AMD manager speaks about Bulldozer, admits failure

[Magic Carpet]

You can't expect AMD Managers to know more than AMD fans on forums. :awe:

I laughed hard :biggrin:

 

[Abwx]

You can't expect AMD Managers to know more than AMD fans on forums. :awe:

And they will surely know even much less than anti AMD
junk spreaders....

 

[USER8000]

Regarding,Andrew Feldman,he was CEO of SeaMicro and considering that they use low power CPUs,and AMD is now focussed towards using low power X86 and future ARM CPUs,of course they are going to make sure that they big that up,but at the same time appropriating any blame for failings towards the previous management,ie, why the company was not working earlier on them before.

Anyway,companies do admit when they have made mistakes.

Intel did with Larabee:

http://www.pcper.com/news/Graphics-...bee-was-impractical-and-power-hungry?nid=9275

Both,Larabee and Knights Ferry never entered production despite what Intel claimed at the time.

Yet an evolution of it continues with Knights Corner which did enter production and seems to get over many of the initial problems:

https://en.wikipedia.org/wiki/Intel_MIC

Intel had problems with the Pentium 4,yet the Northwood B was a decent desktop CPU against the Athlon XP CPUs of its day.

Whether or not AMD will be able to evolve its current desktop CPU architecture to get over its current issues is one thing,but I suppose it as least makes things interesting,LOL.

 

[skipsneeky2]

Its that time again! :awe:

LOL,i almost pissed myself laughing at this.:biggrin:

 

[Ajay]

http://www.pcper.com/news/Graphics-Cards/Intel-admits-Larabee-was-impractical-and-power-hungry?nid=9275[/URL]

Both,Larabee and Knights Ferry never entered production despite what Intel claimed at the time.

Yet an evolution of it continues with Knights Corner which did enter production and seems to get over many of the initial problems:

https://en.wikipedia.org/wiki/Intel_MIC

Intel had problems with the Pentium 4,yet the Northwood B was a decent desktop CPU against the Athlon XP CPUs of its day.

Whether or not AMD will be able to evolve its current desktop CPU architecture to get over its current issues is one thing,but I suppose it as least makes things interesting,LOL.

Larabee was re-targeted from GFX to Compute with the Xeon Phi, that was a smart pivot on Intel's part and a good re-use of engineering resources.

I would say that Northwood A was an unmitigated failure. Intel did it's best of improving the design and ultimately got out of trouble with the 'Core' architecture. If AMD wasn't supply constrained, Intel would have been in a terrible situation by 2005.

AMD could build up a new uarch to put themselves in a better position on big core x86 if they had the $$s to burn, but they don't.

There are plenty of tools in AMD's toolbox (including Intel x-licensed IP) to make larger jumps than Intel was able to do with the P4, but they are constrained by R&D dollars and GFL. XV almost certainly will need a 20nm process to make a large jump in performance, but when will that be available? It would probably make sense if 20nm is available in 1H2015, but given GFL's track record, sometime in 2016 is much more likely. If the later is the case, say bye bye to AMD's big core lineup.

 

[BallaTheFeared]

LOL,i almost pissed myself laughing at this.:biggrin:

That's not even the good one!

 

[skipsneeky2]

That's not even the good one!

LOL,that is a better one and wow so funny and true.:awe:

I haven't paid much attention to the haswell release cause the random stuff i hear sounds like a snore fest while the only decent thing amd has made has been their fm1 and fm2 chips.

Love intel and their processors but holy hell,their processing power is so awesome and refreshes so lackluster i can truck my i5 2500 non k all the way to broadwell unless BF4 rapes my chip.

 

[Lepton87]

LOL,that is a better one and wow so funny and true.:awe:

I haven't paid much attention to the haswell release cause the random stuff i hear sounds like a snore fest while the only decent thing amd has made has been their fm1 and fm2 chips.

I think the only decent things that AMD recently released were GCN cards, by they were recently trounced by NV7XX series, at least in the 400$ space and up.

 

[Arkaign]

It will be like the Pentium 4 - in a few years time, those who supported it will be too embarrassed to admit it.

I just love the fact that, to fix BD, they end up undoing some of the changes, making an SR core more like a Phenom II core.

It's not really comparable to Pentium 4 though. Some people have extremely inaccurate memories of that entire lineup.

