John Fruehe (JF-AMD) no longer with AMD

[pelov]

Rory seems to be a fan of putting up the white flag and moving onto something else that hasn't sunk... yet. I really hope he has a trick up his sleeve and will show it in the next year or so. I am getting impatient.

He's probably the only CEO AMD has ever had that actually cares about how much money they make. He won't just push something through if there's no demand for it. That horrible mistake with the oversupply of Llano, while unbelievably stupid, also had a silver lining. Both Dirk and Ruiz would have dropped the price of the chips even though they aren't EOL, particularly as the HD4000 has closed the gap and the supply was too high, but Read actually kept the prices the same... who does that? He's clearly more worried about AMD's bottom line than the retailers and OEM's, who will be buying Trinity chips soon anyway. It's a page right out of Otellini's book: don't drop the price of the older gen stuff just because you've got something new out soon, and if you do you should do it by small decreases over a longer period of time.

I think he's a perfect fit for AMD, personally. The man has an altar dedicated to the dollar bill and an influx of funds is exactly what AMD need.

 

[thilanliyan]

Yes :sneaky: that is clearly what it must have happened. Could not have had anything to do with faildozer, the timing was just a coincidence.

You could be right. It is just something I have heard, which seemed to make sense considering the growing size of the mobile market.

 

[AtenRa]

28% more performance with 33% more cores and 100 extra mhz is FAIL for a new arch.

There is no other way to state that....sorry.

So let me get this straight,

You have a new product that is 28% faster than your last generation product but it is FAIL ???

Let me remind you that SandyBridge (2600K) was less than 28% faster in multithreaded apps than last gen Core i7 950 and yet i dont remember anyone saying that it was a FAIL

http://www.anandtech.com/show/4083/...core-i7-2600k-i5-2500k-core-i3-2100-tested/17

Cinebench 10 MT, 2600K (3.4GHz) is ~22% faster than Core i7 950(3.06GHz) while having 350MHz advantage and only 12% faster than Core i7 975 with 100MHz advantage.

Povray

Core i7 2600K (3.4GHz) is ~22% faster than Core i7 950(3.06GHz) while having 350MHz advantage and only ~11% faster than Core i7 975 with 100MHz advantage.

7 zip

Where is the Core i7 950 in this chart ??? Core i7 975 is faster than Core i7 2600K

x264

Core i7 2600K (3.4GHz) is ~19% faster than Core i7 950(3.06GHz) while having 350MHz advantage and only ~10% faster than Core i7 975 with 100MHz advantage.

 

[Hypertag]

So let me get this straight,

You have a new product that is 28% faster than your last generation product but it is FAIL ???

If you say bulldozer is 28% faster, then it must be true! Unless of course it isn't

 

[AtenRa]

I doubt the headcount losses at AMD in-and-around the release of Bulldozer came as a result of bulldozer performing to internal expectations.

I have no idea what bulldozer was targeted to perform at, but the fallout stemming from its actual performance certainly lends credence to the conclusion that it did in fact underperform.

Obviously John Fruehe was internally guided to expect IPC to not decrease, and yet IPC decreased, I would call that under-performing.

It underperformed internal expectations, and it underperformed external expectations that were set as a matter of corporate policy/agenda/marketing.

How much more clearly could the case be made that bulldozer underperformed and failed to reach its internal objectives?

IDC you contradicting your self, you say that we have no idea about the targeted performance and then you say that it is clear that BD underperformed and failed to reach the internal objectives.

But i though that we didnt know the internal objectives, then how it is clear that it underperformed ???

 

[AtenRa]

That doesn't put the Bulldozer into a better light. Per thread, Bulldozer is slower.

Really ??? they introduced a new product (Opteron 6200 series) that is 28% faster in Integer and 16% faster in FP and this doesn't put the new product into a better light ?? See above for Sandybridge.

What this shows is that AMD, for all that time and effort they put into Bulldozer, would have been better off making a dual-eight core K10h on their new 32 nm process, and it would have been faster than their K15h effort.

Yes we know that, they knew that but they where also designing the Bulldozer with Fusion in mind. They wanted to have a single architecture design for both Server and Fusion for the next 4-5 years.

