Monday, July 12, 2010

nVidia GTX460, The New Mainstream GPU King

source: hardwarezone.com
A True Mainstream GPU?

It's been almost a good four months since the launch of Fermi with the GeForce GTX 480, and we have yet to see a true mainstream Fermi GPU. The recently launched GeForce GTX 465, as we have noted in our review, is really a crippled GeForce GTX 470, so that doesn't really count. And with a launch price of US$279, it was also far too costly to be considered a viable mainstream option for casual gamers and regular computer users alike.Fortunately, it seems that NVIDIA has heard our cries and have given us the new GeForce GTX 460.

Positioned by NVIDIA as the card for the gaming masses, the new GeForce GTX 460 is not only more affordable, but perhaps more importantly, it is powered by a brand new chip, codenamed GF104. Notably, the GF104 chip contains 1.95 billion transistors and this is good news because lesser transistors usually mean a smaller die size, which in turn often translates to a cooler and less power hungry chip.

The GeForce GTX 460 is based on the GF104 chip, a modified version of the GF100 chip, that NVIDIA says will better cater to the mainstream market.


But more importantly, NVIDIA has restructured the GF104 somewhat to make the GeForce GTX 460 a competitive card. The GPC (graphics processing cluster) is the chip's dominant hardware level block, and the GeForce GTX 460 has two GPCs. For comparisons, the top of the line GTX 480 model has four GPCs. So you can tell that the GF104 is a massively reconfigured chip from the GF 100. Delving down to the details, the GeForce GTX 460 has a total of seven SM (streaming multiprocessors) - the GeForce GTX 480 has 15. And if you recall from our review of the GeForce GTX 480, each SM has 32 CUDA cores, however, NVIDIA has reconfigured the GeForce GTX 460 to give it 48 CUDA cores per SM instead, giving it a grand total of 336 CUDA cores.

Additionally, while the GeForce GTX 480 has four dedicated texture units per SM, the GeForce GTX 460 doubles this by having eight texture units per SM. This gives the GeForce GTX 460 a total of 56 texture mapping, which is identical to the GTX 470.

That's not all, to keep the GeForce GTX 460 competitive, it also gets relatively high clock speeds of 675MHz at the core, 1350MHz at the shaders and 3600MHz DDR at the memory.

And there's more, the GeForce GTX 460 will also be launched in two different flavors. It will come with either a 1GB frame buffer or a smaller 768MB frame buffer size. Both variants will utilize ultra-fast GDDR5 memory, have the same hardware configuration, and also clock speeds, but crucially, the 768MB variants will be equipped with a narrower 192-bit wide memory bus width. The 1GB cards, on the other hand, will have another memory controller enabled to have a wider 256-bit memory bus width.

So far, the GeForce GTX 460 is looking to be a pretty interesting addition from NVIDIA, but before we continue with our review, here's a quick look at how it measures up against its closest rivals.

Screenshot from GPU-Z detailing the specifications of the GeForce GTX 460. Note that some of the info shown here is incorrect and will require a newer software edition to ID the card properly. The GTX 460 has 336 shaders not 224.