When P4 launched, it was an overpriced mess on a weak Willamette core. Even then, it still won a number of benches, though it was a fairly terrible choice. Being tied to RDRAM was the real insult, prices were sky-high and it didn't offer enough over even SDRAM to be worthwhile.

However, it didn't take too long to get to Socket 478, 845 chipset, and DDR support along with Northwood. From that point on, P4 was either tied for first, or led outright (check P4 3.2Ghz vs. 3200+ AXP for example). They also overclocked extremely well, letting you run a much cheaper cheap such as a 1.6A and taking it to very high levels of performance for cheap. AMD's Athlon XP was also excellent, but ran out of steam in the end not being able to credibly fight off the 800FSB Hyper-Threading P4 Northwoods. Thus, we got AMD64, which immediately won, right? Actually no, the first 3000+, 3200+ chips actually split the benches with P4s. And socket 754 was a quick dead end just like socket 423 was. The big win that everyone remembers was actually Socket 939 and later AM2, and when Prescott fell flat on it's face with the disappointing lack of gains from .13 AMD simply ran away with the performance lead, which lasted all the way until Conroe (unless you were some weirdo running a mobile Intel chip on a desktop, which was expensive but silly fast).

Even then, things weren't perfect. AMD went from being a value hero to pricing their chips like they were made of gold. The absolute cheapest dual-core they had was $300, and they went up massively from there, with huge leaps at every SKU all the way up to $1k+. It was so bad that I actually ended up choosing Pentium D 805 for some customer builds, simply because saving ~$200 on the CPU let a much better video card fit into their budget.
It would be like today if Intel's cheapest quad was $300+, and AMD had something 80% as fast, also quad, for $100. Overclocked, the PD-805 was competitive with the AXP 3800 (stock of course), so it actually was a respectable choice if you didn't have the $ to go with the superior AMD option. Prices were bad back then though, I remember paying over $200 for 2GB of DDR2 memory.

Anyway, P4 is not the universal failure some remember it as, it's mostly the poor beginning and poor ending that people remember in hyperbolic terms. For the meat of its life, it was the top seller, with top performance to match. During that era, if you had a P4 2.4B @ 3Ghz+, or a AXP Mobile @ 2.4Ghz+, you were pretty much gold either way.

 

[Idontcare]

Even then, things weren't perfect. AMD went from being a value hero to pricing their chips like they were made of gold. The absolute cheapest dual-core they had was $300, and they went up massively from there, with huge leaps at every SKU all the way up to $1k+.

I so badly wanted an X2, at the time having owned AMD chips since the K6-2 launched. But it was stupid high in pricing, so I had no choice but to buy a Northwood and OC it. AMD was not the enthusiasts friend at that time, they sucked every dollar they could out of the channel.

 

[Arkaign]

I so badly wanted an X2, at the time having owned AMD chips since the K6-2 launched. But it was stupid high in pricing, so I had no choice but to buy a Northwood and OC it. AMD was not the enthusiasts friend at that time, they sucked every dollar they could out of the channel.

Most definitely. Their products were extremely good, though their pricing was brutal, much like Intel's pricing at their worst (P3s for example). It was laughable that they held that pricing even a little after Core 2 Duo launched.

 

[Ajay]

I so badly wanted an X2, at the time having owned AMD chips since the K6-2 launched. But it was stupid high in pricing, so I had no choice but to buy a Northwood and OC it. AMD was not the enthusiasts friend at that time, they sucked every dollar they could out of the channel.

I stuck with Intel until the K7 was released. I bought about three K7s (original, TBird and Barton) and they were a phenomenal value. When the K8 came out I decided to spend the money on a custom LC loop overclocked my Barton to within an inch of its life and stuck with it till Intel's first Core CPUs came out because I thought AMD overpriced its CPUs. Especially considering inflation, prices on CPUs are great nowadays; except Intel's 'X' CPUs, but even then you can get a non X enthusiast CPU for ~ 1/2 the money.

 

[ShintaiDK]

I so badly wanted an X2, at the time having owned AMD chips since the K6-2 launched. But it was stupid high in pricing, so I had no choice but to buy a Northwood and OC it. AMD was not the enthusiasts friend at that time, they sucked every dollar they could out of the channel.

I am so happy we got the price turnaround in july 2006. Performance segment chips went from 600-850$ down to 200-300$.

 

[aigomorla]

dude... bulldozer was fail.

