AMD - Piledriver chitchat

[DanceswithUnix]

Are these supposedly genuine AMD slides? I couldn't tell when I looked at a couple of them on Twitter.

Supposedly yes, but from a paper rather than marketing bumpf.

[scaryjim]

At last ...

Indeed!

I played this video on my laptop last night. As soon as it started the wife was like "what on earth is that awful sounds?". So I explained it was the old AMD CPU cooler - then when it switched I explained that was the new one and she said "What is?"

Then I turned the volume up, and while the difference in sound power doesn't seem that great, the overall pitch is so much lower that the perceived volume is significantly lower. Although I'd be more impressed if I hadn't been messing with an old s775 system recently and therefore had a good reminder of just how quiet the stock Intel coolers for years...

[DanceswithUnix]

New APU speed bump incoming? http://www.fudzilla.com/news/process...k-flagship-apu

Seems a bit late in the lifecycle for new AM3+ motherboards though.

[shaithis]

New APU speed bump incoming? http://www.fudzilla.com/news/process...k-flagship-apu

Seems a bit late in the lifecycle for new AM3+ motherboards though.

Makes me wonder if they will be releasing at least some "Zen" CPUs on AM3+.......

[watercooled]

Not unless they include a DDR3 memory controller and HyperTransport links on the CPU die, which I don't think is likely.

[watercooled]

If anyone's interested, going from the AMD slides which put the Opteron A1100 package at 27x27mm and pixel-counting, the die size looks to be about 130mm2.

[scaryjim]

... Seems a bit late in the lifecycle for new AM3+ motherboards though.

Indeed, it's a rather worrying sign for AM4 availability. Pushing new FM2+ and AM3+ motherboards now implies that we're not going to see AM4 for quite a while.

If that's the case, I wonder if they will consider some new AM3+ CPUs, maybe with Steamroller or Excavator cores? Might be a bit late in the day for a new FX series though, and note the announcement doesn't claim any new chipsets; sounds like it's *just* pushing mobo manufacturers to stick extra chips on existing AM3+ boards. Being stuck with the same CPU line up for the best part of another year really isn't what AMD needs right now...

Plus anyone else concerned that we're getting another Kaveri speed-bump rather than any new Carrizo silicon on FM2+? Are AMD just waving a white flag in the desktop space?

EDIT:

Not unless they include a DDR3 memory controller and HyperTransport links on the CPU die, which I don't think is likely.

I wouldn't be at all surprised if they did a hybrid DDR3/DDR4 memory controller in the first gen, so the only question would be if they plan on keeping a HT link for MCMs and multi-CPU systems. Given AMD traditionally use the same silicon to serve both the consumer and server markets, it's by no means inconceivable that AM4 silicon will still have HT links baked in. Or, they might be able to use an interposer to back-fit the silicon to AM3 - depends on whether they look to move more towards system-on-chip, or system-on-interposer...

[DanceswithUnix]

Plus anyone else concerned that we're getting another Kaveri speed-bump rather than any new Carrizo silicon on FM2+? Are AMD just waving a white flag in the desktop space?

This is quite hot on the heels of the not really announced FX 6330. I suspect none of this is really intended for western consumption, we are probably only buying A8-7600 chips from them at this point anyway for things like HTPC use.

[watercooled]

If that's the case, I wonder if they will consider some new AM3+ CPUs, maybe with Steamroller or Excavator cores?

Considering the time to release a new die is fairly long, my thinking is we would have heard something by now, it's not something they could do last-minute. But even if they couldn't get 28nm to clock that well, it is puzzling why they didn't release anything at all for HEDT/Server based on Steamroller/Excavator as the performance is quite a lot better than Piledriver per core/clock. I would've thought they'd easily make their money back and then some. Maybe it was down to staffing i.e. they literally didn't have enough people to work on it at the same time as the cat/construction APUs, console semi-custom, early work for ARM/Zen, etc.

Plus anyone else concerned that we're getting another Kaveri speed-bump rather than any new Carrizo silicon on FM2+? Are AMD just waving a white flag in the desktop space?

I think Bristol Ridge is the first time we'll see Carrizo on the desktop, on AM4.

I wouldn't be at all surprised if they did a hybrid DDR3/DDR4 memory controller in the first gen, so the only question would be if they plan on keeping a HT link for MCMs and multi-CPU systems. Given AMD traditionally use the same silicon to serve both the consumer and server markets, it's by no means inconceivable that AM4 silicon will still have HT links baked in. Or, they might be able to use an interposer to back-fit the silicon to AM3 - depends on whether they look to move more towards system-on-chip, or system-on-interposer...