 

[AtenRa]

The man has an altar dedicated to the dollar bill

LoL ehehehe

 

[AtenRa]

If you say bulldozer is 28% faster, then it must be true! Unless of course it isn't

I have clearly stated that the 28% performance was for Integer in SPECint-2006.

I have never said that BD was 28% faster in desktop.

next time read more carefully

 

[pelov]

IDC you contradicting your self, you say that we have no idea about the targeted performance and then you say that it is clear that BD underperformed and failed to reach the internal objectives.

But i though that we didnt know the internal objectives, then how it is clear that it underperformed ???

He isn't, read my post.

They were targeting +25% over the i7 920 which would have put it within SB territory. AMD was also gunning for mid 4ghz clock speeds and IPC holding steady at Deneb/Thuban levels but they missed both goals. There's no doubt that it underperformed and I'm sure that was partly the reason for some of the delays.

http://www.realworldtech.com/bulldozer/

the single threaded performance should actually be higher than the previous generation Magny-Cours and comparable to current Intel designs.

They fell short.

Bulldozer is a high frequency optimized CPU, a so called speed demon. This approach has fallen out of popularity in the x86 world, due to Intel’s misadventures with the Pentium 4. In all fairness though, many of the Pentium 4’s problems were unrelated to high clockspeed and more closely tied to the actual microarchitecture. In the high-end server world, IBM has successfully pursued high clock speeds with the POWER7. So a speed demon approach can work out successfully. Bulldozer has a fairly lengthy pipeline, to minimize the gate delays per stage. AMD was unwilling to share any specifics on gate delays, although some discussions at comp.arch suggest a target of ~17 gate delays vs. ~23 for Istanbul. To tolerate the increased latencies necessary for a high frequency target and to efficiently share resources between cores, Bulldozer introduces decoupling queues between most major stages in the pipeline.

They didn't increase the clock speed significantly when compared to the pipeline length. Stock clocks actually stayed the same.

The hierarchical sharing and high frequency design should help AMD achieve higher performance per mm2 of silicon – this is necessary for success, since Intel is roughly 12-18 months ahead of AMD’s manufacturing partner. The ultimate question is how does the novel architecture in Bulldozer translate into performance (both single threaded and multi-threaded), power and die area. Most of the physical characteristics of Interlagos, such as frequency and die area, are unknown. This leaves a great deal of uncertainty, as performance is highly dependent upon frequency as well as many of the details of the architecture which have been withheld for competitive reasons. The biggest questions about Bulldozer are the frequency, branch prediction, various queues and buffers in the front-end, handling of 256-bit AVX instructions, coherency protocol and northbrige microarchitecture.

The misprediction penalties plague the cache, the frequency never hit their target goals and the "hold the line" on IPC didn't succeed as they went backwards roughly 10% or so.

Some, including sources inside AMD, have blamed Global Foundries for not delivering higher clocked SKUs. Sure, the clock speed targets for Interlagos were probably closer to 3GHz instead of 2.3GHz. But that does not explain why the extra integer cores do not deliver. We were promised up to 50% higher performance thanks to the 33% extra cores, but we got 20% at the most.

From Johan's article.

Atenra, Bulldozer flopped. It's a horrible microarchitecture that's thirsty, was expensive at release and offers little as far as most computing goes. Trinity looks pretty good, particularly as far as power consumption goes. The 35W Trinity chips sip less power than equivalent 35W Intel CPUs, somewhere between the 35W and 17W ULVs which is incredibly impressive. The stock clocks on the high end 8350 signify that there won't be a drastic increase in single-threaded performance, but I'm expecting Vishera to at least be a decent CPU.

 

[dma0991]

This is now officially a Bulldozer John Fruehe discussion thread. :sneaky:

 

[Idontcare]

In Read's defense, he's still new as far as the semiconductor market goes. In order to see any of his plans we'll have to wait at least 3 years to even see a single PDF slide with his influence stamped on it. Looking for "direction" from a new CEO in a market that takes 3-5 years to see a single project through is a bit silly. He had no choice but to see BD through, but he's already taken steps with regards to Steamroller. On the fab end he's cut ties to GloFo and even embraced bulk so you've certainly got to give him some credit.

