The AMD Execution Thread [2007 - 2017]

[psurge]

I worked at Bing for a year at one point, you'd be surprised how low CPU utilization was even on the compute clusters, you were almost always constrained by how efficiently you could move the data, rather than any processing you might be doing.

Wouldn't going with a smaller number of fast cores to reach a given level of compute performance result in less data having to go over the network than using a larger number of slower cores? Or are you referring to moving data between disks and cores?

 

[3dilettante]

i dont see how this is a problem for a BBOSN full of ARM cores. First your network latency is going to be the very vast majority of a web server latency. Second why is the ARM core going to be high latency, assuming a suitable memory sub system? Your going to have far more execution resources per clock on a 16 core arm vs a comparably priced X86 server.

Has AMD indicated it is going to go for 16 cores? ARM's platform keeps the option open, but AMD isn't obligated to push things that far and going too far beyond the core count of the majority of small A-57 chips adds engineering cost, risk, and possibly time to market penalties. I'm not as sanguine about AMD's committment to the effort.

thats a straw man and nothing more.

It's an extreme exaggeration for effect, yes.
Various benchmarks and services do have response time requirements, and AMD has listed more usage cases in its slides than these.

Let alone a 16 core A-57 @ 2-3ghz with 16/32/64 gigs of ram. You have done nothing to justify the notion that application latency is going to suffer with arm.

I've already stated my misgivings on core count for AMD and the last I saw was up to 2.5 GHz at 20nm.
The performance capability at a core level is a regression in absolute terms of about 6 years, and potentially a regression compared to existing Seamicro products that use x86 chips.

i dont think your understanding the workloads or the business drivers of this market. big scalable coherent compute is far more niche then what a BBOSN is targeting. i have never seen a workload that can scale across 16 threads but not across boxes. It why 1 and 2 socket boxes massively out strip 4+ in sales.

Memory capacity available per instance goes up with more sockets that remain coherent, and the expansion in 1-2 sockets coincides with the rise of very performant high-count multicore chips.
Has AMD indicated it will have core counts that high, noting that those cores are 1/2-1/3 as powerful?

 

[jimbo75]

We won't see Kaveri until 2014 I fear.

http://www.donanimhaber.com/islemci...-ailesi-icin-2013te-guncelleme-gorunmuyor.htm

 

[function]

I thought that Richland was supposed to be Steamroller based? And that Kabini was supposed to be early 2013?

So AMD will be pushing 40nm Brazos as their Windows 8 tablet platform until late 2013? Hahahaha .... sigh.

I really hope this is fake.

 

[Zaphod]

What in that slide, if real, says it has to be late 2013? Assuming GF, they announced tapeouts on 28nm in July and say they'll ship for revenue in "early 2013". AMD could well (they'd better) have these ready for products that are slated to start shipping this summer.

 

[function]

What in that slide, if real, says it has to be late 2013?

The little bit in the very bottom right hand corner of the slide.

 

[Zaphod]

We already have working silicon for many of our new 2013 products in house, including our next-generation 28nm Kabini APU, which is the successor to our highly successful Brazos platform and our first true SoC design. We are making good progress with the bring up of Kabini, which remains on track to launch in the first half of next year.

AMD Q3 results conference call @ 8m 55s.

 

[function]

That's a relief then. I'm waiting on Jaguar for a Windows 8 tablet and I don't want to be waiting long.

x86 tablets are one area where AMD can still be highly competitive with Intel (Atom is very slow and the IVB platform while fast is too hot, fat, heavy and expensive for mainstream tablets IMO). Every day past the Win8 launch that they don't have a highly competitive tablet part out is an opportunity lost.

 

[jimbo75]

Kabini being 30 months after Brazos is still a bad fucking joke tbh.

Some of AMD's recent moves are downright baffling. Papermaster was talking about Steamroller 2 months ago at Hot Chips. Now it's just gone? It would make sense to assume that most of those 15% being laid off are in server/high end desktop but they were quick to say that they were staying with x86 in server after the ARM announcement. So what are they doing if it's not Steamroller?

None of this makes any kind of sense.

 

[3dilettante]

The word "Stable" as a descriptor for the FX series is better PR than "Stagnant", but it is also the word used for AMD's business class processors, which are basically SKUs it guarantees to produce for a year or so for corporate OEM designs.
Stretching Vishera that long, and calling it stable, makes it sound like AMD is keeping some obligations fulfilled.

