AMD - Piledriver chitchat

[DanceswithUnix]

More reporting on AMDs ARM chips:

http://www.fudzilla.com/home/item/32...ing-in-2h-2014

10Gb Ethernet, not going to fit in FM2+ then.

They must think it is fairly quick to bother putting a pipe that fat on it.

[DanceswithUnix]

Rather nice FM2+ ITX board from Gigabyte.

http://www.fudzilla.com/home/item/32...2%20-itx-board

Normally I would consider WiFi & Bluetooth a waste on desktop machines, but I wonder how useful they would be on a Steam box.

[scaryjim]

They must think it is fairly quick to bother putting a pipe that fat on it.

They're aiming at data centre usage, so presumably they're thinking NAS/SAN usage and high density clusters, so massive network bandwidth can only help in those situations. Not surprising to see that kind of feature support on what is primarily a server chip...

[DanceswithUnix]

Richland is faster under Windows 8.1

http://semiaccurate.com/2013/10/22/w...s-apus-faster/

Fails to say whether i3/i5 is also faster under Windows 8.1, but then if this comes from an AMD press release they wouldn't mention that would they

[CountBartok]

AMD announces Heterogeneous Queuing tech http://www.bit-tech.net/news/hardwar...10/22/amd-hq/1.

New Mac pro has two AMD GPUs http://www.theverge.com/2013/10/22/4...pro-2013-price.

Which seems to go with this announcement http://www.bit-tech.net/news/hardwar...ion-graphics/1

[DanceswithUnix]

Interesting piece on ARM SOCs:

http://www.fudzilla.com/home/item/32...nd-interesting

Most interesting item for me was at the end, don't think it spoils the plot to say they put the Exynos 5 Octa core as $28. That isn't a market that AMD are targetting yet with their up-coming ARM chips, but there really isn't much margin down there is there.

[CAT-THE-FIFTH]

Interesting piece on ARM SOCs:

http://www.fudzilla.com/home/item/32...nd-interesting

Most interesting item for me was at the end, don't think it spoils the plot to say they put the Exynos 5 Octa core as $28. That isn't a market that AMD are targetting yet with their up-coming ARM chips, but there really isn't much margin down there is there.

Its more of a problem for Intel. They spend loads of money on process nodes(more than TSMC or GF),so it makes me think something has got to give if they want to compete. It will probably mean they need the desktop and laptop large core based products to subsidise the lower end.

[kompukare]

Its more of a problem for Intel. They spend loads of money on process nodes(more than TSMC or GF),so it makes me think something has got to give if they want to compete. It will probably mean they need the desktop and laptop large core based products to subsidise the lower end.

This being Intel, I'm sure they'll come up with creative (think monopolistic) way of taking marketshare away from ARM.

Just read that the Apple A7 is 102mm². That makes it only a bit smaller than Cape Verde or GK107 and bigger than Tegra4 (sort of hard to get die-sizes for most SoCs). While what Samsung are charging Apple per waffer is (presumably) unknown, it looks like Apple are spending a fair bit on their processor (since it's their own design it's hard to say how much buying PA Semi and all that cost them).

EDIT: Ah, Apple watching seems to be major hobby for analysts:
http://allthingsd.com/20130924/teard...73-for-the-5c/
The breakdowns quoted there think the A7 is $19. No word on how much Apple's "PA Semi" team costs of course. Also, looks like for the iPhone 5, Apple spent a lot more than £19 on radio chips. Guess that is where Qualcomm and the like make most of their profit.

[DanceswithUnix]

This being Intel, I'm sure they'll come up with creative (think monopolistic) way of taking marketshare away from ARM.

lol, I'm sure they will try. Their usual tricks probably won't be effective.

They are trying FUD already telling people they have better battery life etc, no-one seems to be listening.
They can't withhold design information & licenses like they did with VIA and Nvidia because it is a different eco system so none of that counts.
They can't cripple compiler support because their compilers don't support ARM in the first place.
They can threaten to cut allocation of silicon to people who make competing designs like they did against AMD motherboard builders, but in a stagnant PC market I don't think anyone will care.
They can try a patent trolling exercise, but one of the people they will go up against is Apple so good luck to them with that

Interesting times...!

