We recently covered the mobile GPU landscape, with a discussion of the various performance levels and price segments. The price/performance ratio is actually pretty similar between AMD and NVIDIA mobile GPUs (at least until we hit the top-tier models), and both have a decent number of design wins with notebook ODMs. The current mobile performance crown goes to NVIDIA’s recently launched GTX 580M, but along with the performance crown comes a hefty price and performance bill that needs to be paid. AMD’s top mobile part prior to today’s announcement is the 6970M, which is basically a lower clocked version of the desktop Barts core with some of the Stream processors disabled (essentially a mobile HD 6850). Our testing has shown the 6970M to offer just slightly less performance on average compared to the GTX 485M, but interestingly enough NVIDIA managed to use less power in low/idle loads than AMD. Of course, even the HD 6970M is a trimmed Barts core, and there’s still the desktop 6950/6970 Cayman core that has yet to see a mobile variant, which brings us to today’s announcement.

If you were hoping to see a truly crazy mobile GPU running off the Cayman architecture, we’re unfortunately not getting that. Unlike the desktop 6990, we’re also not talking about a dual-GPU in a single card solution. Instead, the HD 6990M will be a full Barts core, with all 1120 shaders enabled. (The closest desktop equivalent is the HD 6870, which comes clocked at 900MHz, 25% higher than the 6990M.) Besides the now-standard DX11 support that AMD has been shipping since the first HD 5000 parts, 6990M also includes with HD3D (stereoscopic 3D) and OpenCL 4.1 support.

Looking at the mobile parts, the shader count gives the 6990M an immediate 17% boost in performance relative to the 6970M, and with a slightly higher cores clock as well (715MHz on the 6990M vs. 680MHz on the 6970M), we’re looking at up to 23% higher performance than the 6970M. Both the 6970M and 6990M continue to feature 3600MHz GDDR5 memory, although the 6990M comes with 2GB instead of 1GB. AMD also enabled OverDrive up to 740MHz for the 6990M if you want to try some quick overclocking. Here’s how performance between the AMD parts stacks up, according to AMD’s internal testing (using a desktop 3.4GHz Phenom II CPU):

On the other side of the fence, NVIDIA’s GTX 580M has the same number of CUDA cores as the GTX 485M (384 cores), but with an 8% increase in clock speed. (The closets desktop equivalent is the GTX 560 Ti, which comes clocked 37% higher than the 580M.) Our earlier testing of the 485M and 6970M resulted in nearly identical average gaming performance across eight tested games, with both sides winning a few titles. In theory, then, HD 6990M should retake the mobile performance crown given the greater increase in compute and clock speeds relative to the second-tier parts. The following slide uses simulated performance (e.g. a downclocked desktop GTX 560 Ti GPU running at mobile speeds and with only 1GB GDDR5, again with a 3.4GHz Phenom II CPU), so take these results with a grain of salt:

Both the AMD and NVIDIA parts should be plenty fast for 1080p mobile gaming, so the real question is more likely to be who offers the best overall value. Sure, value in a gaming notebook is something of an oxymoron, but unless you absolutely need CUDA/PhysX support on the NVIDIA side or are looking at Bitcoin mining on the AMD side, performance is going to be close enough that pricing will sway the vote. Availability of the 6990M starts today, with the Alienware M18x coming in both single and CrossFire configurations. Clevo will also support the HD 6990M in their P170HM, P150HM, and X7200 notebooks, which means we’ll see whitebooks from the usual suspects like Eurocom, AVADirect, and others. Here's AMD's complete high-end mobile GPU lineup:

As we've noted in the past, the 6000M consists of parts from both the Evergreen and Northern Islands series of graphics chips. For many users, the difference between the two isn't all that important, but Northern Islands does upgrade the video engine to UVD3 where Evergreen is UVD2.2. Also worth remembering is that the 6800M parts are really just renamed 5800M parts with slightly altered clocks in some cases, so they're not as attractive as the 6900M parts. Finally, the 6800M parts can come with either GDDR5 or DDR3, the latter being significantly slower and thus less desirable. Our table only uses the specs from GDDR5 variants, so if you're shopping for a 6800M make sure you get a GDDR5 model.

