Richland ES now out in the wild.

[CTho9305]

OK I went back and was able to find a youtube video someone uploaded of the entire webcast...but I must admit I got maybe 3 minutes into it and my skin was crawling, I absolutely cannot stand listening to Rory speak.

It is something about his speech gait or enunciation or word choice, or something, but it makes my brain spasm. It can't take that much leadership all at once! Leadership overload.

Was there a particular timeframe in the webcast that I am looking for? I'm pretty sure I'd sooner burn my laptop than sit through 40+ minutes of that webcast. Seriously, is Rory like that all the time or was he just nervous/distracted/unfocused?

edit: here is the youtube vid I was watching, anyone brave enough to bask in the bright light of pure unadulterated Rory leadership can get them some starting just 10 seconds into the video (I am pretty sure I am technically supposed to flag this link as NSFW, it is truly that bad IMO)

I think part of the problem is that, when left to his own devices, he's really animated (he always seemed genuine to me), but somebody told him not to be. By the way, the first question (at 47 minutes 20 seconds) is pretty good. Somewhere in this Q&A Rory loses his composure and reverts to his normal, excited self for a moment, which is amusing to watch. You kinda have to feel sorry for the guy... trying to run a struggling company, and having someone pick at his mannerisms.

 

[zlatan]

I thought Facebook said it didn't care about proprietary architectures, because the change rate of technology is so high

In software. But we talking about the hardware side.

 

[SocketF]

Kaveri is 28nm bulk hkmg right?

Yes.

I thought the full-blown Steamrollers for server/workstation are meant for 22 nm... now which ES do you think is more realistic to come out of GloFo considering the date?

No, they were planned with 28nm, too. There was an official roadmap showing that:
http://www.abload.de/img/fileiaa00.jpg

22nm was canceled long ago at GF and TSMC. Their next node is 20nm.

 

[Idontcare]

I think part of the problem is that, when left to his own devices, he's really animated (he always seemed genuine to me), but somebody told him not to be. By the way, the first question (at 47 minutes 20 seconds) is pretty good. Somewhere in this Q&A Rory loses his composure and reverts to his normal, excited self for a moment, which is amusing to watch. You kinda have to feel sorry for the guy... trying to run a struggling company, and having someone pick at his mannerisms.

Interesting, that puts things into a different perspective for me, thanks for sharing that.

So you think his halting speech pattern is more out of an attempt to micro-manage his own excitement and personality because he's been guided to filter it a bit. I can see that, it would certainly be something he would tackle to his best abilities.

I went back and listened to the Q&A as you suggested and you are right, it was much more comprehensive and reasonable, also didn't grate on me like the opening part. One thing I noticed, must be a technical glitch, but the audio is out of sync with the video by quite a few seconds. That issue may have been what was giving me the mental block issues with Rory at the beginning of his opening talk. (maybe?)

What I got out of the Q&A is that it is far more likely that AMD will be licensing freedomfabric to other ARM designers versus AMD acquiring a boutique ARM designer (as Apple did with PA Semi) for inhouse ARM design. So maybe we'll see a Calxeda or an Nvidia licensing freedomfabric?

 

[CTho9305]

I went back and listened to the Q&A as you suggested and you are right, it was much more comprehensive and reasonable, also didn't grate on me like the opening part. One thing I noticed, must be a technical glitch, but the audio is out of sync with the video by quite a few seconds. That issue may have been what was giving me the mental block issues with Rory at the beginning of his opening talk. (maybe?)

The actual live stream was fine...it's just an issue with this guy's recording (and it is noticeably worse towards the end of the video than at the start). I can see why Rory's mannerisms would bug people, but they're the least of my concerns about AMD.

