bit-tech.net

AMD details Radeon HD 6000M series

AMD details Radeon HD 6000M series

AMD's latest laptop GPUs offer impressive performance on paper.

AMD has officially confirmed the specifications of the Radeon HD 6000M series of mobile GPUs, ahead of their unveiling at the Consumer Electronics Show (CES) later this week.

Designed to offer manufacturers the full gamut of performance options, the new Radeon series comprises the budget-level Radeon HD 6300M, 6400M, 6500M, the mid-range 6600M, 6700M, and the high-end 6800M and 6900M.

At the bottom end of the range, the Radeon HD 6300M will be clocked between 500MHz and 750MHz, while the DDR3 memory will be clocked between 800 and 900MHz. The chip has 80 stream processors, eight texture units, 16 Z/stencil ROPs and four colour ROPs. At its highest specs, AMD claims that the 6300M offers 120 gigaflops of processing power, throughput of 187.5 million polygons per second and 14.4GB/sec of memory bandwidth.

Stepping up a level, the Radeon HD 6400M clocks in between 480MHz and 800MHz, with DDR3 or GDDR5 memory running between 800 and 900MHz. The chip features 160 stream processors, eight texture units, 16 Z/stencil ROPs and four colour ROPs. According to AMD, performance on the higher-end implementations is rated at 256 gigaflops, 200 million polygons per second, and memory bandwidth of 25.6GB/sec when using GDDR5 memory.

Hitting the mid range, the Radeon HD 6500M will be clocked between 500MHz and 650MHz, with DDR3 or GDDR5 memory running between 800 and 900MHz. This chip has a respectable count of 400 stream processors, 20 texture units, 32 Z/stencil ROPs, and eight colour ROPS. Despite the comparatively low clock speed, the Radeon HD 6500M's extra hardware pushes it past its cheaper brethren. AMD claims that it's capable of pushing out 520 gigaflops and 650 million polygons per second, while the memory bandwidth is 57.6GB/sec when using GDDR5 memory.

The AMD Radeon HD 6600M and 6700M share similar specifications, both featuring clock speeds between 500MHz and 725MHz, and DDR3 or GDDR5 memory clocks speeds between 800MHz and 900MHz. These chips also feature 480 stream processors, 24 texture units, 32 Z/stencil ROPs, and eight colour ROPs. The increase in clock speed over the Radeon HD 6500M, coupled with the slight boost in the number of stream processors, increases the claimed performance to 696 gigaflops and a throughput of 725 million polygons per second, although the memory bandwidth remains unchanged at 57.6GB/sec.

At the higher end of the scale, the Radeon HD 6800M is clocked between 575MHz and 675MHz, while its GDDR5 memory can be clocked between 900MHz to 1000MHz. The chip boasts a massive count of 800 stream processors, 40 texture units, 64 Z/stencil ROPs, and 16 colour ROPs. At the highest specifications, the Radeon HD 6800M offers up to 1,080 gigaflops of processing power, 675 million polygons per second throughput, and up to 64GB/sec memory bandwidth. While the performance for GPGPU tasks is clearly nothing to be sneezed at, the drop in clock speed appears to hurt the raw polygon performance.

Finally, there's the chip that you're going to see in the money-no-object desktop replacement machines; the Radeon HD 6900M. Clocked between 580MHz and 680MHz, and featuring a 900MHz GDDR5 memory clock, the top-end GPU offers 960 stream processors, 48 texture units, 128 Z/stencil ROPs, and 32 colour ROPs. AMD claims that its processing performance can hit 1.3 teraflops in single precision mode, while throughput is rated at 680 million polygons per second, and the memory bandwidth hits a whopping 115.2GB/sec.

All models in the range support DirectX 11 and OpenGL 4.1, and will also include the full range of AMD graphics technologies, including Eye-Definition, Eyefinity, EyeSpeed and HD3D support, along with the company's UVD 2 video playback acceleration technology.

Sadly, while AMD has been happy to give away the technical specifications, there are no models currently on the market to benchmark. With various partner companies expected to launch laptops based on the new 6000M-series at the CES later this week, we hope we won't have to wait long before we can bring you performance figures, as well as the prices.

Do you think that AMD's latest laptop graphics chips sound like winners, or has the company missed a trick? Share your thoughts over in the forums.

13 Comments

Discuss in the forums Reply
yakyb 4th January 2011, 17:36 Quote
i would be happy with ~ 9800 performance with Low power draw but there is little comparison ever done between desktop and Laptop Components

I.e how does the MSI GX740 compare to my gaming machine
Hustler 4th January 2011, 17:47 Quote
So, anything below 6500m is pretty worthless as gaming chips....

