Introducing AMD’s Radeon 7000M and NVIDIA’s GeForce 600M Mobile GPUs
by Jarred Walton on December 7, 2011 2:45 AM ESTIntroducing AMD’s Radeon Mobility 7400M, 7500M, and 7600M
Traditionally AMD and NVIDIA have launched their new series of graphics products at the high-end and worked their way down. This means launching products like Cypress (Radeon HD 5800 series) and GF100 (GeForce 470/480) first, and following it up with smaller products like Redwood (Radeon HD 5600 series) and GF106 (GeForce 450) later. This owes to the fact that high-end GPUs are the flag bearers of a generation, with new architectures being built on these large chips first before lesser products are derived from them. As a consequence of building the biggest chips first, new architectures have always launched on the desktop first and have come to the mobile space later once the lesser derivatives were ready.
Today AMD will be launching their first Radeon HD 7000 series products, and in a significant deviation from normal they’re starting on the mobile side first. We’ve had some indication that this would happen—AMD chose to demo the mobile version of their 28nm GPU instead of the desktop version back in September—so this confirms AMDs intentions. However the 7000M series launching today is not quite what we had in mind.
We expected Southern Island products based on TSMC’s new HKMG 28nm process, but the fact of the matter is that TSMC’s HKMG 28nm process is running late—yields and production capacity just aren’t good enough for the production of high volume retail products for 2011. We may yet see some kind of 28nm product before the year is out so that AMD meets their stated commitment, but a complete 28nm launch in 2011 is off the table. However, AMD is concerned they need to launch new mobile products at the end of this year whether they have new GPUs or not, meaning they need to make do with what they already have.
As a consequence we’re facing another rebadging situation: the 7000M series launching today is based on AMD’s Turks and Caicos GPUs, the same GPUs that make up part of the 6000M series. Thus AMD may technically be launching the 7000 series today, but it’s the 7000 series in name only. The launch of the 28nm Southern Islands architecture will happen soon enough, but it won’t be happening today. Ignore the product number—if you wanted to see new GPUs from AMD [cue Obi-Wan], these aren’t the GPUs you’re looking for.
Naming shenanigans aside, the particularly frustrating part of all of this is that what was already a two architecture series just became a three architecture series. At the high-end we will of course see Graphics Core Next, AMD’s next-generation architecture intended to move the company away from VLIW. Meanwhile for integrated GPUs AMD’s Trinity will be using a VLIW4 design derived from AMD’s 6900 series Cayman GPU, and at the same time it stands to reason that at least some of AMD’s 7000 series will be VLIW4 in order to have something to CrossFire with Trinity. However, with the latest addition of Turks and Caicos on the 7000M, VLIW5 just got thrown into the mix and any kind of consistency just went out the window.
The one silver lining here is that even with the architecture differences, AMD’s VLIW5 architecture is still a modern architecture. Compute performance is lacking compared to the latest and greatest, but from a graphics perspective GCN, VLIW4, and VLIW5 are all Direct3D 11+ designs. The launch of Direct3D 11.1 will shake this up later next year—particularly if GCN is a D3D 11.1 design—but thankfully there won’t be a massive feature gap like we’ve seen in the past with other rebadging efforts. The graphics feature set will be mostly consistent, even if the underlying architectures are not.
With the above discussion out of the way, let’s hit the actual feature and spec sheets for the 7000M parts launching today. We’re including some information from the existing 6000M lineup as a reference point, and because we don’t have specifics on the actual models that are launching. If the past is anything to go by, we’d expect two or three models (maybe even four) in each series (e.g. the 6430M, 6450M, 6470M, and 6490M are all part of the 6400M lineup).
AMD Mobility Radeon 7400M, 7500M, and 7600M Lineup | ||||||
Radeon HD 7600M | Radeon HD 6750M | Radeon HD 7500M | Radeon HD 6630M | Radeon HD 7400M | Radeon HD 6470M | |
Core Name | Whistler Pro (?) | Whistler Pro | Whistler LT (?) | Whistler LT | Seymore XT (?) | Seymore XT |
Stream Processors | 480 | 480 | 480 | 480 | 160 | 160 |
Texture Units | 24 | 24 | 24 | 24 | 8 | 8 |
ROPs | 8 | 8 | 8 | 8 | 4 | 4 |
Core Clock | — | 600MHz | — | 485MHz | — | 700MHz |
Memory Clock | GDDR5/DDR3 | 900MHz (3.6GHz) GDDR5 | GDDR5/DDR3 | 800MHz (1.6GHz) DDR3 | GDDR5/DDR3 | 800MHz (1.6GHz) DDR3 |
Memory Bus Width | 128-bit | 128-bit | 64-bit | 128-bit | 64-bit | 64-bit |
Memory Bandwidth | — | 57.6GB/s | — | 25.6GB/s | — | 12.8GB/s |
Unfortunately, the details for the new 7000M parts are lacking right now—all we have to go on is the general configuration. We would assume the newer 7000M parts will have slightly higher clocks than the 6000M parts they’re replacing, but we really can’t say much more than that. All the current 7000M parts have the ability to support DDR3 or GDDR5 memory, and we expect to see higher-end models with GDDR5 and more budget friendly offerings with DDR3.