 

[wlee15]

Update

2012-11-20 19:18:44 < Skyler_> BugMaster: which was the test you wanted me to run on the kaveri(?) ?
2012-11-20 19:18:50 < Skyler_> something to do with XSAVEOPT
2012-11-20 19:19:09 < BugMaster> cpuid eax = 7 ecx = 1
2012-11-20 19:23:12 < BugMaster> wait
2012-11-20 19:24:40 < Skyler_> 28b030 28b034 28b038 28b03c
2012-11-20 19:24:50 < BugMaster> eax = 0dh (13) not 7
2012-11-20 19:25:19 < Skyler_> okay
2012-11-20 19:27:25 < Skyler_> nvm, I am horribly dumb and obviously not in my right mind this morning, I forgot to give printf its arguments
2012-11-20 19:27:32 < Skyler_> eax=0xd, ecx=0x1 gives 0 0 0 0
2012-11-20 19:27:34 < BugMaster> btw. what you posted looks like pointers
2012-11-20 19:28:13 < BugMaster> hm. all zeros. than that is probably no bdver3
2012-11-20 19:28:43 < Skyler_> are we sure?
2012-11-20 19:30:05 < BugMaster> not sure but according to http://gcc.gnu.org/git/?p=gcc.git;a...77aff08fc23d66b6eeb8a68c3d5c784f6ec6d07a#l545 that was one of the ways to find you if it run on bdver3
2012-11-20 19:30:21 < BugMaster> http://gcc.gnu.org/git/?p=gcc.git;a...77aff08fc23d66b6eeb8a68c3d5c784f6ec6d07a#l467
2012-11-20 21:08:17 < Mandarinka> regarding the AMD ES discussion... in their Hot Chips 2012 slides, AMD claims that Steamroller module will have increased L1 instruction cache size. Of course, it'S not guaranteed, but perhaps it could help to identify the core
2012-11-20 21:08:36 < Mandarinka> sadly the slide isn't specific.
2012-11-20 21:10:11 < Skyler_> the chip says
2012-11-20 21:10:14 < Skyler_> 4x16kb data cache, 4-way
2012-11-20 21:10:19 < Skyler_> 2x64kb instruction cache, 2-way
2012-11-20 21:10:23 < Skyler_> 2x2048kb L2 cache, 16-way
2012-11-20 21:10:28 < Skyler_> according to puc-z
2012-11-20 21:10:30 < Skyler_> *cpu-z
2012-11-20 21:10:58 < Mandarinka> 64kb, that sounds like Bulldozer/Piledriver, doesn't it.
2012-11-20 21:11:23 < Mandarinka> unless cpu-z doesn't read the actual sizes but instead guesses based on older chips - no idea how it poperates
2012-11-20 21:11:38 < Mandarinka> *operates, sorry
2012-11-20 21:12:14 < Skyler_> I'm guessin it's a bulldozer then... ugh
2012-11-20 21:13:26 < Skyler_> *guessing
2012-11-20 21:13:54 < Mandarinka> there was some leaked information that AMD postponed Kaveri and will instead only introduce a refreshed Trinity APU (but that was based on desktop roadmap). On this monday, some site reported that AMD cancelled both Steamroller and the Excavator arch that comes after it
2012-11-20 21:14:39 < Mandarinka> AMD denied that but what do we know...
2012-11-20 21:14:48 < Skyler_> semiaccurate was the one reporting that though
2012-11-20 21:14:57 < Mandarinka> yeah
2012-11-20 21:14:57 < Skyler_> they are known for being... very speculative
2012-11-20 21:15:02 < Skyler_> (and often extremely wrong)
2012-11-20 21:15:28 < Mandarinka> yes, but they did get some thing right too, so it can'T be discounted completely
2012-11-20 21:15:30 < kierank> how come amd were being so opaque to you about the chip
2012-11-20 21:15:44 < Skyler_> left hand doesn't know what the right hand's doing
2012-11-20 21:15:50 < Mandarinka> with AMD, the worst thing imaginable usually tends to happen
2012-11-20 21:16:40 < Mandarinka> the Trinity refresh was supposed to be called Richland though, I didn't expect them to put the Kaveri name on it (their roadmap from this summer clearly promised GCN + Steamroller + 28nm for Kaveri)
2012-11-20 21:18:32 < Mandarinka> so there are no meassurable performance improvements above Trinity/Piledriver either? /unless you prefer to keep that secret, which I'd respect/
2012-11-20 21:18:41 < Skyler_> I haven't done detailed benchmarks
2012-11-20 21:18:47 < Skyler_> and probably shouldn't publish any
2012-11-20 21:19:24 < Mandarinka> yeah, I think so too I was already surprised I learned what I saw in the log

Doesn't appear to be a Steamroller based so a little bit disappointing.

