So here’s the story on IPC from my perspective.
The original design goal was higher IPC. Back before we had taped out the first processors the discussion came up. I am not allowed to say anything in public that hasn’t been vetted past engineering. I specifically asked the engineering team about IPC and they replied that they expected higher IPC and I was cleared on that statement.
In my estimation, I made the IPC stament on XS and I don’t recall making it other places (but I am sure that my comments were reposted.)
This is not a case of me lying, this is a case of me being wrong. And that happens to all of us. There is no way that I would intentionally say something that I did not believe to be true because a.) the interenet is a permanent record and b.) eventually you’ll have to answer to those comments.
In these different forums that I would frequent, I have made
thousands of posts about the product, always trying to help people understand the technology. Everything was done with the intention of being informative. I never made client performance predictions and was very clear on correcting people who took server benchmarks and tried to correlate client performance, because it just doesn’t work like that.
Since the launch I haven’t really been in any forums because I was in Seattle all week with MSFT. Once the product launched, I started receiving lots of hate mail. If I had a dollar for every message that used the word f*** I would have enough money to buy myself a bulldozer server.
Early on there were a lot of people that said I was crazy for trying to talk to the enthusiast community, but I persisted, the occasional idiot is something that everyone has to put up with from time to time. But the volume of hate messages that I received has unfortunately convinced me that I was wrong.
I was doing this on my own time, it was not part of my job. And because of that, I don’t really feel compelled to take any more abuse.
All of the statements that I have made at the time were statements that I believed to be true. I am no more a liar than Paul Otellini when he said IB would ship in Q4. He is not a liar, he is giving the best information he has at the time and things change. If you are going to extend courtesy to others when their statements don’t pan out, you should do that with everyone.
Thanks to exarkun and some of the others that have had a
reasonable response in all of this. Too many people were waiting to attack, it’s almost like they don’t really care, maybe they just like to fight.
Well, what to make of it? He dug the hole for himself, apparently others (I assume inside the company?) even warned him about it. He went on anyway, then fell into the hole that he dug himself.
I think a man in his position should have known better. It’s the responsibility that comes with such a high-profile position at a high-profile company.
Aside from that, I think it’s a bit too easy to just say “Well, engineering told me that, it’s not my fault”. We have no way to verify if engineering REALLY said that (they could have said something slightly different, and Fruehe may just have misinterpreted it, then ran with it). I personally don’t think that engineering would say that, but that’s obvious, since I’ve said all along that I see some big performance issues with how AMD decided to allocate the execution units in their modules. Therefore I would think that the engineers at AMD would know about this as well, and would not bring forth such information. In fact, I even pointed out some AMD slides that said the opposite:
“Throughput advantages for multi-threaded workloads without significant loss on serial single-threaded workload components”.
As I already said about that remark, at the time:
“Well, that pretty much warns us already that serial single-threaded performance (in other words: IPC) may take a small hit… albeit not a ‘significant’ one.”
The logic being the following: it is a marketing slide. There is no reason to mention negative things such as ‘loss on serial single-threaded workload’ as long as you are confident that the resulting product will perform better anyway. My spider-sense told me that they put this in as a warning, to not get people’s expectations up (which JFAMD then completely undermined).
As old as these slides are, they are still correct today. A small, but not significant loss in serial single-threaded workloads is what we see, compared to Phenom II (I equated that to IPC since it appeared that clockspeeds of Phenom and Bulldozer seemed to be about equal. Which is still the case, although Bulldozer does have more turbo potential. Which does not compensate enough of the IPC to beat Phenom II in single-threaded workloads).
However, even if we assume that they DID say so originally (Fruehe made his posts around August 2010), clearly at some point during development those engineers would have realized that they were running into problems and were going to come up short. Why has Fruehe then insisted for so long that IPC will increase? AMD didn’t find out at launch day that IPC was lower than Phenom II. They must have known this months in advance. But even though leaked benchmarks started appearing, indicating the IPC problem, Fruehe continued to insist that the benchmarks were faulty, reinforcing his earlier IPC statements (so at this point he was indeed perpetuating the wrong information, hence lying).
So, apology not accepted. Firstly, I think Fruehe (and AMD in general, also including people such as Richard Huddy, who have occasionally passed on misinformation on forums, or in interviews) just has to learn the hard way that it was not a smart thing to post such information in public forums in the first place. It’s your own fault, you should have known: “what goes on the Internet, stays on the Internet”. Just like Randy Allen, this will haunt you for the rest of your life.
And secondly, I don’t think Fruehe has done enough to prevent the inevitable outcome. Had he retracted his IPC statement and at least warned the people before launch, then I think people may have been a bit more forgiving. Had he done it far enough in advance, the storm may have blown over by the time of the actual launch, and he might have actually somewhat saved Bulldozer’s reputation (and his own). I suppose he only made it worse by actively debunking leaked benchmarks, and even being insulting in the process (implying that people were paid by Intel to post these things… which is ironic enough, posted by someone paid by AMD). That’s the marketing game, John. You should know how to roll with it.