NVIDIA GeForce GTX 460 and competitive SKUs compared
Model

NVIDIA GeForce GTX 460 1GB / 768MB GDDR5

NVIDIA GeForce
GTX 465 1GB GDDR5

NVIDIA GeForce GTX 470 1280MB GDDR5

ATI Radeon HD 5850 1GB GDDR5ATI Radeon HD 5830 1GB GDDR5ATI Radeon HD 5770 1GB GDDR5
Core CodeGF104GF100 GF100Cypress ProCypress LEJuniper XT
Transistor Count1950 million3200 million 3200 million2150 million2150 million1040 million
Manufacturing Process 40nm40nm40nm40nm40nm40nm
Core Clock675MHz607MHz 607MHz725MHz800MHz800MHz
Stream Processors336 Stream Processors352 Stream Processors 448 Stream Processors1440 Stream processing units1120 Stream Processing Units800 Stream Processing Units
Stream Processor Clock1350MHz1215MHz 1215MHz725MHz800MHz800MHz
Texture Mapping Units (TMU) or Texture Filtering (TF) units5644 56725640
Raster Operator units (ROP)2432 40321616
Memory Clock3600MHz GDDR53206MHz GDDR5 3348MHz GDDR54000MHz GDDR54000MHz GDDR54800MHz GDDR5
DDR Memory Bus 192 / 256-bit256-bit 320-bit256-bit256-bit128-bit
Memory Bandwidth86.4 / 115.2GB/s102.6GB/s 133.9GB/s128GB/s128GB/s76.8GB/s
PCI Express InterfacePCIe ver 2.0 x16PCIe ver 2.0 x16 PCIe ver 2.0 x16PCIe ver 2.0 x16PCIe ver 2.0 x16PCIe ver 2.0 x16
Molex Power Connectors2 x 6-pin2 x 6-pin 2 x 6-pin2 x 6-pin2 x 6-pin1 x 6-pin
Multi GPU TechnologySLISLI SLICrossFireXCrossFireXCrossFireX
DVI Output Support2 x Dual-Link2 x Dual-Link 2 x Dual-Link2 x Dual-Link2 x Dual-Link2 x Dual-Link
HDCP Output SupportYesYes YesYesYesYes
Street PriceLaunch Price: US$229 (1GB) / US$199 (768MB)Launch Price: US$279~US$349~US$300~US$220~US$160



The NVIDIA GeForce GTX 460
The card has been billed as the first mainstream affordable Fermi and it utilizes a traditional block-like cooler unlike older NVIDIA cards, with its fan located dead centre. However, perhaps the most striking thing you'll notice about the GeForce GTX 460 is that it is significantly smaller than previous Fermi cards. The reference GeForce GTX 460 measures only 8.25 inches in length, which makes it easier for it to fit into compact mid-tower casings.

Also, the GeForce GTX 460 requires less power, with NVIDIA stating that its maximum rated TDP is only 160W. This is much lower than any other Fermi card, considering even the GeForce GTX 465's maximum rated TDP is a considerable 200W. The GeForce GTX 460 can therefore get away with a wider range of PSUs, which also means users need not purposely upgrade their PSUs to accommodate the card.

At 8.25 inches long, the GeForce GTX 460 is a compact card. As you can see, it's considerably shorter than a Radeon HD 5850.

The reference versions of the GeForce GTX 460 we got from NVIDIA, both 768MB and 1GB variants, are completely identical.

Even the back of the cards are similar. We were hoping to see extra memory chips on the rear of the 1GB card, but that's not the case. Instead, the extra memory chips were found on the front side of the card, under the cooler.

The GeForce GTX 460 retains the same twin DVI and single mini-HDMI port for video output.

Despite its more lenient power requirements, the GeForce GTX 460 (both 768MB and 1GB) still requires two 6-pin PCIe power connectors.

The reference cards only came with a single SLI connector, which means only 2-way SLI is possible.

The GeForce GTX 460 uses a radial heatsink design, and peering closely, we can see the use of copper heat pipes to quickly draw and dissipate heat away from the GPU core. On casual glances, the aluminum heatsink design resembles that of an old Intel stock cooler, but that's just a side note for techies.



Test Setup

The cards will be tested using our high-end X58 system with the following specifications:

  • Intel Core i7-975 (3.33GHz)
  • Gigabyte GA-EX58-UD4P motherboard
  • 3 x 1GB DDR3-1333 OCZ memory in triple channel mode
  • Seagate 7200.10 200GB SATA hard drive
  • Windows 7 Ultimate

In terms of technical specifications, the GeForce GTX 460 is highly comparable to the GeForce GTX 465, so it'll be interesting to see how the two SKUs will match up. It'll also be intriguing to see where it stands against the more powerful GeForce GTX 470 too.

