NVIDIA GeForce 500M: Refreshing the 400M

When NVIDIA announced their complete 400M lineup last September, we were pleasantly pleased: top to bottom DX11 support, with reasonable performance and the option for laptop manufacturers to use Optimus Technology to provide a nice balance of battery life and performance. There was one problem with that last item, however: the high-end Clarksfield CPUs used in most gaming notebooks didn’t have integrated graphics, so you were left with a notebook continually powering a discrete GPU. The result was less than stellar battery life but good performance, illustrated by notebooks like the ASUS G73Jw.
At the lower end of the performance spectrum, Arrandale offerings provided plenty of performance for mainstream users, but no one wanted to take a chance and pair a fast dual-core Arrandale (i.e. i7-640M) with a high-end GPU (i.e. GTX 460M) and enable Optimus graphics switching technology. That’s a shame, as it would have been an interesting alternative, but with such a short shelf life it probably doesn’t matter. As we illustrated earlier this week, Sandy Bridge has completely altered the mobile landscape. Now you can get near-desktop performance, quad-core processing with Hyper-Threading, decent integrated graphics, and battery life that’s none too shabby—all at roughly the same price as the old Core i7-740QM! SNB’s integrated graphics allow for a combination “have your cake and eat it too” notebook that combines a fast CPU with a fast GPU and doesn’t have to throw battery life under the bus. That’s the theory at least; we’ll have to wait for actual hardware to see who manages to pull it off.
With the Intel HD Graphics 3000 shipping in all mobile Sandy Bridge processors, Intel also effectively killed off anything below the 420M level. Yes, you can still get DirectX 11 support and better drivers from NVIDIA and AMD GPUs, but for those who only need “good enough” graphics there’s no real need to pay more. GT 425M still ended up being 55% to 96% faster than HD 3000 on average (at low and medium details, respectively), but while twice as fast is good, that 55% number is a bit of a concern. Why not widen the gap a bit by increasing clock speeds and memory bandwidth? Enter the GeForce 500M line.

NVIDIA’s New 400M/500M Parts
	GeForce GTX 485M	GeForce GT 555M	GeForce GT 550M	GeForce GT 540M	GeForce GT 525M
Codename	GF104	GF106	GF108	GF108	GF108
CUDA Cores	384	144	96	96	96
Graphics Clock	575MHz	590MHz	740MHz	672MHz	600MHz
Processor Clock	1150MHz	1180MHz	1480MHz	1344MHz	1200MHz
RAM Clock	750MHz (3.0GHz)	900MHz (3.6GHz)	900MHz (3.6GHz)	900MHz (3.6GHz)	900MHz (3.6GHz)
RAM Type	GDDR5	GDDR5/DDR3	GDDR5/DDR3	GDDR5/DDR3	GDDR5/DDR3
Bus Width	256-bit	192-bit	128-bit	128-bit	128-bit
Bandwidth (GB/sec)	96.0	43.2	28.8	28.8	28.8
SLI Ready	Yes	No	No	No	No

Let's get this out of the way first: this is just rebranding the old 400M series and chaning the clocks. Everyone clear on that? AMD is launching their 6000M parts as well, except they failed to provide us with any information ahead of time. I guess NVIDIA wanted to do a number update to keep people excited, but anyway....
Starting at the high-end, we also get a new 400M part to go with the 470M and 460M. The GTX 485M now uses the full GF104 chip, so you get 384 CUDA cores, and the clock speed is 575/1150MHz for the core/shaders, with 750MHz GDDR5 (3.0GHz effective). That puts it ahead of the old 480M (352 cores at 425/850/600 Core/Shader/RAM clocks), and it should have better power characteristics as well. To be specific, the GTX 485M ends up with 48% more computational power than the 480M and 25% more memory bandwidth. The 470M and 460M will continue their roles as high-end parts.
The first of the new 500M chips is the GT 555M, based on the GF106. Unlike the old GT 445M, the GT 555M is always a 192-bit interface, although there’s still the potential to curtail memory bandwidth by using DDR3 instead of GDDR5. So, while there’s plenty of confusion with the overlapping 400M and 500M parts, at least in this one instance we have some clarification. Core clocks are the same as the old GT 445M, but the GDDR5 memory is now specced to run at 900MHz (3.6GHz effective) instead of 800MHz.
The next three cores are all just a speed-bumped version of existing 400M parts. 550M, 540M, and 525M replace the 435M, 425M, and 420M respectively. All have 96 CUDA cores, just like the old 400M chips they replace, but memory bandwidth is up from 800MHz (25.6GB/s effective) to 900MHz (28.8GB/s)—a 12.5% increase. Clock speeds are also up, in this case it’s a 14% increase for the 550M vs. 435M, 20% for 540M vs. 425M, and 20% for the 525M over the 420M—not too shabby. As before, everything that doesn’t start with “GTX” lacks SLI support, which is fine by us; we’d rather have a single fast GPU over two slower GPUs in SLI just for ease of configuration if nothing else.