What I got out of the Q&A is that it is far more likely that AMD will be licensing freedomfabric to other ARM designers versus AMD acquiring a boutique ARM designer (as Apple did with PA Semi) for inhouse ARM design. So maybe we'll see a Calxeda or an Nvidia licensing freedomfabric?

AMD has stated that they won't license their fabric to other cpu vendors.
http://www.theinquirer.net/inquirer...cense-seamicros-freedom-fabric-to-cpu-vendors

I seem them using the gpu model where AMD designs and sells all of the CPU and also makes reference designs for the daughtercards and compute systems which the OEM may choose to use or adapt for their own designs.

 

[Ajay]

Wow, so Richland. What is even different in Richland?

-----------------------------------------------------
Samsung is apparently working on an 8-core ARMv8 design, and they have their own fab. Maybe there is something to this ARM 64b server market, no need for an 8-core consumer ARM CPU.

 

[nforce4max]

Interesting read, I hate to say it but I am still holding out some tiny bit of hope that maybe Steamroller does make it to market. I do remember reading an article somewhere of some of the changes they had planed from Piledriver. Mainly a few fixes and some new logic design that reduced the transistor count without sacrificing performance. The rest was about new design tools instead of traditional methods.

 

[Ajay]

Interesting read, I hate to say it but I am still holding out some tiny bit of hope that maybe Steamroller does make it to market. I do remember reading an article somewhere of some of the changes they had planed from Piledriver. Mainly a few fixes and some new logic design that reduced the transistor count without sacrificing performance. The rest was about new design tools instead of traditional methods.

They increased xtor densities. They didn't reduce xtor count AFAIK. AMD says Kaveri isn't cancelled, but everything else about AMD's current financial situation seems to say it has to be cancelled. We'll see.

 

[Idontcare]

.... but they're the least of my concerns about AMD.

Any chance of me convincing you to expand on your thoughts there?

 

[ShintaiDK]

Wow, so Richland. What is even different in Richland?

Just an slightly updated IGP. I assume its an electric respin too.

 

[Abwx]

Just an slightly updated IGP.

What is this "slightly" exactly ?....

Frequency .?...number of compute units ?..

 

[ShintaiDK]

What is this "slightly" exactly ?....

Frequency .?...number of compute units ?..

You cant do much since its the same processnode. I assume its just a change in GPU uarch to CGN since they keep talking about the HSA (enhancement) fairy tale. So I wouldnt put hopes up for anything.

 

[Abwx]

Going from VLIWX to GCN is GPU wise as using another core CPU wise.

It s somewhat a little more than just a slight update.

 

[Idontcare]

Interesting read, I hate to say it but I am still holding out some tiny bit of hope that maybe Steamroller does make it to market. I do remember reading an article somewhere of some of the changes they had planed from Piledriver. Mainly a few fixes and some new logic design that reduced the transistor count without sacrificing performance. The rest was about new design tools instead of traditional methods.

Hey, even cedar mill (65nm netburst) made it to market despite the presence of Core 2 Duo. I even bought one for $54 as a throw-away disposable test chip for veryifying my striker mobo (a ridiculously priced $450 nightmare) had the right bios before plugging on the latest QX6700...so hey served a purpose

The concerns I have with steamroller's future is that it has reportedly not even taped out yet. That is a big red flag in terms of project management and the prospects of upper management continuing to fund the project through complete. The tapout milestone is only significant because that is the milestone that precedes the first-silicon milestone. Again first silicon is only relevant because that precedes the debug and verification phase...which itself takes about a year to fully hash out if more steppings and respins are necessary (and they almost always are).

Management pulling the plug now on steamroller prior to tapeout would be the right time if they wanted to cut short the losses that they are going to incur if they commit themselves to resourcing another year of steamroller development through the costly verification and debug stage.