[kompukare]

Interesting times...!

That's Intel: so desperate they've resorted to ancient Chinese curses!

But all your points are perfectly true. Another I've thing which fills me with a bit of schadenfreude is that for years Intel crippled Atom so badly and now they have to reap that.

Of course the main reason for Atom was to keep old Intel fabs busy churning out chipsets on old nodes. However bad Atom was, its early chipsets were ten times worse. But then again 'perfect' Intel does seem to a have a history of things going wrong with their chipset: after the billions they spent on the SB chipset recall the Haswell USB suspend bug should never made it live.

[DanceswithUnix]

Tom's hardware hardware says the 9370 now comes with a bundled water cooler?

http://www.tomshardware.co.uk/gaming...iew-32806.html

A quick look around (Scan, Ebuyer) it looked like you still get just the CPU, but if this actually happens at the same price then that is insanely good value.
But then this is Tom's, so slightly better than reading it in the Daily Mail

[CountBartok]

[QUOTE=CAT-THE-FIFTH;3068189]Its a shame if Kaveri is not released this year!!

It seems AMD has won a contract with Verizon:

http://venturebeat.com/2013/10/07/ve...micro-servers/

Shame there almost all with Intel chips.
http://www.extremetech.com/computing...-win-for-intel
Although AMD will still make some profit from it.

[CAT-THE-FIFTH]

[QUOTE=CountBartok;3087477]

Its a shame if Kaveri is not released this year!!

It seems AMD has won a contract with Verizon:

http://venturebeat.com/2013/10/07/ve...micro-servers/

Shame there almost all with Intel chips.
http://www.extremetech.com/computing...-win-for-intel
Although AMD will still make some profit from it.

http://www.moorinsightsstrategy.com/...verizon-cloud/

On an analyst conference call, I asked Verizon whose processors were being used, AMD or Intel, or both, and here is what Pat and I got back, word-for-word…

The answer is both, we are using the Intel Xeon class processors and a bunch of our infrastructure is using the Intel. Recently we switched to AMD Opterons, and as you wonder why, one of the things we are looking for is increasing the memory per host and the Opterons allow us in a single socket configuration to address more memory, and so all of the new deployments we are putting out there are carrying 64GB per host and the 8-core Opteron processors.

[CAT-THE-FIFTH]

Kaveri is released in February:

http://wccftech.com/amd-launches-kav...#ixzz2ixA6PLMi

Fail.

Edit!!

I do hope the Kaveri we get has the rumoured RevB SR core and has the rumoured 700+ shader IGP section.

Second Edit!!

The slides say SR RevB,so hopefully a decent IPC increase,especially considering Kaveri should have been out a long time ago AFAIK.

Third Edit!!

It seems NDA might be over on the 5th of December??

[CAT-THE-FIFTH]

I was looking at this AMD presentation on Kaveri which was released last month:

share.csdn.net/uploads/5232b691522ba/5232b691522ba.pdf

If you look at page 21,you can see a representation of Kaveri. It looks like the GPU is depicted as having 6 CUs,which means 384 shaders in total.

[scaryjim]

384 shaders is the sensible number to ally the R5 graphics core with an R7 240/250 in Dual Graphics. If it's using GCN 1.1 we should still see a decent lift over VLIW4 Trinity/Richland. While they're on a dual channel DDR3 platform it makes no sense to stick a significantly better graphical processor in an APU: it'll just get starved for bandwidth. The R7 240 uses a 128bit 1600MT/s DDR3 interface, which is basically identical to the interface Kaveri will have - in fact Kaveri may be able to access slightly more bandwidth as it will be able to use a higher memory clock (although AMD will need to improve the efficiency of their IMC if it's going to realise that potential). So they should provide very similar performance, making them ideal for dual graphics pairing. Assuming the top end Kaveri comes in around £100 again, just over £150 will grab you 768 shader's worth of dual graphics performance. Should be quite tasty

Anyway, compare the performance of an R7 240 and an R7 250 (or indeed a DDR3 7750 and a DDR5 7750), and then tell me why AMD would put an R7 260X's worth of shaders on a platform with a 128bit DDR3 memory interface. They wouldn't. Makes no sense.