is acting like them is valid. It's not disrespectful. His point wasn't they are killing Jews by the millions (or anyone else). His point was devs of cards are a bit peeved. I.E. only 2 have announced 7950 BOOST editions that I'm aware of. First, because they are already selling 900mhz+ versions that AMD doesn't want to see in the market (which is why I said Ryan should have benched one of his cards at this speed, what fool would buy REF or boost @850 when you can get a free 50-100mhz overclocked by the vid card makers already?), and second they don't care about want AMD wants after being shackled and wanting free reign, like you see on 660 TI's...all kinds of speeds from the launch, with rarely a REF CLOCKed card in sight! Do you get it now? It's not about the fan, it's about the speed the maker is willing to BACK out of the box by default and still warranty it without complaining about what you did to void their warranty. Cerise isn't putting AMD out of business, AMD is.
  • Galidou - Wednesday, August 22, 2012 - link

    I was speaking to CeriseCogburn, he sees the company as the prime evil, just read ny of his post and you will see the hate, the knifes in his eyes. I'd be AMD and I'd have him arrested by the police, he's a madman and almsot goes up to throw menaces at the company, Using terms like greed, Nazis and so much more that I'll leave it up to you to read him from page 1 to 10 on this forum. I'd be an AMD employee reading this and I'd be like ''WTF, I'm just a human being working my best to feed my family, I'm not working for the devil...''

    ''severely restrictive control freak nazi master dom smacking ''
    ''Evil, greedy, tyranny control, amd''

    used just in the text above mine, how are those words any useful when defending an opinion, that's repression, lack of respect and total madness..... We're not speaking of an army that tries to take control of the world by domination coming up in your home and killing your children FFS.
  • CeriseCogburn - Thursday, August 23, 2012 - link

    Keep lying and crying crybaby.
  • Galidou - Thursday, August 23, 2012 - link

    I'd like to lie, but the posts have not been deleted and everyone can read them.
  • Galidou - Thursday, August 23, 2012 - link

    I don't know if you realised but they way you display your arguments, accuse so easily, attack and confront even the owners of the cards, that uses it right now and telling them everything crashes and is unplayable while they are using the cards with no problem, it discredit any or your credibility.

    You alone are making it worse, continue, I have no problem with that, the more you add things like your last post, the worse it is. I won't, I'm sorry for the way things are turning, I'm not going down to your level because I'm just a simple enthusiast.

    And yet posting only to say ''Keep lying and crying crybaby'' is just another proof that SOMETIMES, your really just answering so you can have the last words. That was the most useless post I've ever seen....
  • CeriseCogburn - Thursday, August 23, 2012 - link

    Oh you're such a fool - it's nine months AFTER amd released their crap clocked cards with the LOCK on their core speed.
    dude, get a freaking clue
    "oh dey released it juzz azz much wit oc,....i'm so stpooopid and such a liar.."
    How about manning up : " I was wrong in front of my daughter every time, and I'm mad about that. She's going to be just as foolish as me (I hope)".
    No nevermind, you KNOW IT.
  • ionis - Thursday, August 16, 2012 - link

    "NVIDIA did see a lot of success last year with the GTX 560 driving the retirement of the 8800GT/9800GT"

    Just curious, where do you get these stats? I still haven't found a reason to upgrade from the 8800GT. It plays everything I throw at it great, on high, at 1680x1050.
  • Sufo - Thursday, August 16, 2012 - link

    Interesting, as the 8800 GTX can barely crack 30fps in BF3 at that res on low. I guess they were talking about people who play mainly modern games.

Log in

Don't have an account? Sign up now