If AMD didnt still admit to it, id say they were a very terrible company.
A good company admits there mistakes... pays for them.. and then fixes it.
A bad company wont admit there mistake... tell there customers they are wrong.. and manipulates information to there favor as best as possible.

it is nice to see AMD starting the first path on a correct company.

 

[Vesku]

I think the only decent things that AMD recently released were GCN cards, by they were recently trounced by NV7XX series, at least in the 400$ space and up.

$650 space and up actually.

 

[Rvenger]

I so badly wanted an X2, at the time having owned AMD chips since the K6-2 launched. But it was stupid high in pricing, so I had no choice but to buy a Northwood and OC it. AMD was not the enthusiasts friend at that time, they sucked every dollar they could out of the channel.

You should of bought a dual core opteron and overclocked it and saved 40-50%. For some reason the x2s were ridiculously priced but the opterons carried a much lower clock speed but were able to catch up by increasing the FSB... They were selling those OEM cpus dirt cheap!

 

[mrmt]

Anyway,companies do admit when they have made mistakes.

Intel did with Larabee:

http://www.pcper.com/news/Graphics-...bee-was-impractical-and-power-hungry?nid=9275

Both,Larabee and Knights Ferry never entered production despite what Intel claimed at the time.

Interesting rumor about Xeon Phi:

http://www.hpcwire.com/hpcwire/2013...d-shattering_supercomputer_.html?featured=top

We do not generally report on rumors.

But we are making one exception right now.

With so many sources, all of them credible and from across a rather wide swath of the HPC spectrum, we couldn’t ignore the unconfirmed word of what might be a rather stunning victory for China in the race to produce the fastest supercomputer on the planet.

Again, while unconfirmed, from what we have been led to believe from our sets of sources in both industry and academic circles, China may have something on the order of a 50 petaflop system based on the MIC architecture waiting in the wings for June’s Top 500 unveiling. Sources have claimed that Top 500 brass have already been deployed to validate the results and have emerged with the verification.

Ed:

More here:

*UPDATE* 11:41 a.m. Eastern -- Confirmed from three credible sources--there are 48,000 MICs on the system

*UPDATE* 2:37 p.m. Eastern -- From Dr. Satoshi Matsuoka "1PB memory & 12.4PB Storage. 576 port x 13 Core SW, NW derived from Tianhe-1A & improved (still IB QDRx2 equiv.)"


(...)

The reported performance (we have four highly credible sources confirming) is between 53-55 peak and between 27-29 LINPACK sustained performance. This is actually better than we were led to believe yesterday when it felt a little wrong to make the "50 petaflop" claim, despite our best sources telling us it was so.

 

[krumme]

It really was an unmitigated disaster:

A, you boast of having this small module while having a huge overall die size.

B, you saved a lot of module die size by sharing resources, unfortunately it translated into a little $ saved, a not insignificant performance penalty, and a big loss on ASP and market share.

C, you burden your customers and chipset group with the cost of supporting 2 sockets.

D, overall thinking is minimizing die size/core even though the WSA makes incremental die basically free.

Reading between the lines on these forums, AMD is actively addressing these issues. Its unlikely to help the overall market grow again but it should stabilize share and raise ASP's.

I dont know why BD ever hit the street. PD came already 1 year later and was so wastly improved for that timespan, it looks like BD was a broken product. I dont know if you could call a nearly broken unfinished product a failure, but bringing it to market, is just so difficult to understand and that is with a nice word a unmitigated disaster. What is the reason for it?

BD is so bad its difficult to know even the potential of it, its to unfinished to be called a failure, even though PD points clearly in that direction.

SR, Excavator whatever might by slight possibility turn out to be very strong performers, especially for the desktop and client side. But for what purpose? I think its safe to say Intel hold the perf/power advantage and thereby the still profitable mid and highend servermarket. That opportunity seems closed to amd. And then it doesnt matter if they come with fx. a big fat bad ass worldclass fpu, because there is no customers for it.

What they needed more, was just speeding up their small core strategy. And have done that 2½ years when it was obvious bobcat was going to sell like hot cakes. Jaguar is a nice improvent, but they just need to improve constantly here. This is where the world class competence should go.

 

[USER8000]

Anyway,companies do admit when they have made mistakes.

Intel did with Larabee:

http://www.pcper.com/news/Graphics-...bee-was-impractical-and-power-hungry?nid=9275

Both,Larabee and Knights Ferry never entered production despite what Intel claimed at the time.

