NVIDIA Announces GTX 480M: Mobile Fermi Coming Soon
by Jarred Walton on May 25, 2010 5:37 PM ESTNVIDIA Announces GTX 480M: Mobile Fermi Coming Soon
NVIDIA has been trailing AMD in terms of graphics performance and features since the launch of the Radeon 5000 series late last year. Missing the holiday shopping frenzy undoubtedly hurt, and AMD has enjoyed being the sole purveyor of DX11 hardware for six months. In late March, NVIDIA finally answered AMD with their launch of the desktop GTX 480 and GTX 470. While they did surpass AMD in raw gaming performance, the victory was not without a cost—specifically, the GTX 400 cards use a ton of power and run quite hot. Imagine our surprise, then, when we heard that GF100 chips would be going into notebooks next month.
We're told that the total graphics power draw (GPU + RAM) for each 480M MXM is 100W, which is higher than the 75W draw of the GTX 285M. While the GTX 480M will in theory support Optimus Technology, it's unlikely any vendors will actually implement Optimus. We expect the majority (all of them?) will go the discrete-only route as quad-core mobile i7 processors don't have the requisite IGP, and pairing the top-end GTX 480M with a Core i3/i5 would potentially introduce CPU bottlenecks. We expect with the GPU running constantly battery life is likely to be in the ~1 hour range, but that's typical of the gaming notebook market. Far more important than battery life is the performance characteristics, and here's where NVIDIA should trump AMD's Radeon Mobility HD 5870 quite handily.
Like the desktop GTX 400, the GTX 480M puts a heavy focus on tessellation hardware and compute shaders. NVIDIA states that the 480M will deliver "up to five times" the tessellation performance of any other GPU, though how much that will truly matter remains to be seen. GTX 480M will also support the usual slew of NVIDIA features like CUDA, OpenCL, and PureVideo, and it will be part of the Verde Driver Program. On the desktop, the additional shader performance does appear to be the main reason NVIDIA is ahead of AMD, with an average lead of around 15% at high resolutions. In DX11 titles, the lead is often even greater, coming in at 20 to 25% faster. With that in mind, let's look at what the GTX 480M offers.
GeForce GTX 480M Specifications | |
CUDA Cores | 352 |
Graphics Clock (MHz) | 425 |
Processor Clock (MHz) | 850 |
Texture Fill Rate (billions/sec) | 18.7 |
Memory Clock/Effective (MHZ) | 600/2400 |
Standard Memory Configuration | GDDR5 |
Memory Interface Width | 256-bit |
Memory Bandwidth (GB/sec) | 76.8 |
CUDA gigaflops | 598 |
PhysX capable | Yes |
OpenCL support | Yes |
PureVideo HD 1080p Full HD | Yes |
H.264, VC1, MPEG2 1080p video decoder | Yes |
Full spec Blu-ray decode | Yes |
DirectX 11 support | Yes |
DirectCompute Support | Yes |
OpenGL 2.1 support | Yes |
OpenGL 3.2 support | Yes |
Windows Vista, XP and 7 | Yes |
GDDR5 support | Yes |
LCD – LVDS support | Up to 2048x1536 |
VGA analog display support | Up to 2048x1536 |
DisplayPort multimode support | Up to 2560x1600 |
HDMI 1.4 support | Yes |
HDCP content protection | Yes |
7.1 channel HD audio on HDMI | Yes |
PCI Express 2.0 support | Yes |
SLI Ready | Yes |
Package | MXM board |
(Note: Some of the above figures may be different from what you'll find on NVIDIA's current spec sheet. We spoke with NVIDIA to confirm the numbers, and what we have posted should be correct. It sounds like there was a math error/typo initially, so the GFLOPS data is correct and we have clarified the memory clock information.)
Beyond the obligatory DX11 support, the paper specs of the 480M look to be a good sized jump over the previous generation GTX 285M. CUDA Cores have increased from 128 to a whopping 352, which could mean a dramatic increase in performance. It's interesting to note that with 352 shader cores, that means NVIDIA is disabling five of the available 16 "Streaming Multiprocessors" we discussed in our GF100 Architectural Overview. 352 is still a big jump from 128, but we have to keep clock speeds in mind as well. The GTX 285M had a core clock of 576MHz with a shader clock of 1500MHz, and up to 1020MHz GDDR3 memory (2040MHz effective). The GTX 480M has a core clock of 425MHz, a shader clock of 850MHz, and 600MHz GDDR5 memory (2400MHz effective).
