Showing results for 
Search instead for 
Did you mean: 

Archives Discussions

Adept III

290X vx 280X from a OpenCL compute point of view

Are there any documents detailing the differences between the 290X and the 280X from a compute point of view?

I need double precision performance (which favors the 280X) , but want to know what I'm giving up.

7 Replies


I am not getting particular document found right now.

But i found some of the link which actually gives you the details about Radeon 290X and 280x  with reviews

Please have a look

there are severa pages in this ...check all may be useful for you

Adept III

This is all I have at the moment.

From Radeon R9 290X Review: AMD's Back In Ultra-High-End Gaming - Hawaii: A 6.2 Billion Transistor GPU Fo...

"There are a few tweaks to the design though, including device flat addressing to support standard calling conventions, precision improvements to the native LOG and EXP operations, and optimizations to the Masked Quad Sum of Absolute Difference (MQSAD) function, which speeds up algorithms for motion estimation."

And AnandTech Portal | The AMD Radeon R9 290X Review

"Finally GCN 1.1 also introduces some new instructions, including a Masked Quad Sum of Absolute Differences (MQSAD) and some FP64 floor/ceiling/truncation vector functions."

and of course the fact that the DP ratio is 1:8 on the 290X and 1:4 on the 280X

But I'd like the full technical detail so I can assess how these extra features might benefit my application.

Does "device flat addressing" mean the 290X supports HSA over Pcie?


I read in some reviews that firepro cards will do 1:2 ratio between sp and dp. But of course I am not sure how trustworthy the reviews may be. Why dont you get a firepro card when it is available?


Is there a technical reason why AMD has crippled (1/4 to 1/8 ratio) the double-precision performance on the R9 290X GPU?


R290X is quite big chip. maybe AMD want save some size.



As of now i can say that its hardware limiation. But we have an OpenCL extension cl_khr_fp64 or cl_khr_fp16 for it.

Double-precision is now an optional core feature instead of an extension in OpenCL 1.2.

Also i read in one of site that

"AMD changed the double-precision rate from 1/4 to 1/8 on the R9 290X, yielding a maximum .7 TFLOPS. The FirePro version of this configuration will support full-speed (1/2 rate) DP compute, giving professional users an incentive to spring for Hawaii's professional implementation"

Adept III

I believe the 290X has the features mentioned in

The section "Differences Between Southern Islands and Sea Islands Devices"

I want two gpus so I can write and debug multi gpu code.

In Australia I can get two 280X's for $399x2 = ~800 which has 2000Gflops of DP verses a single 290X for $700. which has 700Gflops of DP.

I might have considered 2x290X for $1400 if I was getting 1400*2 = 2800 Gflops of Dp, but that has not happened.

The only other affordable option for multi gpu gcn 1.1 is two r7 260X which at most will have ~500Gflops of DP ( I believe 250Gflops *2).

>The FirePro version of this configuration will support full-speed (1/2 rate) DP compute, giving professional users an incentive to spring for Hawaii's professional implementation

That would be great... but I'm guessing $4000+ for the card and I still need two of them.