Considering its launch price, the GeForce GTX 460 is in the same league as the Radeon HD 5830 and close to the Radeon HD 5770, so we'll be looking at their performance closely.

To round up our analysis, we have also added the Radeon HD 5850 and the GeForce GTX 285 into our mix of results.

The full list of tested cards and their driver versions:

  • NVIDIA GeForce GTX 460 1GB GDDR5 (ForceWare 258.80)
  • NVIDIA GeForce GTX 460 768MB GDDR5 (ForceWare 258.80)
  • NVIDIA GeForce GTX 465 1GB GDDR5 (ForceWare 257.21)
  • NVIDIA GeForce GTX 470 1280MB GDDR5 (ForceWare 257.21)
  • NVIDIA GeForce GTX 285 1GB GDDR3 (ForceWare 257.21)
  • ATI Radeon HD 5850 1GB GDDR5 (Catalyst 10.6)
  • ATI Radeon HD 5830 1GB GDDR5 (Catalyst 10.5)
  • ATI Radeon HD 5770 1GB GDDR5 (Catalyst 10.5)

The list of benchmark used are as follows:

  • Futuremark 3DMark Vantage
  • Crysis Warhead
  • Far Cry 2
  • Warhammer: Dawn of War 2
  • Battlefield Bad Company 2
  • “Heaven" from Unigine v1.0
  • S.T.A.L.K.E.R.: Call of Pripyat


3DMark Vantage Results
The GeForce GTX 460 got off to a promising start with the 1GB variant outpacing the GeForce GTX 465 and also NVIDIA's own previous flagship, the GeForce GTX 285. The 768MB variant, on the other hand, found itself matching the GeForce GTX 465 despite having less frame buffer and having a narrower 192-bit memory bus width.

Against the cards from ATI, the GeForce GTX 460 was beaten by the Radeon HD 5830. But as we know, performance on 3DMark Vantage is usually not a fair representation of real world performance.



Crysis Warhead & Far Cry 2 Results
The GeForce GTX 460 continued its promising start with the 1GB variant consistently outperforming the GeForce GTX 465. The 1GB GeForce GTX 460 was on average about 5% quicker than the GTX 465. But what's more impressive is that the 768MB variant of the GeForce GTX 460 was able to continue to match the GTX 465. In fact, the 768MB variant of the GeForce GTX 460 only lost out when 4x anti-aliasing was enabled.

Both GeForce GTX 460 cards were also competitive against ATI's offering, as both convincingly outperformed the Radeon HD 5830. The Radeon HD 5850, however, proved to be too much even for the 1GB GeForce GTX 460.

On Far Cry 2, the 1GB GeForce GTX 460 continued to outclass the pricier GeForce GTX 465 and the 768MB variant also continued to keep the GeForce GTX 465 close in check. NVIDIA cards have always performed well on Far Cry 2, so it wasn't a surprise to see both GeForce GTX 460 cards managing higher frame rates than even the Radeon HD 5850.



Dawn of War 2 & Battlefield Bad Company 2 Results
Dawn of War 2 has traditionally been tightly contested and the 1GB GeForce GTX 460 extended its winning run against the GeForce GTX 465. The 768MB variant was competitive on the first run, but as we upped the intensity by increasing the resolution, it began to lose ground. Clearly, frame buffer size is important for anti-aliasing performance.

Also, both GeForce GTX 460 cards were comparable to the Radeon HD 5830, but on the higher resolutions, we can see that the 1GB variant was gradually outpacing the Radeon HD 5830, whereas the 768MB variant began losing its ground because of its smaller frame buffer size.

On Bad Company 2, the 1GB GeForce GTX 460 continued to outshine the GeForce GTX 465, although there wasn't much difference between them. The 768MB variant, by virtue of its smaller video memory size, was slower than the GeForce GTX 465.

