AMD - Piledriver chitchat

[CAT-THE-FIFTH]

Here is a review of the Toshiba Satellite L855D-100:

http://translate.googleusercontent.c...HzxQCw#content

It uses the Trinity A8 CPU which has a slower IGP with 256 shaders.

Edit!!

Here is the Toms Hardware review:

http://www.tomshardware.com/reviews/...iver,3202.html

They compare the 35W TDP A10-4600M against the 35W A8-3500M. The highest performing 35W TDP Llano part is the A8-3520M which is clocked 100MHZ higher than the A8-3500M.

It will be interesting to see if AMD releases 45W TDP parts and how well they will do against the 45W TDP A8-3550MX.

[CAT-THE-FIFTH]

It seems HandBrake supports OpenCL now:

http://www.anandtech.com/show/5835/t...ds-trinity-apu

It seems AMD VCE is slower than Quick Sync although I do wish there was some comparisons of what the final encode look like. However,the improvement in HandBrake performance is impressive.

[badass]

It seems HandBrake supports OpenCL now:

http://www.anandtech.com/show/5835/t...ds-trinity-apu

It seems AMD VCE is slower than Quick Sync although I do wish there was some comparisons of what the final encode look like. However,the improvement in HandBrake performance is impressive.

It was my understanding that AMD's VCE was like intel's quicksync - i.e. some fixed function logic and some use of the shaders and an SDK provided to make use of that combination. If that's the case then VCE should be faster than indicated by that graph.

[kalniel]

It was my understanding that AMD's VCE was like intel's quicksync - i.e. some fixed function logic and some use of the shaders and an SDK provided to make use of that combination. If that's the case then VCE should be faster than indicated by that graph.

No reason to suspect AMD's fixed function encoder would be anything like as fast as Intel's. But we've yet to see it in GCN yet (this is still a 6000 series GPU). However, I'm not holding my breath - it's still missing from GCN when it was promised ages ago.

[badass]

From what I have gathered from just the Anandtech review:
Single threaded performance is better than Llano. This was direly needed. Looks like around 20% better.
Multi threaded performance is a wash. That is IMHO irrelevant. Few laptops need 4 threads.
Power consumption is substantially better.
GPU performance is better, though not by as much as AMD had claimed ~30%

TBH I'm disappointed. The A10-4600M is great by the looks of things but I was hoping it would have more than a 20% average lead over Ivy Bridge GPU wise.
However I still think that Trinity is far more relevant than 4c Ivy Bridge for any laptop.

No reason to suspect AMD's fixed function encoder would be anything like as fast as Intel's. But we've yet to see it in GCN yet (this is still a 6000 series GPU). However, I'm not holding my breath - it's still missing from GCN when it was promised ages ago.

WHat I was trying to say is that is using openCL - i.e. it's just using shaders where appropriate. If there is fixed function logic to accelerate transcoding that should be faster still than using the shaders.

[CAT-THE-FIFTH]

It was my understanding that AMD's VCE was like intel's quicksync - i.e. some fixed function logic and some use of the shaders and an SDK provided to make use of that combination. If that's the case then VCE should be faster than indicated by that graph.

Its hard to say what is happening,but until more articles are released,it does look like Quick Sync is faster. However,we need to see if video output quality is the same. If VCE is better quality then,it might be slower.

However,the HandBrake results are quite impressive especially for Trinity and HandBrake is not only widely used but free. AFAIK,the HD4000 is not as good as the AMD IGPs in compute and GCN is even better in this regard with regards to OpenCL performance.

Edit!!

Kaveri is suspected to use a 512 shader IGP similar to the GPU in the HD7750:

http://www.tomshardware.com/reviews/...k,3135-13.html



Even with the lower clockspeeds and bandwidth reduction, could you imagine how useful the IGP is going to be for things like HandBrake??

The major issue is how AMD is going to deal with the bandwidth needed. Haswell will probably use stacked RAM and AMD seems to be looking at it too(there is a picture of an AMD GPU with it).

[CAT-THE-FIFTH]

The TechReport review is up:

http://techreport.com/articles.x/22932/1

Overall,it seems the latency of the HD7660G is much better than the HD4000.

[CAT-THE-FIFTH]

It seems Diablo3 runs much better on the HD7660G:

http://www.legitreviews.com/article/1928/2/

Edit!!

Picture of a Trinity ultrathin laptop:

http://www.expertreviews.co.uk/proce...ors-launched/2

[Biscuit]

Its hard to say what is happening,but until more articles are released,it does look like Quick Sync is faster. However,we need to see if video output quality is the same. If VCE is better quality then,it might be slower.

Doubt it will be any different to be honest. The quality will be defined by the encoding algorithms used rather than the hardware its used on and im assuming handbrake doesnt pick and choose between different algorithms depending on the hardware.

[kalniel]

Edit!!

Picture of a Trinity ultrathin laptop:

That's where I'm excited. We need more competition in this space as there's a massive premium being charged at the moment.

[badass]

Its hard to say what is happening,but until more articles are released,it does look like Quick Sync is faster. However,we need to see if video output quality is the same. If VCE is better quality then,it might be slower.

However,the HandBrake results are quite impressive especially for Trinity and HandBrake is not only widely used but free. AFAIK,the HD4000 is not as good as the AMD IGPs in compute and GCN is even better in this regard with regards to OpenCL performance.

Edit!!

Kaveri is suspected to use a 512 shader IGP similar to the GPU in the HD7750:

http://www.tomshardware.com/reviews/...k,3135-13.html



Even with the lower clockspeeds and bandwidth reduction, could you imagine how useful the IGP is going to be for things like HandBrake??

The major issue is how AMD is going to deal with the bandwidth needed. Haswell will probably use stacked RAM and AMD seems to be looking at it too(there is a picture of an AMD GPU with it).

I hope they (Kaveri APU's) either have a large ish on die shared L3 cache, a Large ish on die L3 cache exclusive to the GPU or some kind of very large cache (>64MB) on package with a suitably fast, low latency interface to the APU.

I suspect that with Desktop Trinity when it comes out, GPU performance will increase almost linearly with RAM speed again.

[DanceswithUnix]

Charlie seems to like Trinity.

http://semiaccurate.com/2012/05/17/i...an-it-appears/

[CAT-THE-FIFTH]

Charlie seems to like Trinity.

http://semiaccurate.com/2012/05/17/i...an-it-appears/

He usually lambasts AMD CPUs too.

[Noxvayl]

Hopefully the Piledriver enhancements for desktop parts are impressive enough for similar remarks.

[dfour]

Hopefully the Piledriver enhancements for desktop parts are impressive enough for similar remarks.

Reality says they wont scratch ivy bridge but it would be good for AMD if they matched olde sandy bridge in the desktop space.

[Biscuit]

To me it seems the same situation as llano. The performance for most people is enough if not more than enough but the public isn't informed enough to know that. Intel win the business because in benchmarks they rule.
The point Charlie made is quite good in that the actual real life power usage for Trinity is impressive, perhaps that along with the fact that it seems to be marketed in similar 'ultra slim' and stylish packaging but at a lower price point will be enough to damage Intel.