AMD - Piledriver chitchat

[scaryjim]

hmmm, just goes to show the stackoverflow geeks don't know everything

I reckon SHA512 should be sufficiently secure for storing passwords for my mate's website though. It's not like he's actually paying me for doing it....

[DanceswithUnix]

hmmm, just goes to show the stackoverflow geeks don't know everything

I reckon SHA512 should be sufficiently secure for storing passwords for my mate's website though. It's not like he's actually paying me for doing it....

If someone cracks into the site to steal the password file, and then feeds that file through a bunch of video cards to brute force the plaintext of the passwords, will it have been worth the attacker's time?

I think for most people what you have done will be fine.

[DanceswithUnix]

Rumours of Kaveri delays:

http://www.fudzilla.com/home/item/32...-delay-sort-of

[watercooled]

CD explicitly says otherwise, 5th paragraph:

http://semiaccurate.com/2013/08/07/a...aii-in-hawaii/

[scaryjim]

Depends on how you see the roadmap as being read. If they paper-launch Kaveri in December, they've still hit Q4 2013, even if it doesn't get into the channel until February.

More interesting thing is that the fudzilla article points out the lack of Kabini in the channel: so far I can only find one device, an Acer netbook, for £350 at Argos, and tbh it strikes me as being a bit pricey for an 11.6" device with the 1GHz version, would've expected nearer £300 for what is basically a netbook replacement. It'd be nice to see one of the higher-clocked quad cores (A4-5000/A6-5200) and a 13.3" screen for < £500; that might well get my money out of my pocket

[kompukare]

You'd think they'd have learned their lesson about paper launches in December. Aside from the speculation never being good for current sales, the 7970 launch (which AFAIK happened in December because Rory had said they'd ship 28nm GPU for revenue that year) had two major problems:
1) drivers weren't really ready (but waiting until June would have been crazy) and
2) the yields at the time were not so good so the launched 7950/7970 at the kind of speeds which made them look silly when 670/680 launched.

Of course a short time after that, the GHz editions and the new drivers changed all that but the damage with the initial reviews had already been done and Nvidia were able to sell their slower 670/680 for higher prices.

So launching Kaveri while yields are poor or with too high a voltage (& hence power consumption) might not be a good idea.

[watercooled]

I'm also puzzled by the lack of Kabini products, surely yield can't be a major issue now - it's only a smallish die on a fairly mature 28nm process! I wonder if there's any foul-play, or if manufacturers are cautious about releasing new netbooks/tops with the popularity of similarly-priced tablets?

Nonetheless it would be nice to see them used in things like NAS boxes, and at least a few mini-ITX boards would be nice!

On the subject of the Kaveri delay, like scaryjim says I think it's more of a case of interpretation. A lot of places are just jumping on the bandwagon claiming there's been some delay/setback. The more recently 'revealed' plans, if true, might have been planned all along.

'New CPUs not as soon as we assumed when we looked at a roadmap leaked by some other site' doesn't make for as much of an attention-grabbing/sensationalistic headline as 'moar AMD delays!'

[CAT-THE-FIFTH]

AMD has gained CPU marketshare:

http://investorvillage.com/mbthread....6&tid=12970932

Retail AMD CPUs appear to be quite radiation tolerant:

http://semiaccurate.com/forums/showthread.php?t=7322

[teppic]

I've read some rumours that AMD isn't going to release an AM3+ Steamroller CPU (just the APUs) - is there anything to support that? It'd really surprise me given that they've talked about it being such an improvement over Piledriver. Assuming it is released, is there any realistic possibility it won't be AM3+?

[kompukare]

I would have though it really depends on which Opterons are released. According to the Wiki list of Operons, Bulldozer had three AM3+ models and Piledriver had four. The other Opterons were C32 for dual socket and G34 for 2/3 socket.

Basically, I expect them to release AM3+ consumer FX CPUs of anything which they release as Opteron for the same reason that Intel sells LGA2011: once the server chips have been validated selling a consumer version must be fairly little extra work. I was going to add acaveat about release times of FX vs Opterons but having just looked at the Wiki on FX CPUs it seems they were released at almost the same time.

EDIT: although the AM3+ chipsets are looking at bit dated in terms of features and power consumption; unsure if AMD plan to fix that though. Certainly they don't seem to think of chipsets as a cashcow unlike Intel.