Good points - Skylake has a hybrid IMC however it's limited to LP-DDR3, that could pose a risk for people doing drop-in upgrades from Vishera with standard voltage memory. AMD could do without a ton of people not RTFM and killing their memory controllers. Also yeah it would make sense to include some form of HyperTransport as it would allow them to scale it to multi-socket systems on top of what you said.

[scaryjim]

AMD's hybrid DDR2/DDR3 memory controller was very voltage-tolerant, so I see no reason to think their DDR3/4 one wouldn't be. Don't forget that some of intel's DDR3 controllers were susceptible to problems using 1.65v DDR3. AMD seem to just make generally more robust IMCs, as far as I can tell...!

[CAT-THE-FIFTH]

Hmm,noticed this today:
http://cdn.wccftech.com/wp-content/u...10-635x357.jpg

Greater than 40% IPC now stated for Zen.

[scaryjim]

... Greater than 40% IPC now stated for Zen.

That rings a couple of alarm bells to me, mainly because the claim is listed under the "datacenter" heading.

Firstly, AMD's currently data centre CPUs are piledriver-based, rather than one of the newer construction cores. If that's the generation the slides are talking about, then the IPC increase against Excavator/Steamroller/wherever we're at now will be somewhat lower - according to anandtech bench a 40% IPC increase over an FX-8350 still wouldn't catch the i5 4690k on single threaded benchmarks.

Secondly, datacenters tend to run on lower-power, lower-clocked processors. For the performance claims for Zen to be made specifically in that segment makes me wonder about target clock-rates for Zen: if it won't scale up to the 4GHz+ rates that we currently have in FX CPUs, it could end up rather underwhelming...

[DanceswithUnix]

That rings a couple of alarm bells to me, mainly because the claim is listed under the "datacenter" heading.

Firstly, AMD's currently data centre CPUs are piledriver-based, rather than one of the newer construction cores. If that's the generation the slides are talking about, then the IPC increase against Excavator/Steamroller/wherever we're at now will be somewhat lower - according to anandtech bench a 40% IPC increase over an FX-8350 still wouldn't catch the i5 4690k on single threaded benchmarks.

Secondly, datacenters tend to run on lower-power, lower-clocked processors. For the performance claims for Zen to be made specifically in that segment makes me wonder about target clock-rates for Zen: if it won't scale up to the 4GHz+ rates that we currently have in FX CPUs, it could end up rather underwhelming...

The key is how believable that 40% is in the first place. My worry is from the big figures that were bandied around for the dual instruction decoders in Steamroller, which in the gaming benchmarks that people actually care about only came to about 10% better performance than Piledriver.

Still, that 40% (which could be 35% rounded up as much as 44% rounded down) is supposed to be over Carrizo. Carrizo's Excavator is supposed to be about 10% faster than Steamroller, which is supposed to be around 10% faster than Piledriver. Compounding that lot up gets you 1.1x1.1x1.4 = 1.7 times faster than Piledriver, and 1.7 times faster than an FX 8370 gets you to around Haswell performance. That should be enough if AMD can give us a 16 thread FX chip with the FX L3 cache reinstated which turbos up to 4.5Ghz. If they give us a 4 core 8 thread chip that tops out at 3.5GHz and no L3, then we will be underwhelmed.

[watercooled]

a 4C Zen on 14nm would be really small, even with L3 we'd probably be looking at sub 50mm2. So I think the 8C plus L3 claim is more plausible, especially given the target audience.

But of course, clock speed is what we're all wondering about for the desktop side, and something AMD are keeping quiet on.

[abaxas]

Maybe AMD are finally targetting the mainstream?

People want thin laptops that will plays games, while not roasting their nuts. Hopefully AMD can provide the CPU/APU/whatever they call it, for this.

[scaryjim]

Maybe AMD are finally targetting the mainstream?

People want thin laptops that will plays games, while not roasting their nuts. Hopefully AMD can provide the CPU/APU/whatever they call it, for this.

FX 7600p + R7 260m laptop are already available, are relatively thin, run pretty cool, and are very capable gaming machines - at least on the low-end panels that get used in mainstream laptops

AMD have been targetting the mainstream for a couple of generations - that's why Steamroller and Excavator have made it into APUs but not CPUs. The Kaveri mobility APUs run cool and can easily handle 720p gaming: more if they're coupled with a discrete card. Producing good mainstream processors has never been AMDs problem: getting them into OEM machines has. The success of Zen is going to be as much about the number of design wins as the performance of the cores.