Outside of their newest mobile GPU, we also asked AMD about the current state of their switchable graphics on Intel platforms. AMD says they should have some partners releasing laptops with application based switching (e.g. similar to NVIDIA’s Optimus), but that will likely be with lower performance GPUs. In contrast, NVIDIA is touting Optimus support on certain GTX 580M configurations, though as always it’s up to the notebook vendors to utilize the feature. We haven’t had a chance to get hands on time with any form of AMD switchable graphics for some time, so the jury is still out. We hope to have an appropriate laptop for testing in the not-too-distant future, at which time we’ll be able to provide a better answer on which solution is the overall winner.

As for the question of who actually takes home the mobile gaming performance crown, we hope to have both GTX 580M and HD 6990M notebooks for testing in the coming weeks. On paper and using our previous 6970M and GTX 485M results, it looks like the 6990M should come out on top, but with various driver updates in the past several months we’re not ready to declare an official winner. If you’re looking for more than a few slides and potentially biased game selections, stay tuned: we’ll provide our usual in-depth look at real-world performance as soon as we can get hardware into our labs. Our money is still going to be on whoever can come in at a lower price point, and if recent history is any indication, that will likely be AMD with the 6990M.

The recent refresh of the GeForce GTX 460M into the 560M was a welcome one, bringing Optimus support and higher clocks to what's liable to be their crown prince of budget mobile gaming. The GTX 560M is beginning to materialize in the market, but we were still left with an odd hole in NVIDIA's lineup at the top. That hole has now been filled, and thankfully it's at least a little more than the usual speedbump that goes along with a new moniker these days. Now every part in the 500M series can switch over to Sandy Bridge's integrated graphics while running on the battery, offering the best of both worlds.

NVIDIA's new top dog is the GeForce GTX 580M. Every spec on the press sheets is listed as "up to," but that's really par for the course. While NVIDIA's reps declined to comment on what chips are being used for the GTX 580M, you'll find the specs are a dead giveaway. The 580M seems to be a mobile version of the GF104/114 (more likely GF114), with 384 CUDA cores, with a main clock of 620MHz (yielding 1240MHz on the shader cores). Riding shotgun is a 256-bit memory bus supporting up to 2GB of GDDR5 clocked at an effective 3GHz. Those of you keeping score at home will note that's an improvement of 45MHz on the core (and thus 90MHz on the shaders), with the memory speed remaining constant. All told I'd expect performance in the neighborhood of the desktop GeForce GTX 460 1GB version (or at least, the original NVIDIA spec): not too shabby, but still an incremental improvement on the shipping GTX 485M. The major selling point is, again, Optimus support in the 580M.

On the heels of the GTX 580M is the GeForce GTX 570M. The 470M was a bit of an oddity in that while the 480M and 485M were easy enough to find, the 470M was largely a rarity. While the 580M sees the minor speedbump we've come to expect, the 570M is a major improvement over its seldom-used predecessor. The 570M likely uses the same silicon as the 580M (which is, again, more than likely a GF114), but while the 470M only had 288 CUDA cores, the 570M gets a healthy upgrade to 336. The 470M's 535MHz core clock also sees a boost to 575MHz, with the shaders clocked at 1150MHz. The 570M still retains the 470M's 192-bit memory bus, and will be configured with either 1.5GB or 3GB of GDDR5 clocked at an effective 3GHz, a marked improvement on the 470M's 2.5GHz memory clocks. All told performance should be somewhere between the desktop GeForce GTX 460 768MB and the GTX 460 SE. For mobile gaming, that's still not bad at all, and again it benefits from Optimus support.

NVIDIA also was able to point out specific models of notebooks that will be shipping with these parts. Availability in 17" Clevo notebooks should surprise no one, with the P270WN in particular supporting both 3D Vision and SLI'd GeForce GTX 580M's. Alienware's new M18x (which we have en route for review) will also be supporting the GeForce GTX 580M in SLI. And finally, the MSI GT780R shown above will be shipping with the GeForce GTX 570M.

NVIDIA expects notebooks featuring the new GPUs to be available for order from OEMs today.