[DanceswithUnix]

I've read some rumours that AMD isn't going to release an AM3+ Steamroller CPU (just the APUs) - is there anything to support that? It'd really surprise me given that they've talked about it being such an improvement over Piledriver. Assuming it is released, is there any realistic possibility it won't be AM3+?

A while back someone at AMD said something along the lines of "We know we don't have the fastest CPU at the moment, so we are pushing mobile tech where we have an advantage".

It is a long way from there to "AMD will never release another server class CPU", but people will interpret things they way they want to.

The roadmap does imply a new 2P/4P platform Opteron in early 2014, and it would be easy for AMD to release an AM3+ version of that.

http://community.amd.com/community/a...market-in-2014
http://www.theregister.co.uk/2013/06..._server_chips/

[teppic]

I really hope they do release it. So many people will be disappointed otherwise, as everyone's been quite patient with Piledriver's flaws knowing that an upgrade to Steamroller is just a BIOS update for the board. The 990FX chipset seems ok though?

[kompukare]

990FX is a good chipset aside from the lack of USB3 (easily added to most motherboard). Native SATA3 ports and plenty of PCI-E lanes. And good value and longevity for early adopter which is always nice.

No, my problem is the upto 20W TPD for a chipset. I guess it competes with X79 which is also far from frugal (those recent FX9550 reviews reminded me how much power LGA2011 uses), and the Intel's promises of Haswell's LGA1150 chipset being more power efficient doesn't seem to have materialised. Despite Intel's near infinite budget, FM2 still has better idle power than Haswell.

[teppic]

Yeah I noticed that the Sabertooth can run PCI-E 3.0 at 16x/16x, which Z87 can't do. The FM2 chipsets seem great - if only it'd take the FX processors.

[watercooled]

I've read some rumours that AMD isn't going to release an AM3+ Steamroller CPU (just the APUs) - is there anything to support that? It'd really surprise me given that they've talked about it being such an improvement over Piledriver. Assuming it is released, is there any realistic possibility it won't be AM3+?

There's nothing much to support an argument either way, it's all just speculation really. I don't think there are any valid 2014 desktop roadmaps available, but looking at the server roadmap it looks like they're pushing the Berlin APU platform into the single-socket market, with no mention of 8 core CPU-only parts.

990FX is a good chipset aside from the lack of USB3 (easily added to most motherboard). Native SATA3 ports and plenty of PCI-E lanes. And good value and longevity for early adopter which is always nice.

No, my problem is the upto 20W TPD for a chipset. I guess it competes with X79 which is also far from frugal (those recent FX9550 reviews reminded me how much power LGA2011 uses), and the Intel's promises of Haswell's LGA1150 chipset being more power efficient doesn't seem to have materialised. Despite Intel's near infinite budget, FM2 still has better idle power than Haswell.

What are you talking about, Haswell is so low power it needs new special power supplies!

Seriously though, AM3+ does the job but it could do with an overhaul really - it's not terribly power efficient compared to modern chipsets like FM2, and is still a two-chip solution. I'm not sure how that would realistically be worked out though.

I guess it would be possible to release FX parts for the FM2 platform without any GPU (there are already those Athlon parts); having all 8 cores, associated cache and an AMD-sized IGP might run into die size problems.

Maybe they're planning on interposers for the next FX? Either way, they probably want to get Kaveri at 28nm shipping with decent yields before they consider the server stuff.

[watercooled]

Has anyone heard anything about VCE lately? Besides the release info, and the fact driver support was delayed, I've read very little about it, media attention has all but disappeared lately. Presumably driver support is there now, and I've heard of other potential uses for it e.g. video streaming. It would be good if hardware encode could be utilised for things like remote desktop amongst other things.

IIRC some software which claimed to support it (or was it just reviewers claiming that) was actually 'only' using OpenCL rather than the fixed-function logic.

I saw a few benchmarks on Tom's, claiming to use VCE, but they weren't hugely faster than software encoding. I'd like to see power draw measurements during those tests though - hardware encoding can use a small fraction of the power required for software encoding. Even mobile phones have some hardware encoders for things like the camera, and if it supports wireless display. I've not seen anything about 'hybrid mode' either, supposedly allowing higher performance but lower efficiency by using the GPU shaders.