Monday, May 30, 2011

NVIDIA's GeForce GTX 560: The Top To Bottom Factory Overclock

NVIDIA’s GF104 and GF114 GPUs have been a solid success for the company so far. 10 months after GF104 launched the GTX 460 series, NVIDIA has slowly been supplementing and replacing their former $200 king. In January we saw the launch of the GF114 based GTX 560 Ti, which gave us our first look at what a fully enabled GF1x4 GPU could do. However the GTX 560 Ti was positioned above the GTX 460 series in both performance and price, so it was more an addition to their lineup than a replacement for GTX 460.

With each GF11x GPU effectively being a half-step above its GF10x predecessor, NVIDIA’s replacement strategy has been to split a 400 series card’s original market between two GF11x GPUs. For the GTX 460, on the low-end this was partially split off into the GTX 550 Ti, which came fairly close to the GTX 460 768MB’s performance. The GTX 460 1GB has remained in place however, and today NVIDIA is finally starting to change that with the GeForce GTX 560. Based upon the same GF114 GPU as the GTX 560 Ti, the GTX 560 will be the GTX 460 1GB’s eventual high-end successor and NVIDIA’s new $200 card.
GTX 570 GTX 560 Ti GTX 560 GTX 460 1GB
Stream Processors 480 384 336 336
Texture Address / Filtering 60/60 64/64 56/56 56/56
ROPs 40 32 32 32
Core Clock 732MHz 822MHz >=810MHz 675MHz
Shader Clock 1464MHz 1644MHz >=1620MHz 1350MHz
Memory Clock 950MHz (3800MHz data rate) GDDR5 1002Mhz (4008MHz data rate) GDDR5 >=1001Mhz (4004MHz data rate) GDDR5 900Mhz (3.6GHz data rate) GDDR5
Memory Bus Width 320-bit 256-bit 256-bit 256-bit
Frame Buffer 1.25GB 1GB 1GB 1GB
FP64 1/8 FP32 1/12 FP32 1/12 FP32 1/12 FP32
Transistor Count 3B 1.95B 1.95B 1.95B
Manufacturing Process TSMC 40nm TSMC 40nm TSMC 40nm TSMC 40nm
Price Point $329 ~$239 ~$199 ~$160

The GTX 560 is basically a higher clocked version of the GTX 460 1GB. The GTX 460 used a cut-down configuration of the GF104, and GTX 560 will be doing the same with GF114. As a result both cards have the same 336 SPs, 7 SMs, 32 ROPs, 512KB of L2 cache, and 1GB of GDDR5 on a 256-bit memory bus. In terms of performance the deciding factor between the two will be the clockspeed, and in terms of power consumption the main factors will be a combination of clockspeed, voltage, and GF114’s transistor leakage improvements over GF104. All told, NVIDIA’s base configuration for a GTX 560 puts the card at 810MHz for the core clock and 4004MHz (data rate) for the memory clock, which compared to the reference GTX 460 1GB is 135MHz (20%) faster for the core clock and 404MHz (11%) faster for the memory clock. NVIDIA puts the TDP at 150W, which is 10W under the GTX 460 1GB.

With that said, this launch is going to be more chaotic than usual for an NVIDIA mid-range product launch. While NVIDIA and AMD both encourage their partners to differentiate their mid-range cards based on a number of factors including factory overclocks and the cooler used, these products are always launched alongside a reference card. However for the GTX 560 this is going to be a reference-less launch: NVIDIA is not doing a retail reference design for the GTX 560. This is a fairly common situation for the low-end, where we’ll often test a reference design that never is used for retail cards, but it’s quite unusual to not have a reference design for a mid-range card.