How long before bulldozer's release did you know of Bulldozer? Fusion? Bobcat?

You are talking about market release dates of products, I am talking about releasing a strategy to their analysts.

If a strategy existed then it would have been detailed to analysts within the first 100 days of Rory taking the CEO position, and we would have been inundated with dailytech articles on the matter.

The fact that there has been no clear redirection of the company, a year on, is very telling to everyone who follows these things. The mobile strategy excuse was just that, it was a way to admonish Dirk without cutting off their own nose in the process because they still had to sell those same bulldozer chips that led to his firing in the first place.

IDC you contradicting your self, you say that we have no idea about the targeted performance and then you say that it is clear that BD underperformed and failed to reach the internal objectives.

But i though that we didnt know the internal objectives, then how it is clear that it underperformed ???

Contradicting? Hardly. Do you know the mass of the sun? No, no one does. But we can confidently say it must be greater than the mass of the earth.

Is it a contradiction to say we know the universe is larger than the earth and yet at the same time we do not know the exact measurements of both to within a Plank length? No, I do not think so. The universe is clearly larger than the planet earth and at the same time we do not need to rigorously know the exact measurements of both to conclude as much.

It is true that I do not know the internal targets of bulldozer, but we can confidently conclude that it was targeted to be higher than what it ended up being given the number (and org chart positions) of heads that have rolled in its wake in addition to the fact that John Fruehe spoke of lower-estimates of its performance (IPC does not decrease) from an official capacity as a spokesperson for AMD via his corporate blogs and so forth.

Unless you take John Fruehe to be a liar, then you must agree that he was guided to believe the internal targets for bulldozer were higher than what they turned out to be. Why else would he so staunchly argue that IPC would not decrease only to then find out it did? Only to then move on to another position outside the company?

 

[pelov]

How long before bulldozer's release did you know of Bulldozer? Fusion? Bobcat?

You are talking about market release dates of products, I am talking about releasing a strategy to their analysts.

If a strategy existed then it would have been detailed to analysts within the first 100 days of Rory taking the CEO position, and we would have been inundated with dailytech articles on the matter.

The fact that there has been no clear redirection of the company, a year on, is very telling to everyone who follows these things. The mobile strategy excuse was just that, it was a way to admonish Dirk without cutting off their own nose in the process because they still had to sell those same bulldozer chips that led to his firing in the first place.

Their main direction is HSA/Fusion. Here are the PDFs from their FA day earlier this year.

What I got from it was:

1 - Fabless. They can't afford to invest the money nor be married to an underperforming GloFo. Going bulk really hammers this point home. Dirk would have kept going SOI-HKMG until hell froze over.

2 - They're putting a lot of emphasis on the mobile end, namely 17W-and-under and 35W. The desktop CPUs are gone from all slides after Vishera. As far as AMD's direction is concerned, desktop CPUs don't exist and only desktop APUs are shown. Had Read not been CEO then we'd have seen a desktop CPU, the fact that we don't even see a single slide regarding an AM3+ successor is a sign of a direction, don't you think?

3 - Regain some market share in server, mostly on HPC as it's a booming business.

4 - New interconnect via SeaMicro, putting a larger emphasis on GPU's and the PCIE ties. You'd figure that this would mean they'd embrace PCIE 3.0 but they'll still be using PCIE 2.X through 2013. Whatever they have planned for server is likely small and the added throughput per-PCIE isn't likely to bottleneck, meaning maybe Trinity/Bobcat type APUs rather than big (bigger) GCN things.

5 - Willingness to take in third-party IP, including ARM cores. Essentially, AMD is setting themselves up like IBM (where Read spent quite a long time). By willing to design products and incorporate third-party IP, Read is stating that AMD is willing to do custom designs for those that need the x86 license or on-die GPU or what-have-you.

The mobile strategy excuse was just that, it was a way to admonish Dirk without cutting off their own nose in the process because they still had to sell those same bulldozer chips that led to his firing in the first place.