The most interesting (and not necessarily in a good way) series is the 7500M, which looks to straddle the ground between the entry-level 7400M and the more capable 7600M. It combines the 480 core GPU of the 7600M with the 64-bit memory interface of the 7400M. The goal is to bring prices down on mainstream hardware, but unless pricing is significantly lower the loss of memory bandwidth is going to hurt. Of course, the GDDR5 equipped models can provide the same bandwidth over a 64-bit bus as a DDR3 model with a 128-bit bus, but we’ll have to wait and see what laptops actually ship with the GPUs and how much they cost before we can come to any firm conclusions.
And that sums it up. AMD is launching 7000M GPUs today at the entry-level and midrange segments, but it’s only a rebadging of existing 6000M GPUs. We assume there will be some increased core clocks on the higher end SKUs, but overall there’s no significant change to the performance on tap.
64 Comments
View All Comments
Roland00Address - Wednesday, December 7, 2011 - link
The 6630m to 6770m has 480 shaders (vliw5) a die size of 118mm^2 (40nm) and it is effectively a 6670 that has been downclockedThe 6990m has 1120 shaders (vliw5) a die size of 255mm^2 (40nm) and it is effectively a 6870 that has been downclocked
You can easily find cheap sub 800 dollar notebooks with one of the 6630m to 6770m graphic card in there.
So assuming a die shrink of 40nm to 28nm (with perfect scaling, which I know is impossible) the
6990m replacement will have a die size 125mm^2 or very similar to the current 6630m to 6770m.
And while a perfect scaling is impossible, some of the 7000 series gpus will be vliw4 instead of vliw5. I am assuming the 6990m replacement will be vliw4, thus I believe it is likely to get 6990m performance in the 800 dollar range in the next 6 months or so.
SydneyBlue120d - Wednesday, December 7, 2011 - link
Is there any differences or is it always the same feature set? I red rumors of new UVD version with decrypt on board and maybe Nvidia may put the latest PV version on the old/new cards ? TnxRyan Smith - Wednesday, December 7, 2011 - link
Same GPUs. Same features.theangryintern - Wednesday, December 7, 2011 - link
So where are the 7000 and 600 series desktop cards?tzhu07 - Wednesday, December 7, 2011 - link
That's what I want to know too!I recently ordered a Sandy Bridge build from newegg sans video card. I plan on using the integrated HD3000 graphics in the meantime and later drop in a 7000/Kepler GPU. I mostly use my computer for productivity and light general use, so the lack of a real video card isn't too agonizing...Bleh.
Still, I'd like some official news of the HD7970 and GTX680 top end cards.
antef - Wednesday, December 7, 2011 - link
Hey I am right there with you. I don't really have time to build during the school semester so I am debating between either building over Christmas without a video card or delaying the whole build until after the spring semester. I can't decide! If the new cards are right around the corner I would build now, but people have been saying that since May! If they are delayed until March or later I might as well just wait.dj christian - Thursday, December 8, 2011 - link
Coming in Marssotoa - Wednesday, December 7, 2011 - link
Craptastic. Rebadge again.What gets me is... Where's the competition? How does one company know what the other is doing and they match each other STEP FOR FREAKIN STEP???
CarrellK - Wednesday, December 7, 2011 - link
Seeing as how I no longer represent AMD, I think I can comment on this.As the person responsible for AMD's graphics roadmap for many years, I know for a fact that neither company knows what the other does. You might read one of Anand's articles where I was interviewed. I believe I discussed this in passing in one of those articles. That isn't to say that I didn't make my best estimates for what I thought nV would do (and was right most of the time), I did. That rarely drove my decisions.
Doing what I thought we could do that would best address the market drove the decisions. We both use nearly identical technologies to build our products, and if we both do the best we can with those technologies, guess what the outcome is?
No collusion or evil intent.
Anand Lal Shimpi - Wednesday, December 7, 2011 - link
Carrell - thanks for dropping by :)FYI, CarrellK == http://www.anandtech.com/show/2937