As a result, in lieu of a reference card to refer to we have a bit of chaos in terms of the specs of the cards launching today. As long as you’re willing to spend a bit more in power, GF114 clocks really well, something that we’ve seen in the past on the GTX 560 Ti. This has lead to partners launching a number of factory overclocked GTX 560 Ti cards and few if any reference clocked cards, as the retail market does not have the stringent power requiements of the OEM market. So while OEMs have been using reference clocked cards for the lowest power consumption, most retail cards are overclocked. Here are the clocks we're seeing with the GTX 560 launch lineup.
GeForce GTX 560 Launch Card List
Card Core Clock Memory Clock
ASUS GeForce GTX 560 Top 925 MHz 4200 MHz
ASUS GeForce GTX 560 OC 850 MHz 4200 MHz
Palit GeForce GTX 560 SP 900 MHz 4080 MHz
MSI GeForce GTX 560 Twin FrozrII OC 870 MHz 4080 MHz
Zotac GeForce GTX 560 AMP! 950 MHz 4400 MHz
KFA2 GeForce GTX 560 EXOC 900 MHz 4080 MHz
Sparkle GeForce GTX 560 Calibre 900 MHz 4488 MHz
EVGA GeForce GTX 560 SC 850 MHz 4104 MHz
Galaxy GeForce GTX 560 GC 900 MHz 4004 MHz

This is why NVIDIA has decided to forgo a reference card altogether, and is leaving both card designs and clocks up to their partners. As a result, we expect every GTX 560 we’ll see on the retail market will have some kind of a factory overclock, and all of them will be using a custom design. Clocks will be all over the place, while designs are largely recycled GTX 460/GTX 560 Ti designs. This means we’ll see a variety of cards, but there’s a lack of anything we can point to as a baseline. Reference clocked cards may show up in the market, but even NVIDIA is unsure of it at this time. The list of retail cards that NVIDIA has given us has a range of core clocks between 850MHz and 950MHz, meaning the performance of some of these cards is going to be noticeably different from the others. Our testing methodology has changed some as a result, which we’ll get to in depth in our testing methodology section.

With a wide variety of GTX 560 card designs and clocks, there’s also going to be a variety of prices. The MSRP for the GTX 560 is $199, as NVIDIA’s primary target for this card is the lucrative $200 market. However with factory overclocks in excess of 125MHz, NVIDIA’s partners are also using these cards to fill in the gap between the GTX 560 and the GTX 560 Ti. So the slower 850MHz-900MHz cards will be around $199, while the fastest cards will be closer to $220-$230. Case in point, the card we’re testing today is the ASUS GTX 560 DirectCU II Top, ASUS’s highest clocked card. While their 850MHz OC card will be $199, the Top will be at $219.

For the time being NVIDIA won’t have a ton of competition from AMD right at $200. With the exception of an errant card now and then, Radeon HD 6950 prices are normally $220+; meanwhile Radeon HD 6870 prices are between $170 and $220, with the bulk of those cards being well under $200. So for the slower GTX 560s their closest competition will be factory overclocked 6870s and factory overclocked GTX 460s, the latter of which are expected to persist for at least a few more months. Meanwhile for the faster GTX 560s the competition will be cheap GTX 560 Tis and potentially the 1GB 6950. The mid-range market is still competitive, but for the moment NVIDIA is the only one with a card specifically aligned for $199.
May 2011 Video Card Prices
GeForce GTX 590
$700 Radeon HD 6990
GeForce GTX 580
GeForce GTX 570
$320 Radeon HD 6970
$260 Radeon HD 6950 2GB
GeForce GTX 560 Ti
$230 Radeon HD 6950 1GB
GeForce GTX 560
$180 Radeon HD 6870
GeForce GTX 460 1GB
$160 Radeon HD 6850
GeForce GTX 460 768MB
$150 Radeon HD 6790
GeForce GTX 550 Ti

Finally, I’d like to once again make note of the naming choice of a video card. I’m beginning to sound like a broken record here and I know it, but video card naming this last year has been frustrating. NVIDIA has a prefix (GTX), a model number (560), and a suffix (Ti), except when they don’t have a suffix. With the existence of a prefix and a model number, a suffix was already superfluous, but it’s particularly problematic when some cards have a suffix and some don’t. Remember the days of the GeForce 6800 series, and how whenever you wanted to talk about the vanilla 6800, no one could easily tell if you were talking about the series or the non-suffixed card? Well we’re back to those days; GTX 560 is both a series and a specific video card. Suffixes are fine as long as they’re always used, but when they’re not these situations can lead to confusion.

No comments:

Post a Comment