Dirk got thrown out because Bulldozer failed but also because he was late with Brazos. AMD could have made some really good money had they released their Bobcat APUs in a timely fashion rather than midway through the netbook craze. Hell, even today their Bobcat chips are some of their bestselling processors. He didn't just not have a mobile plan, but the one product that bested Intel by a considerable margin in essentially every respect was released late and the successor had to be cancelled due to fab issues.

 

[jhu]

Really ??? they introduced a new product (Opteron 6200 series) that is 28% faster in Integer and 16% faster in FP and this doesn't put the new product into a better light ?? See above for Sandybridge.

The problem is that they're using more resources and getting lower performance than if they had stuck with the original. Sure a "16-core" BD is faster than a 12-core k10h, but it's still slower than a theoretical 16-core k10h. Even their Stars cores in Llano are faster than Phenom II. They could have saved all that time they had with BD, copied/paste Stars x4 (or x2 + MCM) into a new real 16-core Opteron and end up with a better processor.

Intel's efforts are not considered "fail" because they're using the same number of cores and achieving higher performance (for the same reason Llano is not a "fail").

Now, had AMD's marketing called BD for what it really is (4 core with CMT), then there wouldn't be such an uproar.

So BD isn't necessarily bad, depending on how it's priced, but AMD definitely could have done better while probably using less resources than they did developing BD.

 

[pelov]

jhu, the K10 cores don't scale up well in frequency but they do scale down pretty well, at least as far as power consumption is concerned. Check out the voltage readings some people were getting with K10stat on the Llano mobile chips. A lot of them are hitting 2.3ghz across all 4 cores loaded at 1.00v, with undervolting even more impressive. And it's not just a few random forum dudes, it's literally every single person attempting to fiddle with the voltage of the 35W and 45W Llanos. The 631/641 Athlons on FM1 had trouble hitting anything past 3.7ghz.

 

[Vesku]

Then there is the fact that under the new CEO, Rory Read, several mobile products have been pushed out farther in their roadmap. Granted it seems AMD launched a hiring program to beef up their SoC expertise but it sure seems they want to do it "right" rather than fast.

Yes :sneaky: that is clearly what it must have happened. Could not have had anything to do with faildozer, the timing was just a coincidence.

I hear it was also just coincidence that Pat Gelsinger left Intel on the eve of Intel announcing that Larrabee was being scuttled as well. The two were not related, Pat was just really really motivated to go work at EMC for personal reasons.

Had bulldozer turned out to be a hit and Dirk was let go then I'd buy the "lack of mobile strategy" cover-story. Given the facts on the ground though, no, I'm not buying it at all. Dirk gambled, and lost, on pursuing bulldozer at the expense of not pursuing alternative internally competing projects.

As captain of a ship that was steered into an iceberg, his early retirement was most certainly directly related to the performance of AMD under his stewardship and not related at all to hypothetical what-ifs that come with the "if only he had opted to develop a mobile strategy in parallel to prioritizing bulldozer".

Further proof that the mobile excuse was just that can be found in the fact that Rory has been CEO for a year now and AMD does not have anything close to a comprehensive mobile plan. If it was such a priority to the BoD as to motivate them to fire Dirk (versus any number of lesser punitive reactions) then to be sure the new CEO would have been well prep'ed to know he needed to roll out such a new mobile-facing strategy within his first 100 days of office.

But unsurprisingly Rory's AMD looks a lot like Dirk's AMD, the roadmap for bobcat and bulldozer remain in place and nothing even close to resembling a mobile strategy has taken shape, despite Rory being under the BoD's thumbs for a year now.

 

[AtenRa]

Intel's efforts are not considered "fail" because they're using the same number of cores and achieving higher performance (for the same reason Llano is not a "fail").

So, AMD have chosen to get higher performance by having more cores and that is FAIL because Intel have chosen a different path with less cores ???

Why it matters to you or anyone else how many cores any given CPU has ??? It is the performance you get that you should care about. Does it matter if you get 28% more performance from a quad core CPU or an 8-core CPU ??

As consumers we should only care about Performance/price and some will also should care about Performance/watt. How many cores each design needs to get that performance it should have zero impact in your purchasing decision.