From what i can see the 6500m is about equal to my 4850 (bandwidth terms), although it is DX11...
proxess 4th January 2011, 18:08 Quote
I think my Mobility 5650 is a nice GPU, curious to see how this cuts out against it.
yakyb 4th January 2011, 19:13 Quote
Quote:
Originally Posted by Hustler
So, anything below 6500m is pretty worthless as gaming chips....

From what i can see the 6500m is about equal to my 4850 (bandwidth terms), although it is DX11...

is this right? do you have any benches
Hustler 4th January 2011, 19:20 Quote
Quote:
Originally Posted by yakyb
Quote:
Originally Posted by Hustler
So, anything below 6500m is pretty worthless as gaming chips....

From what i can see the 6500m is about equal to my 4850 (bandwidth terms), although it is DX11...

is this right? do you have any benches

How could i have benchmarks on a product that isnt even on sale yet????......

When it comes to GPU performance, Bandwidth is everything..my desktop 4850 has 64GB of Memory Bandwidth, this new 6500m has 57GB...

You work it out......
Elton 4th January 2011, 22:15 Quote
Hmm, thing is, if they're using the new kind of shaders (clusters of 4) would it be more SPUs or less? Because a rough comparison to Nvidia's SPU count would be dramatically different if it was in 4s.
alwayssts 4th January 2011, 22:27 Quote
Quote:
Originally Posted by Hustler
How could i have benchmarks on a product that isnt even on sale yet????......

When it comes to GPU performance, Bandwidth is everything..my desktop 4850 has 64GB of Memory Bandwidth, this new 6500m has 57GB...

You work it out......

Performance gained from doubling bandwidth is ~16%. That's hardly spectacular compared to a reasonable bump in core architecture, granted matching capabilities of both is key.

Just out of curiosity, are we to assume 6400 is Caicos, 6600/6700 are (perhaps neutered) Turks? Sure looks like it, and they're both 5D (as expected.)

With Llano coming, half of this stack will be close-enough to redundant in AMD laptops (unless people want to crossfire with these parts, making a respectable gaming laptop) later in the year. I guess we know their true purpose though, eh?

6900 is Barts. Wow, that's a big ol' clock dump to make it (6850) mobile capable.

I'm personally gonna wait for the 192-bit half-of-whatever-Cayman's-successor-is before upgrading my laptop. Pair it with a Trinity and we could have some pretty bad-ass mobile setups......................................in 2012. :'(
l3v1ck 5th January 2011, 00:18 Quote
I'd be interested to see the difference in gaming power between the 6300M and the built in Sandybridge GPU's.
It may give us an idea about Fusion performance in the future.
Goty 5th January 2011, 03:06 Quote
Quote:
Originally Posted by Elton
Hmm, thing is, if they're using the new kind of shaders (clusters of 4) would it be more SPUs or less? Because a rough comparison to Nvidia's SPU count would be dramatically different if it was in 4s.

It would be fewer SPUs for a given performance target, but these are supposedly the older VLIW5 shaders (which is fine, assuming they're really 6000-series chips).
Elton 5th January 2011, 04:41 Quote
I see, I just wanted a simple metric.

Of course the performance can't really be gauged seeing as the Barts core was quite impressive in terms of performance anyhow.
CowBlazed 5th January 2011, 07:06 Quote
The 4850 has 800 stream processors versus the 400 much newer and improved stream processors on the 6500M, all things said and done I'd give the slight advantage to the 4850 performance wise. That's still great for a mid range mobile GPU which has DX11 support, morphological AA etc.
xaser04 5th January 2011, 11:45 Quote
Hmm improvements in the mid range (6500m looks good but only if it uses GDDR5) whilst the high end looks like a incremental jump rather than a revolution. The 256bit memory interface on the 69xx mobility looks to be the only real jump over the previous gen.

It is slightly disappointing when you consider that the HD58xx mobility series were barely faster than the HD48xx mobility series they replaced.

I was really hoping to see at least 1120 shaders (HD6870) in the top end mobility GPU as the mobility lineup is starting to get left behind by the desktop line-up.

What really annoys me though is that with the HD48xx mobility line-up ATI/AMD proved they were able to take the full GPU and shoehorn it into a mobility package. Since then though despite seeing desktop GPUS get more efficient (Cypress vs RV770) the mobility lineup got a mere incremental DX update rather than a true performance improvement (800SP Juniper Part was made into the HD5870m and HD5850m).
Elton 5th January 2011, 12:44 Quote
Well xaser the HD58xx needed to use the lower end chip so it wouldn't melt the casing..or so I think.

that or they couldn't be bothered to make a more powerful GPU. However with that said, the Cypress core had really diminishing returns compared to the Barts and Cayman cores anyways so the smaller SP count is irrelevant given the better arch.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.



Discuss in the forums