• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD 7 series latest spec info from AMD round up and discussion

Farooq

Farooq

Farooq

Associate
Joined
4 Jan 2011
Posts
180
There have been a number of posts recently about the upcoming AMD 7xxx series cards on this forum and others so here is a round up of the details available from AMD and TSMC dated as of 16/10/2011. Im going to get into fine details on this thread for those interested but here is a roundup of what we know from those directly involved.

1)The cards will be made on a TSMC 28nm HP process.

2)There will be two Architecture types firstly VLIW4(very long instuction word) a die shrink and tesselation improvement over current cayman VLIW4 in the 6970 and secondly but more important the new GCN(Graphics core next) originally not planned to be introduced in this generation at all. It was roadmarked for 8xxx series and beyond. However the GCN is understood to be used in only the flagship models 7970 and 7950. This is a radical overhaul of the architecture and moves into massive gpu compute capabilities far beyond just being a gpu. Similar to what nvidia did with fermi and upcoming keplar. Ill link detailed slides from amd's june 16th talk on called upcoming architecture for a new era for those interested in a diagrams.

detailed anandtech roundup of GRAPICS CORE NEXT (GCN) link here

http://www.anandtech.com/show/4455/amds-graphics-core-next-preview-amd-architects-for-compute

fetch and simd details

http://www.nordichardware.com/image..._Islands/fullimages/AMD_Southern_Islands3.jpg

compute unit break down

http://www.nordichardware.com/image..._Islands/fullimages/AMD_Southern_Islands1.jpg

now the latest spec info for GCN cards

Graphics card Radeon HD 7970 Radeon HD 7950
Node TSMC 28nm HP TSMC 28nm HP
Architecture GCN GCN
GPU Tahiti XT Tahiti Pro
Radeon cluster 32 30
Radeon cores 2048 1920
GPU frequency 1000 MHz 900 MHz
Texture units 128 120
ROPs 64 64
Graphics memory 2GB XDR2 2GB XDR2
Memory freq 8000 MHz 7200 MHz
Memory inter 256 bit 256 bit
Memory bandwidth
256 GB/s 230 GB/s
Power consumption
190 watt 150 watt

and the die shrinked and tweaked compiler VLIW4

Graphics card Radeon HD 7870 Radeon HD 7850
Node TSMC 28nm HPL TSMC 28nm HPL
Architecture VLIW4 VLIW4
GPU Thames XT Thames Pro
Radeon cluster 24 22
Radeon cores 1536 1408
GPU frequency 950 MHz 850 MHz
Texture units 96 88
ROPs 32 32
Graphics memory 2GB GDDR5 2GB GDDR5
Memory frequency
5800 MHz 5200 MHz
Memory interface
256 bit 256 bit
Memory bandwidth
186 GB/s 166 GB/s
Power consumption
120 watt 90 watt

prices no public availabilty (suprise suprise)

now the noticable differences between the cards


1) The GCN has XDR2 memory at a staggering 8000MHz the VLIW4 rev b sothern islands uses gddr5 at 5800Mhz for reference current cayman 6970 is at gddr5 5500MHz.

2) According to the amd slides shown above at AMD june 16th talk the GCN(Graphics core next) will handle x86 code and virtual memory as well as full support for any c++ construct including dlls and virtual functions this opens up a whole new bag for developers to use with the bounus that they already code in this way.A lot of applications can be easily patched or coded in very simple ways to use the full power available of the gpu compute on the GCN. This is a radical break from the more specific gpu accelarated tasks currently available on VLIW4 and fermi/keplar platforms.

3) The power footprint is reduced considerably during the die shrink. The 7870 which is a revision of the caymen 6970 core uses less than half the power coming in at a lowly 120 watt compared to the 250 watt of the 6970.
The top end 7970 uses 190 watt still a lot lower than the current cards.

My thoughts

This is the currently available information from AMD and TSMC on the 7xxx series gpu's. I fully expect things to change from now until launch but not greatly from what we see here. That being said it looks like due to fast turnaround of nvidia from its 4 to 5 series and upcoming keplar Q2/Q3 2012 AMD has had to bring forward its GCN to this generation. Perfomance can only be estimated so I wont even go into that until we have benchmarks, but the sweet spot looks like the 7950 GCN for the all round performance plus gpu compute. or for the budget gamer even crossfire 2x 7870 with a power req of 10 watts less than a single 6970 similar at least and if not better performance than a 6990 at a fraction of the estimated cost and will run on a lowly powersupply.

