AMD - Piledriver chitchat

[CAT-THE-FIFTH]

It looks like Intel needs to pay compensation over P4 benchmarking:

http://www.extremetech.com/computing...ng-shenanigans

[watercooled]

Seems a bit pathetic though doesn't it, especially if you consider the damage potentially caused to competition because of stuff like this. And is it USA only?

[CAT-THE-FIFTH]

Nvidia Denver benchmarks out:

http://www.anandtech.com/show/8670/g...inary-findings

[DanceswithUnix]

Nvidia Denver benchmarks out:

http://www.anandtech.com/show/8670/g...inary-findings

Shows some potential, but using firmware from two months ago when Lollipop has only just gone AOSP.

If I understand how the CPU works correctly (and I might not as there haven't been that many details shown so far) driver updates could make it go a lot faster in the future, which is pretty weird for a CPU but between Nvidia and the VISC cpu that was announced recently we might be seeing more of that.

[scaryjim]

... driver updates could make it go a lot faster in the future ....

Driver updates, for an Nvidia SOC on Android? Tell that to people who invested in a first gen tegra device.... (or maybe that was just the LG O2X....)

[watercooled]

Yeah after seeing the benchmarks I came to largely the same conclusion as the author; it's a bit bimodal. It's an especially interesting design of course, and it will definitely be worth waiting to see how the release 'firmware' changes things.

One of the notable comparisons is to Apple's Cyclone core, which is also a wide but more conventional design. It seems Denver can achieve the same sort of speed as Cyclone for certain code, but fall far behind on others. Of course, a lot of the benchmarks used for mobile e.g. Sunspider seem to be impacted just as much by software as they are hardware, so they're not always the most reliable way of judging hardware performance.

To get some meaningful comparisons to other cores we'd need to see die size and power data - it wouldn't be too impressive if the performance is achieved at the expense of these.

Also, the sceptic in me is wondering if the code morphing software is susceptible to benchmark gaming? E.g. we could see very specific optimisations added to the firmware for benchmarks, the sort of thing which wouldn't be done, and wouldn't necessarily even be feasible, for the numerous other applications.

[DanceswithUnix]

Driver updates, for an Nvidia SOC on Android? Tell that to people who invested in a first gen tegra device.... (or maybe that was just the LG O2X....)

My Tegra based HTC One-X got updates for a couple of years, so it is possible.

[watercooled]

http://anandtech.com/show/8715/samsu...a-counter-suit

Well I didn't see that one coming.

[CAT-THE-FIFTH]

Looks like the A8X has an extremely fast IGP:

http://www.anandtech.com/show/8716/a...than-i-thought

[DanceswithUnix]

Looks like the A8X has an extremely fast IGP:

http://www.anandtech.com/show/8716/a...than-i-thought

Very nice, though as always I find myself wondering when they are going to crank up the voltage and stick it in a laptop. Perhaps when they get to 4 cores, I think people expect that many these days.

My Tegra based HTC One-X got updates for a couple of years, so it is possible.

Another thought on this, any code morphing software should be broken out and put on the play store.

[CountBartok]

Allways nice to see some good AMD news: AMD Awarded $32 Million for 'Extreme Scale' High-Performance Computing Research
http://www.techpowerup.com/207244/am...-research.html

[watercooled]

Interesting! I wonder if the recent opening up of HSA on Linux has anything to do with this announcement? It's good to see HSA getting some high-profile attention either way; it seems able to considerably accelerate certain compute tasks but there's still a lack of widespread supporting software.

They also mention memory technology, could be something to do with their partnership with Hynix I guess? Processing-in-memory is something else which has drawn my interest, so it will definitely be interesting to see where that goes!

[bobhope2005]

Good news for AMD. Absolutely need them to keep snapping at Intel's heels if only to make sure Intel keep pushing the boundaries as well.

[CAT-THE-FIFTH]

Looks like AMD Carizzo will be 3.1 billion transistors in size and 244MM2:

http://www.eetimes.com/document.asp?...&page_number=2

AMD will disclose Carrizo, an integrated processor with its latest x86 core. The 28 nm chip measures 244.62 mm2 and packs more than 3.1 billion transistors. Its new Excavator core is 23% smaller and uses 40% less power than AMD’s previous x86 core.

Kaveri had 2.41 billion transistors and the same die area. I suspect the increased number of transistors is for the integrated FCH.

HDL used??

[DanceswithUnix]

Looks like AMD Carizzo will be 3.1 billion transistors in size and 244MM2:

http://www.eetimes.com/document.asp?...&page_number=2



Kaveri had 2.41 billion transistors and the same die area. I suspect the increased number of transistors is for the integrated FCH.

HDL used??

So using just 22% more area than 4 cores, they can now have 6 cores?

[scaryjim]

ts new Excavator core is 23% smaller and uses 40% less power than AMD’s previous x86 core.

40% less power than Steamroller? So we're looking at the the 45W Excavator parts producing the performance of 65W Steamroller? And I guess 65W parts with ~95W Steamroller performance?

I'd take a 65W with equivalent performance to A10-7850 I guess the question is, since they've obviously power-optimised again, what has happened to the performance of the flagship parts. And will we finally see an FX version based on a newer-than-piledriver core? (I'm guessing no to the last question!)

So using just 22% more area than 4 cores, they can now have 6 cores?

I did wonder about that - either small die area increase, or reduce the number of shaders in the IGP (6 cores + 384 shaders? sounds reasonable to me....)