The new Vaio Z is significantly thinner and lighter than the old Vaio Z: It weighs in just 2.6lbs and has height of 0.66” compared to its 3.04lbs and 1.3” predecessor. It’s actually slightly thinner than the MacBook Air (0.66” vs 0.68”). MacBook Air’s design is slanted though so at its thinnest point, it’s much thinner than the new Vaio Z. Vaio Z is also quite noticeably lighter than 13” MacBook Air or Samsung 9 Series, which weigh 2.9lbs and 2.88lbs respectively. 

The main reason why the new Vaio Z is so much lighter and thinner than its predecessor is the lack of a discrete GPU. It comes with Intel HD 3000 which is integrated into the CPU die, whereas the old Vaio Z came with a discrete NVIDIA GT 330M. However, Sony has provided an interesting solution to this. They have announced a Power Media Dock (PMD) which is basically a souped up external optical drive. The external dock has an AMD Radeon HD 6650M with 1GB of DDR3 in it as well. The AMD 6650M allows you to connect up to two displays to the PMD so you can have a total four displays: two attached to the PMD, one attached to the laptop’s HDMI port and obviously the laptop’s integrated LCD. 

To make this worthwhile, Sony uses Intel’s Light Peak technology (yes, Sony calls it Light Peak, not Thunderbolt) which provides up to 10Gb/s of bandwidth in each direction. A GPU requires lots of bandwidth which is why USB, especially 2.0, is not suitable for powerful external GPUs. There are plenty of USB video adapters which are fundamentally external GPUs but they are not suitable for gaming or other GPU heavy tasks by any means. What makes Sony’s implementation different from Apple’s is the fact that Sony uses a combo-port that combines USB and Light Peak into one, instead of Mini DisplayPort connector like Apple. Sony actually calls the port a docking station/USB port. It can function as a regular USB 3.0 port as well. 

The Power Media Dock comes in three flavors. The cheapest one is a normal DVD drive. The second option is a Blu-Ray player. The most expensive option is a Blu-Ray writer as you might have guessed. PMDs with Blu-Ray are only available from Sony’s online store whereas the DVD PMD will be available through resellers as well. The PMD supports up to two external monitors: one via HDMI and one via VGA. There is also a Gigabit Ethernet port, two USB 2.0 ports and one USB 3.0 port. 

One issue with the PMD is its price. The cheapest one is £400 which translates to $640. Take away UK’s VAT which is 20% and we get $512, so the cheapest PMD will most likely retail for $500. That sounds pretty expensive, considering that AMD 6650M is an underclocked AMD 6570 (our review) which goes for around $70. Combine that with $20 DVD drive and you get what Sony calls Power Media Dock. 

Then there's the concern about performance. Sony must be running PCIe over Light Peak, but you only get a maximum of 1.25GB/s of bandwidth to/from the GPU - assuming no additional overhead. While the 6650M is likely much faster than Sandy Bridge's intergrated GPU, it's potentially slower than an on-board 6650M would have been because of the Light Peak bottleneck.

Conclusions

Vaio Z is definitely the most portable of the current 13" ultraportables and it packs in a nice amount of power. However, its pricing makes it too expensive compared to most of its competitiors. 13" MacBook Air starts at $1199 so the cheapest Vaio Z will be at least $700 more expensive if UK's pricing is comparable to US's. Even the Lenovo X1 starts at $1199 nowadays and it features exactly the same CPUs. The biggest market for Vaio Z seems to be in the high-end ultraportable market. No other brand offers 512GB SSD or 1080p screen at the moment for example. 

Power Media Dock sounds interesting but the price tag is everything else but attractive. For the same amount of money, you could grab for instance NVIDIA GTX 580, which is the fastest single chip GPU as of today. With PMD, you will be limited to AMD 6650M which isn't exactly a great GPU for gaming in the first place. At $500, you would hope for something much, much better than a DVD+GPU combo that struggles to play games at decent quality. We have seen external GPUs before but they have always failed miserably. Sony's attempt doesn't seem too promising either but at least we now have a port that is capable of external GPUs: Thunderbolt/Light Peak. There is hope that third parties will release cheaper and faster solutions, the most promising being Sonnet's Echo Express, which supports full size PCIe cards (though no word on GPU support). 

As the world becomes increasingly more mobile, external GPUs may be a solution to the problem of balancing portability with performance. Light Peak is a great way to get data out of a system however we may need to see a next-generation version of the interface, with higher bandwidth to really make sense for high performance external GPU solutions.