 

[Idontcare]

Going from VLIWX to GCN is GPU wise as using another core CPU wise.

It s somewhat a little more than just a slight update.

And in a world where low-cost entry level budget discrete gpu cards didn't exist, this would mean something substantial to AMD's bottom line.

Aren't AMD's financials telling enough that bundling bottom-tier graphics capabilities with bottom-tier cpu capabilities just isn't going to cut it? So what if you raise the bottom end when it comes to graphics capabilities?

Intel is proof that if you want to make money selling CPUs then you better darn well focus on improving the CPU, the graphics part of your APU can be second best (or even worse) and the financial ramifications of that to the company are pretty much irrelevant.

 

[ShintaiDK]

Going from VLIWX to GCN is GPU wise as using another core CPU wise.

It s somewhat a little more than just a slight update.

Good luck with that hope

Llano to Trinity didnt offer that much. Was it 16% average, even with 33% higher GPU clock? Plus the IGP is also bandwidth starved. And there is no DDR4 with Richland.

GPU progression is very heavily dependent on a shrink or a diesize increase. Looking at Trinity you see the same. Diesizes increases, even tho the CPU is changed to a 2C/4T instead of 4C/4T.

 

[Abwx]

Intel is proof that if you want to make money selling CPUs then you better darn well focus on improving the CPU, the graphics part of your APU can be second best (or even worse) and the financial ramifications of that to the company are pretty much irrelevant.

Rather this is the proof that the the dominant player set the momentum
even if it s at the expense of the consumer.

Basicaly , whoever had several laptops the last decade knows that
Intel s HD2000 to 4000 will be obsolete long before Trinity s CPU part will be also obsolete.

Of course this suit Intel since their high volumes lives on fast
obsolescence of the products they are selling.

They know that their CPU has theoricaly a very long life cycle
but they know even better that its GPU part wont allow for the whole
thing to last as long as expected by the gullible consumer that banks
all in a "fast CPU".

 

[Vesku]

And in a world where low-cost entry level budget discrete gpu cards didn't exist, this would mean something substantial to AMD's bottom line.

Aren't AMD's financials telling enough that bundling bottom-tier graphics capabilities with bottom-tier cpu capabilities just isn't going to cut it? So what if you raise the bottom end when it comes to graphics capabilities?

Intel is proof that if you want to make money selling CPUs then you better darn well focus on improving the CPU, the graphics part of your APU can be second best (or even worse) and the financial ramifications of that to the company are pretty much irrelevant.

I think the Intel lesson is more about having a commanding position with your vendors.

 

[AnandThenMan]

I think the Intel lesson is more about having a commanding position with your vendors.

This.

 

[Ajay]

I think the Intel lesson is more about having a commanding position with your vendors.

Good point. That's the way it was with Microsoft (and still is, though Win8 might be changing that for the consumer market).

 

[CTho9305]

Any chance of me convincing you to expand on your thoughts there?

Sorry, not until far enough in the future that nobody is interested any more.

 

[os2wiz]

The next APU will be a Piledriver refresh, unlikely that someone outside from AMD get a Steamroller based APU this year. I guess he didn't realize AMD delayed Kaveri into 2014, so he believes he get one while in reality it's just a Piledriver based APU.

Absolutely false!AMD has reaffirmed Kaveri release in 2nd half of 2013. I do not know where you trolls get your disinformation from.

 

[wlee15]

Absolutely false!AMD has reaffirmed Kaveri release in 2nd half of 2013. I do not know where you trolls get your disinformation from.

Source?

 

[Olikan]

Good luck with that hope

Llano to Trinity didnt offer that much. Was it 16% average, even with 33% higher GPU clock? Plus the IGP is also bandwidth starved. And there is no DDR4 with Richland.

well, from Llano to trinity, amd just made the IGP even more bandwidth starved...what really bumped the performance was the improvments in bandwidth sharing

from vliw 4 to GCN, the increase will be bigger...
at the same clocks/shader/bandwidth: http://www.inpai.com.cn/doc/hard/168475_30.htm

add another improvments in IMC, then it is a decent bump
(but probably will get owned to GT3)