I just wanted to share the current information and hear your thoughts on these cards and the direction of gpu's according to AMD plans.
look forward to reading your comments and thoughts.

ps no flaming I love the green team as well :D cant wait to see what they bring to counter this move by AMD interesting time for GPU's
 
Last edited:
Very interesting, lots more info than I've seen before. These cards look to be really good if they prove to be as good as those specs. The low power use is great too, my choice of a 750W psu seems to have been a good one :)
 
To be honest mate I have no idea and pricing is not the type of thing AMD will make public. That being said the 7870 will be a faster/cooler and lower power version of the 6970 with massivly improved complier so perfomance should be greater even with similar specs so I would expect AMD to stop producing the 6970/6950 in bulk and focus on the 28nm chips and fabrication to ramp up considerably to meet market demand.
 
Last edited:
7870/7850 would be a great choice for CrossfireX as they have 2GB vram per GPU. It's like the past brilliance of the 5770 1GB CrossfireX - GPU raw power not bottlenecked by vram.

However, 7970/7950 might eventually hit the vram wall in multiscreen/2560 res sooner or later, if they only have 2GB vram per GPU. To unleash the full GPU power in CrossfireX I would still prefer non-reference cards with 4GB vram per GPU.
 
We would all like 4gb of memory however its understandable why the reference cards use 2gb. xdr2 or rambus 2.0 at 4gb would be a severe cost increase. They chose to use this kind fo ram for a reason its a lot faster and it can handle small memory requests a lot better than gddr5 as well as point to pont requests which the gddr5 has to use a minimum of 128byte bursts infact this is the same on any non rambus ram ddr2 ddr3 etc. XDR2 can use from 4-8 byte bursts. This allows a lot of perfomance headroom for small read write cycles and helps in gpu compute tasks a lot.

I am going to lift a small section from rambus and copy n paste it in for you guys to help understand the changes from previous generation rambus.

(Changes from XDR DRAM
Signaling

In addition to a higher clock rate (up to 8000 MHz), the XDR2 differential data lines transfer data at 16 times the system clock rate, transferring 16 bits per pin per clock cycle. This "Hexadecimal Data Rate" is twice XDR's 8× multiplier. The basic burst size has also doubled.

Unlike XDR, memory commands are also transmitted over differential point-to-point links at this high data rate. The command bus varies between 1 and 4 bits wide. Even though each bit requires 2 wires, this is still less than the 12-wire XDR request bus, but it must grow with the number of chips addressed.

Micro-threading

There is a basic limit to how frequently data can be fetched from the currently open row. This is typically 200 MHz for standard SDRAM and 400–600 MHz for high-performance graphics memory. Increasing interface speeds require fetching larger blocks of data in order to keep the interface busy without violating the internal DRAM frequency limit. At 16×800 MHz, to stay within a 400 MHz column access rate would require a 32-bit burst transfer. Multiplied by a 32-bit wide chip, this is a minimum fetch of 128 bytes, inconveniently large for many applications.

Typical memory chips are internally divided into 4 quadrants, with left and right halves connected to different halves of the data bus, and top or bottom halves being selected by bank number. (Thus, in a typical 8-bank DRAM, there would be 4 half-banks per quadrant.) XDR2 permits independently addressing each quadrant, so the two halves of the data bus can fetch data from different banks. Additionally, the data fetched from each half-bank is only half of what is needed to keep the data bus full; accesses to an upper half-bank must be alternated with access to a lower half-bank.

This effectively doubles the number of banks and reduces the minimum data access size by a factor of 4, albeit with the limitation that accesses must be spread uniformly across all 4 quadrants)




That all being said im sure board partners will provide us with a various memory options. However the 8000MHz speed is unlikly to change as xdr2 is a design specification from rambus and not manufactured by anyone other than now AMD. If im going into way to much detail im sorry but some on here want cold hard facts not fanboyism.

keep them questions coming. peace
 
Last edited:
Thanks for all this info, certainly cleared up a few things for me except when they will be here! If I can afford one then the 7950 looks good for me but as someone said it needs to be around £220. If not then maybe a 7870\50 CF again depending on cost.:)
 
This might be the first AMD generation I have skipped in a while - depends how much I can get for my 6990 when the 7990 comes out. Only cost me £50 to upgrade from my 5970 to 6990, so fingers crossed :D
 
Those gcn cards are very interesting. Am I right in thinking that means we could potentially accelerate any application with these gpus? Or at least devs would be more likely to implement gpu acceleration as it would be easier to implement? I can certainly say I like the idea of using a powerful gpu to accelerate any task :)
 
Do we know if the move to GCN brings any direct graphics performance increase because otherwise it's only 33% more cores and 10% more MHz over the 6970 cards.
 
Lets hope they dont turn out like bulldozer. Im not planning on graphics card upgrade. But i want to see the results of these new 7series!
 
harmony said:
@Farooq: Are you sure GCN will use rambus? I think I read somewhere that the rambus rumours are fake, and GDDR5 is still more likely.
Click to expand...

GCN(graphics core next) will use XDR2 its been designed around it especially.
the VLIW4 revision cards will use gddr5 at a higher clock.
 
You must log in or register to reply here.
Back
Top Bottom