Yet an evolution of it continues with Knights Corner which did enter production and seems to get over many of the initial problems:

https://en.wikipedia.org/wiki/Intel_MIC

Interesting rumor about Xeon Phi:

http://www.hpcwire.com/hpcwire/2013...d-shattering_supercomputer_.html?featured=top



Ed:

More here:

You missed a bit!!

Larrabee and Knights Ferry never entered full production despite statements at the time. But out of the ashes of those projects came Knights Corner,and they managed to iron out,ahem,many kinks. So a project which might have been fail at the time(even by what Intel said themselves),did end up being useful a few years later after they kept at it.

Larabee was re-targeted from GFX to Compute with the Xeon Phi, that was a smart pivot on Intel's part and a good re-use of engineering resources.

I would say that Northwood A was an unmitigated failure. Intel did it's best of improving the design and ultimately got out of trouble with the 'Core' architecture. If AMD wasn't supply constrained, Intel would have been in a terrible situation by 2005.

AMD could build up a new uarch to put themselves in a better position on big core x86 if they had the $$s to burn, but they don't.

There are plenty of tools in AMD's toolbox (including Intel x-licensed IP) to make larger jumps than Intel was able to do with the P4, but they are constrained by R&D dollars and GFL. XV almost certainly will need a 20nm process to make a large jump in performance, but when will that be available? It would probably make sense if 20nm is available in 1H2015, but given GFL's track record, sometime in 2016 is much more likely. If the later is the case, say bye bye to AMD's big core lineup.

It was more the case how Intel publicly admitted their failings with Larabee,but in the end the project was not a total failure.

Northwood B was actually not too bad at all with regards to performance and what was available on the market too. I think Intel misstepping with Prescott did remove the momentum with the P4 they had though after that.

Regarding AMD,at least they seem to be trying to keep to a yearly cadence in improving their cores,but I suspect they will be concentrating more and more on the smaller cores,which are more easily portable between different processes. It seems the SOCs in the PS4 and XBox One are made at GF and TSMC.

Even the chap who commented on the BD problems,was the CEO of Seamicro.

 

[HurleyBird]

That's not even the good one!

Oh my God that's funny! Hehe.

 

[Ajay]

Interesting rumor about Xeon Phi:

http://www.hpcwire.com/hpcwire/2013...d-shattering_supercomputer_.html?featured=top

Ed:

More here:

I just hope that they don't use the thing to steal even more US (or any other country's) IP more effectively.

 

[mrmt]

Larrabee and Knights Ferry never entered full production despite statements at the time. But out of the ashes of those projects came Knights Corner,and they managed to iron out,ahem,many kinks. So a project which might have been fail at the time(even by what Intel said themselves),did end up being useful a few years later after they kept at it.

I think it is important to distinguish between failures here.

One thing is trying to extend the scope of a given project beyond its suitability, and this is where Larrabee failed. MIC is nice to crunch numbers, but isn't suitable to graphics, but even so Intel pushed the thing as a GPU, and this is what gave the project a bad fame. Had Intel not marketed Larrabee as a GPU but instead as a co-processor, nobody would be calling it a failure. (OTOH, if it wasn't internally market as a GPU it might never got an approval signature... Internal politics, you never know)

Plus Larrabee was never core business for Intel, the company wouldn't live or die by its success or failure, it was a project they could incur in bigger risks.

The other thing is AMD's preferred unmitigated failure. It doesn't excel in things it was supposed to excel, it failed miserably in the very areas it was supposed to perform and, as it was *the* AMD core business, the company paid a heavy price for that failure.

 

[USER8000]

I just hope that they don't use the thing to steal even more US (or any other country's) IP more effectively.

It does make me think how long until we start getting more indigenous efforts from the Chinese too. Such computers would have no export restrictions too,and we are giving them all the tools to do so.

 

[Lepton87]

$650 space and up actually.

Isn't GTX770 practically just as fast as 7970GHz and cheaper? It uses less power to boot. The only saving grace of Radeon is overclock-ability but that is pure luck.

 

[BUnit1701]

Isn't GTX770 practically just as fast as 7970GHz and cheaper? It uses less power to boot. The only saving grace of Radeon is overclock-ability but that is pure luck.

GTX770 = GTX680 Ghz Edition. Nothing new here.