 

[Vesku]

True but some performance suffered from the design choices and people who are affected by that need to take it into account. IMO, orochi server should have gained AMD a bit more market share but they got kicked in the groin by software vendors. But as I said earlier that is something even regular tech forum people had seen coming.

 

[pelov]

As consumers we should only care about Performance/price and some will also should care about Performance/watt. How many cores each design needs to get that performance it should have zero impact in your purchasing decision.

That's a contradictory statement. The higher the core count the higher the power consumption required in order to maintain the same low-threaded performance. The higher the core count the lower the clock speeds for a given TDP, whether MCM or not.

AMD offered slightly better multi-threaded performance (and only when using new ISAs and threads over 6) but worse single-threaded performance (in anything under 6 and not using AVX/FMA4). The price was also way too high considering what the Thubans were going for, roughly $170 for the 1090T, and they performed roughly the same as the 8150s which cost $270.

Intel's hyperthreading approach makes more sense than CMT + increasing core count at the expense of IPC, clock speeds and power consumption (and die size and because of that cost as well). Most workloads still aren't highly threaded, and for AMD to gain an advantage the workloads would have to have 4+ threads, namely 6+ to outrun the Thubans. Atenra, you're being disingenuous. You know full well that Bulldozer sucked it up on nearly all accounts for nearly every user imaginable. It's only for certain workstation workloads that the chip makes sense but those users don't mind spending an extra $50 (at release) for the 2600K or opting to go with the 2011 platform.

 

[AtenRa]

True but some performance suffered from the design choices and people who are affected by that need to take it into account.

Of course, each design has its pros and cons and people should know in order to chose whats best for them. But we should not care how many cores the CPU has if you get the same performance in the same application from a CPU with 4 or 8 cores.

 

[dac7nco]

I also got banned for a few weeks at a time for daring to say that on the balance of probabilities, Bulldozer was going to be a dud and that IPC would be lower than Thuban.

However I think the appalling moderator in that case, may have actually been in love with JF-AMD.

His conduct was the worse I have ever seen on mainstream forums, by a company representative.

And most ludicrous of all, is how much information his supporters must be blissfully ignorant of, to not see his despicable behaviour.

AMD's engineers at a Hot Chips convention stated that one of their primary goals they were working on with Bulldozer was to minimise IPC loss, with the clear inference being that there was going to be some IPC loss.

This was publicly stated months before Bulldozer was released, yet JF-AMD kept insisting that IPC would be better than Thuban's.

You were probably banned due to the fact that quite a few people, including you, seemed to be lashing out at AMD supporters. Lots of us wanted some competition in between CPU manufacturers again in their lifetime. There's no reason to shit on people who's hopes are that level.

JF-AMD not being in the loop with AMD's internal rumors... I doubt it. The fact that they needed an advance front-man to promote the new "FX/Opteron"... Shit. You don't see IntelEnthusiast spreading bullshit. Hypertransport and QPI started life as server products. $4,200 and 64 cores on the AMD side prices-out to 16 cores on Intel, and gets burned with very few exceptions. Anandtech has to cherry pick Pro-AMD benchmarks in the CPU space.

Bulldozer's IPC being less than Thubans? That's shit which happens when you don't know what you are doing. Be happy we aren't still on X58/P55 right now, because Intel is competing with 2009.

Daimon

 

[exar333]

So, AMD have chosen to get higher performance by having more cores and that is FAIL because Intel have chosen a different path with less cores ???

Why it matters to you or anyone else how many cores any given CPU has ??? It is the performance you get that you should care about. Does it matter if you get 28% more performance from a quad core CPU or an 8-core CPU ??

As consumers we should only care about Performance/price and some will also should care about Performance/watt. How many cores each design needs to get that performance it should have zero impact in your purchasing decision.

Thuban 904M transistors
BD 1.2B transistors

Added 33% more transistors and gets 20% less IPC. That's WITH a new arch.

If AMD or NV released a new GPU that had 30% more transistors and was 20% slower than the previous gen AND process node, it would be a MASSIVE fail. I don't see any difference here.

BD is an 'OK' chip. But successful it was definitely not.