Comparing the GeForce GTX 460 against the Radeon HD 5830, we can appreciate that the Fermi architecture is geared towards handling tessellation-heavy applications as we can see both the 1GB and 768MB variants garnering a big lead over the Radeon HD 5830. Even the slower 768MB GeForce GTX 460 was about 15% quicker than the Radeon HD 5830.

If you didn't realize it yet, Battlefield Bad Company 2 is designed to take advantage of DirectX 11 features, out of which tessellation is the biggest immediate pull factor for better performance and improved image quality. Though you can't explicitly specify the DirectX mode to run the game, it does recognize the hardware used and will use the best path possible. As such, our results analysis favors NVIDIA's architectures where more emphasis has been placed on tessellation performance.



Unigine "Heaven" Results
The 1GB GeForce GTX 460 pressed on with its fine performance, outgunning the GeForce GTX 465 on Unigine, especially when we enabled 8x anti-aliasing. With anti-aliasing enabled, the 1GB GeForce GTX 460 was significantly quicker. The 768MB GeForce GTX 460 did well too, managing to keep pace with the GeForce GTX 465, although it was slightly slower overall.

Against the Radeon cards, the 1GB GeForce GTX 460 was very competitive and managed to outperform the Radeon HD 5850 when running using DirectX 11 and with tessellation enabled. However, when we reverted to DirectX 10, the Radeon HD 5850 had the upper hand. Also, the Radeon HD 5830 wasn't a match for the 1GB GeForce GTX 460, and could only claw back slightly even when running on DirectX 10.

DirectX 11 Results

DirectX 10 Results



S.T.A.L.K.E.R.: Call of Pripyat Results
On S.T.A.L.K.E.R.: Call of Pripyat, the GeForce GTX 465 continued to find itself on the losing end when pitted against the new 1GB GeForce GTX 460. The 768MB variant did well too, managing to hold its own against the GeForce GTX 465.

And against the Radeon cards, both the GeForce GTX 460 once again trumped the Radeon HD 5830. The only time the Radeon HD 5830 looked like it had a lead was on the SunShafts run with DirectX 10 API enabled. In fact, overall, the performance of the 1GB GeForce GTX 460 is somewhat comparable to that of the Radeon HD 5850.

DirectX 11 Results

DirectX 10 Results



Temperature
Since the new GF104 chip only has about two-thirds the number of transistors the GF100 has, we were hopeful for some encouraging results. In our testing, the 1GB variant of the GeForce GTX 460 managed a promising 67 degrees Celsius at the GPU core, which is markedly lower than any of its Fermi siblings. The 768MB variant, on the other hand, recorded an even more impressive 60 degrees Celsius. With custom coolers, we are confident these figures will head even further south.


Power Consumption

Power consumption has never been Fermi's strong suit but the GeForce GTX 460 does remarkably well in this aspect, recording the lowest figures we've ever seen for a Fermi-based card. While peak power draw readings are still slightly higher than the Radeon cards, idle power draw figures are very decent and are on a par with ATI's Evergreen offerings.

Overclocking

NVIDIA informed us that the GeForce GTX 460 was an eager overclocker, and they weren't kidding. Bearing in mind that default core clock speeds is 675MHz, we managed to get both cards up well above 800MHz - to be precise, the 1GB variant managed 860MHz, while the 768MB edition made it all the way to 880MHz. This is just amazing. With their clock speeds, the 1GB GeForce GTX 460 recorded a 23% gain in 3DMark Vantage scores, whereas the 768MB GeForce GTX 460 recorded an astonishing 25% gain in 3DMark Vantage scores.



Fermi Done Right, At Last
NVIDIA has been going on and on about how their new Fermi is “DX11 Done Right”, but we have had our reservations about that statement. It is true that the new Fermi cards are powerful and fast, and have better performance in tessellation-heavy applications, but it comes at the price of heavy power draw and high operating temperatures. With the new GeForce GTX 460, however, things have taken a turn for the better.