The APU segment isn't called "Stable", apparently by virtue of the GPU half only. The x86 half is "Stable".
edit: My mistake, AMD already called Trinity's GPU Radeon 2.0. AMD's entire desktop space is "Stable".


The thing is that I'm not sure what place Steamroller would have if Excavator is not also delayed.
2014 was at least initially going to bring in DDR4 and a different socket with Excavator, and that core is using the same libraries as the GPU half to improve density at the expense of clock speed.
Perhaps aligning the x86 core with the GPU design methodology would fix some of the perennial problems AMD has getting both the CPU and GPU manufacturable on the same die.

It does look like AMD is going through the motions with its large core development. Potentially, the projected numbers for Haswell is making AMD afraid it cannot survive the cash burn of massaging Steamroller into shape for 2013 given the reduced expectations for any kind of sales for AMD It may be sunk for fear of putting in a lot of money and getting a repeat of the Llano writedown when the next APU Osbornes Trinity.
Heaven help them if it's just a matter of moving Steamroller into Excavator's place or one or both being cancelled.

 

[Alexko]

Charlie mentioned something like this back in August: http://semiaccurate.com/2012/08/06/amd-to-update-trinity-to-trinity-2-0/

AMD is planning a Trinity refresh in 1H/2013, and Kaveri will be a tad later than most expect.

So maybe it's not that bad, and Kaveri is only delayed a quarter or so. One can only hope.

 

[Kaotik]

I thought that Richland was supposed to be Steamroller based? And that Kabini was supposed to be early 2013?

So AMD will be pushing 40nm Brazos as their Windows 8 tablet platform until late 2013? Hahahaha .... sigh.

I really hope this is fake.

Kaveri is Steamroller, Richland is apparently Trinity with new chipset (Bolton instead of Hudson)

 

[jimbo75]

I mentioned the idea before - more through hope than expectation - but if Steamroller is cancelled in favour of a shoot-for-the-moon at 14nm, then I can at least begin to understand some of this.

Let's be quite frank - AMD isn't even in the position where being cautious helps any more. They need to be bold - extremely bold. It's quite possible that Kaveri showed miserable gains on gpu as well due to bandwidth constraints, so if that's been put back in favour of AMD attempting to wipe the slate clean at 20/14nm then it might not be a total disaster.

Think about it - Trinity/Piledriver is actually pretty good but they've opened up at lower cost than Llano did. Why continue this losing streak with Kaveri/Steamroller when they can just get away with speed bumps and minor tweaks costing a lot less R&D? Chances are Kaveri or this Richland/Piledriver would end up at the same ~$100 mark anyway, with similar die sizes.

I know people get annoyed at the cancellations, but in many ways they really are just throwing good money after bad anyway. I just hope the decisions were made in order to make a big push at 14nm, but I don't really believe it.

 

[Raqia]

I agree, AMD can address the consumer markets more than adequately by selling Trinities and Brazos even well into next year. I'm not excited as an enthusiast, but it's a sound business strategy to pull back R&D when their gains would probably be minimal but disruptive (due to socket changes) especially when existing products are doing fine. It's being quite bold on the server side w/ its ARM strategy, but it will take some effort to build up a reputable software ecosystem.

 

[UniversalTruth]

Posted, discussed?!

AMD Slashes Prices of 17 Socket FM1 and AM3 Processors

http://www.techpowerup.com/175045/AMD-Slashes-Prices-of-17-Socket-FM1-and-AM3-Processors.html

Quite noticeable price reduction.

 

[no-X]

Kaveri is Steamroller, Richland is apparently Trinity with new chipset (Bolton instead of Hudson)

Richland is not a platform, but an APU. Probably a 28nm "shrink" of Trinity.

 

[AlexV]

Richland is not a platform, but an APU. Probably a 28nm "shrink" of Trinity.

Probably just Trinity without any shrinkage, and some magic hoopla bombad ninja bdver2.Pi features that'll get the forums riled up.

 

[3dilettante]

Perhaps the overdue introduction of PCIe 3.0?
It seems odd that AMD had so much trouble introducing that interface to the APU side, since it's not like they don't have chips with compatible controllers.
Maybe a revision that enables it?

 

[Osamar]

More "good" news, this time for Linux users.

http://www.h-online.com/open/news/item/AMD-dismisses-numerous-open-source-developers-1745131.html

 

[ninelven]

Stock nearing a 5 year low... now would be the time to buy...

Something I had not really considered before, but it would be very interesting if Google bought them.