 

[happysmiles]

oh look another endless argument.

 

[AtenRa]

That's a contradictory statement. The higher the core count the higher the power consumption required in order to maintain the same low-threaded performance. The higher the core count the lower the clock speeds for a given TDP, whether MCM or not.

I was talking generally and not about BD, we shouldnt care if one CPU has more cores than the other if they both have the same performance and power consumption in the same application.

AMD offered slightly better multi-threaded performance (and only when using new ISAs and threads over 6) but worse single-threaded performance (in anything under 6 and not using AVX/FMA4). The price was also way too high considering what the Thubans were going for, roughly $170 for the 1090T, and they performed roughly the same as the 8150s which cost $270.

I do agree that FX8150 at $270 was way higher than it should be. But FX8120 was at a much better price point from the start.

Intel's hyperthreading approach makes more sense than CMT + increasing core count at the expense of IPC, clock speeds and power consumption (and die size and because of that cost as well). Most workloads still aren't highly threaded, and for AMD to gain an advantage the workloads would have to have 4+ threads, namely 6+ to outrun the Thubans.

Since those two CPUs (Core i and BD) are of different designs and are manufactured with different process you really cant say if one design is better than the other. We can only compare them the way they are.

Atenra, you're being disingenuous. You know full well that Bulldozer sucked it up on nearly all accounts for nearly every user imaginable. It's only for certain workstation workloads that the chip makes sense but those users don't mind spending an extra $50 (at release) for the 2600K or opting to go with the 2011 platform.

http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested/7

3dsmax 9 : faster than Phenom II 1100T
Cinebench R10 : faster than Phenom II 1100T
Cinebench R15 : faster than Phenom II 1100T
7-Zip Benchmark : faster than Phenom II 1100T (Even faster than 2600K)
PAR2 Benchmark : faster than Phenom II 1100T
TrueCrypt Benchmark : faster than Phenom II 1100T (close to 2600K)
x264 HD 3.03 second pass : faster than Phenom II 1100T (close to 2600K)
x264 HD 3.03 AVX second pass : faster than Phenom II 1100T (Even faster than 2600K)
Adobe Photoshop CS4 : faster than Phenom II 1100T
Compile Chromium Test : Slower than Phenom II 1100T
Excel Monte Carlo : faster than Phenom II 1100T

Also, much lower idle power consumption than the Phenom II 1100T and same performance/power at full load.

From 11 test it was slower only in one of them and it was close with Core i7 2600K in four. Was it worth it to upgrade over a Phenom II 1100T ?? imo NO. But that doesn't make it a FAILed product. In that case, IvyBridge is also a FAILed product because it is not worth it to upgrade over a SandyBridge.

 

[jhu]

So, AMD have chosen to get higher performance by having more cores and that is FAIL because Intel have chosen a different path with less cores ???

Honestly, yes. See below.

Why it matters to you or anyone else how many cores any given CPU has ??? It is the performance you get that you should care about. Does it matter if you get 28% more performance from a quad core CPU or an 8-core CPU ??

It matters because not all computations can be easily parallelized. That 28% increase in integer performance that you keep quoting, it should have been at least 33%. Intel was taken to task for Pentium 4, and so should AMD for Bulldozer.

As consumers we should only care about Performance/price and some will also should care about Performance/watt. How many cores each design needs to get that performance it should have zero impact in your purchasing decision.

This is exactly where the fail comes in. When BD was introduced, the 8150 cost over $250. At the time I got my 1090T for $160. BD did not deliver $100 worth of performance over the 1090T. Now prices are more reasonable - $180 for 8150 on Amazon.

 

[AtenRa]

Thuban 904M transistors
BD 1.2B transistors

Added 33% more transistors and gets 20% less IPC. That's WITH a new arch.

If AMD or NV released a new GPU that had 30% more transistors and was 20% slower than the previous gen AND process node, it would be a MASSIVE fail. I don't see any difference here.

BD is an 'OK' chip. But successful it was definitely not.

Ehm, IPC and performance is not the same. 8-core BD is faster than Phenom II X6, 6-core BD is faster than Phenom II X4 and 4-core BD is faster than Phenom II X2.