For one, we are very impressed with the performance of the GeForce GTX 460. In all cases, the 1GB variant of it has proved itself to be as quick as, if not faster, than the GeForce GTX 465. This is despite it packing less CUDA cores than the GeForce GTX 465. So it seems that the modifications NVIDIA has done to come up with the new GF104 chip works.

Against its closest competitor from ATI, the Radeon HD 5830, the 1GB GeForce GTX 460 completely thrashes it. It was faster than the Radeon HD 5830 in almost all instances and on the newer DirectX 11 games such as Bad Company 2, it could be as much as 10% to 25% quicker.

That said, the 768MB variant of the GeForce GTX 460 is no slouch either. Although it was slower than the 1GB version, the 768MB GeForce GTX 460 is able to go toe-to-toe with the GeForce GTX 465 and the Radeon HD 5830, both of which are significantly more costly. Of course, compared to the other Fermi cards, it loses its edge more significantly when we increased the resolution or enabled anti-aliasing, but that's understandable given its smaller frame buffer size. Furthermore, it also has the distinction of being the coolest Fermi card, recording a cool 60 degrees Celsius in our temperature test.

With its outstanding performance and price, the GeForce GTX 460 might just signal the turn of the tide. A value for money card from NVIDIA? When was the last time anyone said that?

Overall, we are impressed with the GeForce GTX 460. Not only does it offer compelling performance, it is also markedly cooler to run than any other Fermi and less power hungry.. But between the 1GB and 768MB variant, which is the better buy?

To answer that, it's important to consider their price and performance. Launch prices of the 1GB variant has been pegged by NVIDIA at US$229, whereas the 768MB cards will be cheaper at around US$199 - this means the 1GB cards are about 15% more expensive. In terms of performance, the two are fairly close, but the 1GB variants are faster. To be specific, the 1GB cards are about 5% to 6% faster when anti-aliasing is disabled; however, when anti-aliasing is turned on, the gulf between the two increases to about 10% to 15%.

Hence, between the two, the pricing can be said to be quite reasonable. To be honest, considering their performance against competitive SKUs, both cards are value for money and for those who find US$229 too much of a stretch shouldn't worry about going for the less expensive 768MB variant. Unless a high-degree of anti-aliasing comes into play, the 768MB variant is very competitive; and even with anti-aliasing, it holds its own against the more expensive Radeon HD 5830. In fact, its performance on DirectX 10 games is very similar to the old GeForce GTX 285, which, if you recall, was once the fastest single GPU card in the world.

But for those who want a little more performance and future-proofing, and can afford the premium, the 1GB GeForce GTX 460 is a good pick especially if you'll be gaming in full HD resolutions or higher. It is also perhaps one of the most value for money graphics card model NVIDIA has introduced in recent times. At US$229, it's considerably cheaper than the US$279 GeForce GTX 465 which it outperforms; and in some cases, it even challenges and outperforms the US$300 Radeon HD 5850. We've done some internal performance-per-dollar comparisons and the GTX 460 1GB edition comes out a little better than the GTX 460 768MB edition, so our personal choice is clear. In fact, the GTX 460 fares so well overall that we're really wondering why was there even a need to launch a GTX 465 SKU a month earlier.

In closing, the new chip GF104 is certainly very promising, providing class-leading performance with more acceptable levels of power draw and operating temperatures. And we're pleased that NVIDIA has finally got its act together and gave us a proper mainstream Fermi GPU.

Looking ahead, we've heard rumors of a possible GeForce GTX 485 sporting the full 512 CUDA cores that Fermi promises, but when quizzed about the likelihood of such a SKU, NVIDIA said they couldn't comment. There's also talk of an even smaller and less complex Fermi chip, possibly a GF105 or GF106 that would be used with even more affordable mainstream and budget Fermi cards, but once again, NVIDIA neither denied nor confirmed these reports. In any case, if the GeForce GTX 460 is any indication, the future looks encouraging for NVIDIA and we hope to see more of such ideal Fermi-based solutions.

No comments:

Post a Comment