AMD Kaveri Review : Focused on HSA Features

[DGLee]

After I'd dealt with Kaveri's Processor Graphics thoroughly, I focused on its HSA features which as of today have little meaning for everyday computing but also are believed to provide wider chances to HPC-level GFLOPS in sub-$200 chips. (A full CPU-aspect review on Kaveri is here : <redacted>

Today, I want to introduce and share some HSA-based benchmarks. Let's have a look.

(All comparison groups are featured GTX 780 Ti and thus unable to boost their score via Accelerated mode in the most recent version of PCMark 8 except Kaveri : Its internal GPU wouldn't be disabled whether or not a discrete Graphics Card is installed. Actually user can manually choose to use it or not. Indeed, CPU-only OpenCL acceleration actually slows the performance down just like a CPU is trying to handle PhysX by its own)


redacted
Self Promotion is not allowed
Markfw900
Anandtech Moderator

How's your opinion? Is Kaveri an attractive chip?

 

[Arkadrel]

(All comparison groups are featured GTX 780 Ti and thus unable to boost their score via Accelerated mode in the most recent version of PCMark 8 except Kaveri.....

It just shows that HSA works.

The fact that a A10-7850k' IGP is beating a "Core I5 4670k + Geforce GTX 780 ti" in GPGPU workloads where HSA works, is pretty cool.


It depends on how many HSA enabled programs we ll see in the future.
I really hope AMD is doing everything it can to speedup the growth of HSA programs.

 

[PPB]

Wait wait, weren't those OpenCL benchs to begin with?

If not, that means that we have already HSA enabled software?

 

[jhu]

LibreOffice has GPU compute capabilities? Interesting… Need to look into this further.

 

[Paul98]

Interesting, I wonder how much ram timings and speed effect's this performance.

 

[Arkadrel]

redacted
Self Promotion is not allowed
Markfw900
Anandtech Moderator

I have no idea how to write a message to a mod, so bare with me here.
I swear if I get a mod call out infaction.... <.<

You might as well have deleted the entire thread, once you took down the pictures of the benchmarks & their results.

Couldnt you just have removed the link text to the site, instead?
(ironically that gets to stay, but the benchmark results get removed?)

the tread is kinda worthless without seeing how it actually does in HSA programs. Also its something Anandtech didnt even show case, so its usefull stuff, it shows the "concept" behinde the APUs is working.
<accidentally edited, see post below>

 

[Arkadrel]

Im not self promoteing, or promoteing a site.

Here are the benchmark results, without which the thread has zero meaning:
(A10-7850k + its IGP vs "All others with a GTX 780ti" )

 

[inf64]

So wait, if one of us,other AT forum users (so not DGlee) posted a thread with the link to this review, would that be OK?
Results are really good, the chip is crushing everything in HSA enabled workloads.

 

[Arkadrel]

So wait, if one of us,other AT forum users (so not DGlee) posted a thread with the link to this review, would that be OK?

The link to his own site was the only thing that got to stay

The mod removed the benchmark results instead.

I posted the benchmark results without a link above, hopefully that gets to stay.

 

[Erenhardt]

Sorry by I don't believe those graphs - it cant be true, otherwise every site would write an article about it and AMD would be all over the place with its HSA performance.

 

[inf64]

Sorry by I don't believe those graphs - it cant be true, otherwise every site would write an article about it and AMD would be all over the place with its HSA performance.

LibreOffice results are genuine , AMD was even using it on Kaveri launch day...
I see no reason for the rest not to be genuine. Anyone with Kaveri can find the workloads on the internet, DL them and test it out. What would be the point to fake anything these days when everything is testable so easily?

 

[Markfw]

I have no idea how to write a message to a mod, so bare with me here.
I swear if I get a mod call out infaction.... <.<

You might as well have deleted the entire thread, once you took down the pictures of the benchmarks & their results.

Couldnt you just have removed the link text to the site, instead?
(ironically that gets to stay, but the benchmark results get removed?)

the tread is kinda worthless without seeing how it actually does in HSA programs. Also its something Anandtech didnt even show case, so its usefull stuff, it shows the "concept" behinde the APUs is working.
<accidentally edited, see post below>

Actually I missed the link, but its gone now.

So here are the rules. If you like a review (like Anandtech.com or tomshardware.com, etc...) you can link to if so others can share.
If you have a benchmark of your own personal computer that you want to share, you can link a picture of that.

What you CAN'T do is link to your own website, or link to pages of benchmarks of the 5-10 or more computers you have in your "lab", as this person has done twice now. That is self-promotion. It doesn't have to be a company, but its obvious its not his one or two personal computers.

The next time may be ban time.

 

[OCGuy]

AMD executives were having a hissy fit on twitter that reviewers didn't use HSA, so you will be seeing more of a push trying to market this cherrypicked "data".

Anyone dying to see this ad can check other forums where it is spammed.

 

[Erenhardt]

LibreOffice results are genuine , AMD was even using it on Kaveri launch day...
I see no reason for the rest not to be genuine. Anyone with Kaveri can find the workloads on the internet, DL them and test it out. What would be the point to fake anything these days when everything is testable so easily?

To test if the results are real, you need to have kaveri. That means you have to buy it to - which would be a point of misleading article.

If it was real anand would jump on it right away - not the weak game benches they posted.

But that is just my logic.

 

[inf64]

Well I am telling you that LibreOffice results are true. You might not believe that or think "it's too good", that changes nothing. Check Kaveri launch slides and you will find the LibreOffice benchmark data.

Many reviewers still have Kaveri so they can test it out.

 

[Arkadrel]

Sorry by I don't believe those graphs - it cant be true, otherwise every site would write an article about it and AMD would be all over the place with its HSA performance.

Id like a more creditable source too than random guy in forums threads.

I wish anandtech had done some testing to see if amd's HSA claims where true.

 

[rainy]

Sorry by I don't believe those graphs - it cant be true, otherwise every site would write an article about it and AMD would be all over the place with its HSA performance.

Those results are not fake, here is an example: http://www.extremetech.com/computin...-wait-for-the-first-true-heterogeneous-chip/5

 

[Dresdenboy]

Id like a more creditable source too than random guy in forums threads.

I wish anandtech had done some testing to see if amd's HSA claims where true.

How should they do that? Own tests or rerunning the same stuff, others already did? Coding some own tests needs time. Maybe they'll have an article in the future covering just that.

 

[cytg111]

Yea well, it has to be this, this and mantle or bust.

 

[erunion]

Those results are not fake, here is an example: http://www.extremetech.com/computin...-wait-for-the-first-true-heterogeneous-chip/5

That article says those tests are HSA demos provided by amd. That explains why the big sites didn't emphasize them.

 

[BallaTheFeared]

That article says those tests are HSA demos provided by amd. That explains why the big sites didn't emphasize them.

Still fairly impressive.

I only get 813 with my i5-4670k @ 4.8GHz and my HD4600 @ 1.75GHz.

That said I would rather see some non AMD derived tests, obviously.

 

[Markfw]

Locking for spammed URL's on images.