The old G90 architecture at the heart of the GTX 285M could do three shader operations per clock; the GF100 does two per clock, which means that the raw GFLOPS isn't particularly faster when you factor in clock speeds. Memory bandwidth on the other hand has increased by around 20%, despite the relatively tame 600MHz base clock of the GDDR5 memory. If we look at the desktop GTX 480, we can see that even with less texture fill rate and only slightly more memory bandwidth, the other architectural changes resulted in a performance increase ranging from 20% at the low end to as much as 75%, with the average being around 50% faster. We would expect a similar performance increase with the GTX 480M compared to the previous generation 285M. Given that the mobile HD 5870 is only slightly faster on average than the GTX 280M, it looks like NVIDIA is set to retake the mobile performance crown.
Beyond offering better performance, the new 480M also provides full support for 3D Vision displays. Notebooks with dual-link DVI output will be able to drive 120Hz panels without any problem. Likewise, the internal LCD can be a 120Hz panel, and we'll likely see some notebooks with 3D displays down the road. SLI is also possible, though it looks like notebooks with 480M SLI won't be out for another couple of months.
Availability of notebooks with the GTX 480M should start next month, with the first units coming from Clevo. What we don't know yet is pricing information, but if the past is any indication we expect the GTX 480M notebooks to command a pretty penny. Being the fastest notebook on the block is one thing, but what we're interested in seeing is a well-balanced answer to notebooks like the ASUS G73Jh. Also worth noting is that the mobile DX11 at present consists of a single top-end GPU, whereas AMD has top to bottom DX11 support.
49 Comments
View All Comments
Zink - Tuesday, May 25, 2010 - link
If this is only half a GTX 480performance wise, wouldn't a desktop HD 5770 use the same amount of power as this card with similar performance for only $150 dollars? I think that these high end mobile GPUs are relativity expensive so using a stock desktop GPU instead of a $300? GTX 480M would make a killer laptop for much cheaper. You could use the same types of chassis/cooling as the GTX 480M needs.On the other hand the Mobility HD 5870 is just an underclocked HD 5770 designed to use only 60W. Using the same binned RV840 chips used in the Mobility 5870 and going all the way to 100W, ATI could make a mobile GPU that is an overclocked desktop HD 5770 and easily meet nvidia's performance levels.
Nvidia can't really win in the mobile GPU space with these blazing hot chips.
dagamer34 - Tuesday, May 25, 2010 - link
I never really did understand the "super performance" GPUs in laptops when they are such a significantly small segment of the market that you can't make any reasonable amount of money off of them. Even in desktops, far more money is spent in GPUs that cost under $200 than those that are $400+.We need to ditch this pure performance metric and start working in two-dimensional statistics: performance per watt for laptops and performance per dollar for desktops.
GullLars - Wednesday, May 26, 2010 - link
+1, single dimensioned thinking is for small minded people ;)B3an - Wednesday, May 26, 2010 - link
The only small minded people here are the ones like you.You people dont seem to understand that not everyone has YOUR specific needs and wants.
This GPU might not be good for you or even most people, but it is for others.
André - Wednesday, May 26, 2010 - link
Well, this is where it gets funny.Just looking at the gigaflops provided shows that the Radeon Mobility 5870 is vastly superior. So is the Radeon Mobility 5850 and the Radeon Mobility 5830.
JarredWalton - Wednesday, May 26, 2010 - link
But GFLOPS is the single most useless spec around I think, particularly for GPUs. Pure, theoretical performance is one thing; what you can actually get in practice is another. Look at desktop GTX 480 vs. HD 5870 and you'll see the same GFLOPS "deficit", but NVIDIA is consistently faster. Now power on the other hand....sh_kamalh - Wednesday, May 26, 2010 - link
External GPU is the way to go.I wish for a 14'' laptop with Core i5 and an external GPU on the same level of desktop 57xx+
Griswold - Wednesday, May 26, 2010 - link
ATI tried to establish that a couple years ago. Nobody wanted it.ik9000 - Monday, June 7, 2010 - link
well, shuttle seem to think differently: http://tv.hexus.net/show/2010/06/Computex_2010_Sou... notebook boosted by external desktop graphics card being showed off at computex 2010.york50 - Wednesday, May 26, 2010 - link
Google:" DIY Vidock"bandwidth is tight but soon to improve with PCI Express 2.0
works like a charm,
never again to buy a heavy expansive notebook that its gpu turns out dead in less than two years (nvidia) and from a manufacturer (gateway) that does not have spare parts to support its own products.
BAH