Jump to content
HWBOT Community Forums

JF-AMD's Bulldozer Pre-launch FAQ


Recommended Posts

Found this post over at Overclock.net. Although JF-AMD is working for the server division, the post applies to more than just the server segment. In my opinion, it gives a pretty nice insight on future release strategies.


The IPC question/answers are quite relevant for the overclocking community. In general, I'd say overclockers are interested in CPC performance (clock-per-clock), but in the end they'll take whatever hardware will give them the highest score. For instance, Sandy Bridge is much faster clock-per-clock in 3DMark05, but with Gulftown reaching 500-700MHz higher clocks we choose the 990X to set our records.


OK, let me begin by saying "I'm gonna post this once."


Everyone, please just point people back to this post when the 1,957,837,842 questions show up that are all the same. This will be easier than me posting it up over and over.


Admin, please lock this thread ASAP, I am not going to answer questions here.




Q. I saw a benchmark on xyz website. Is that how bulldozer will perform?


A. No. Nothing posted before launch will be representative of actual performance. To get actual performance, you need:


Final production silicon

Final processor microcode

Final system BIOS

Final OS optimizaitons

Final drivers

An app compiled with the latest flags

A person who understands the app and configures the test properly


Without these things (and there are probably more), you cannot get an accurate benchmark. Any extrapolation of a crappy benchmark gives you a crappy estimate of actual performance. Period.


Oh, and many of the benchmarks that you see were probably not run, those are just charts made in excel. It's really easy to make a chart in excel - what do you want, bulldozer faster by 3716%? Intel faster by 293%? Sure, I can do either one in 10 seconds. Why are you arguing about benchmarks that probably aren't real?


Q. When are you launching? Why don't you release the date?


A. When we launch, we launch. I will not comment on dates, I will not comment on schedules. We do not release dates prior to launch, at most we give quarter granularity. Giving the date out will stall demand. We have a business to run. While you might think that it will make your life easier to not have to guess, the reality is that we have a business to run and the minute you let the date out, sales stall. For everything. The cost impact of announcing the launch date is always bigger and drives these decisions.


Q. What are the prices?


A. Look at the above question and you have your answer.


Q. Why don't you release benchmarks before launch? You could steal so much business away from the other guy?


A. Again, releasing benchmarks before launch will simply stall sales. Believe me, if the competition thinks they are out of position, will they just sit back and say "oh well" or will they react? Handing them benchmarks is simply giving them time to form a strategy. I am not in the business of helping them, they are on their own on this one.


Q. Will bulldozer be faster than....?


A. I don't need to finish the question, please read above.


Q. I read on xyz site that you were launching on xxxxx?


A. Yeah, and I read on another site that elvis was still alive. The reason I don't comment on date rumors is that there are a limited number of days in the quarter. Once you say no to some, and suddenly say "no comment" or don't answer that one, immediately everyone thinks that is the date. So, no matter how crazy it sounds, you can't answer any of them.


Q. I saw someone selling bulldozer parts online, that must mean the launch is happening, right?


A. No. First off, many of the people that are advertising parts for sale ahead of time do not have parts in hand. Buyer beware. If they are selling engineering samples, we will take care of that. Occasionally parts are loaded into disti databases, and if the flag is set wrong, it can flow through EDI to a partner's database and show up online. That does not mean parts are available. Oh, and sometimes distis use planned pricing and part numbers as placeholders, so don't believe what you see. Nobody is allowed to advertise parts ahead of launch.


Q. I know someone who knows someone who has a cousin who lives next to...?


A. Ask yourself this. Would you take investment advice from someone like that? If not, put a big grain of salt into those conversations.


Q. Why haven't you guys launched?


A. The key to a successful product is lauching it once you are ready. When you are trying to put hundreds of millions of transistors onto a piece of sand about the size of your thumbnail, there is a lot going on. It's a pretty complicated busines.


Q. Why don't you comment on client stuff? Why don't you get a client guy to post on here?


A. I am in the server world, I comment on my products. I will not comment on their products, because I don't want them commenting on mine. The few times I have ventured into the client world, I have caught flack for my comments. It is not worth the trouble. Based on the filth that seeps into my inbox from a very small segment of the enthusiast world, I can understand why the client people might not want to make their names known. When you are attacked in a vicious way for no reason, and you are doing this on your own time, you start to ask "why bother"?


Q. I see you have <feature x> on server, so we will see that on client too?


A. I can't say for sure. We both productize different features based on our customer needs. Wait for the launch to see for sure.


Q. Why aren't you doing an unlocked Opteron so I can do 2P overclock?


A. First the market for 2P client systems is tiny, and getting smaller every day. While you would like to buy one, there just aren't enough people to do that. Second, I want to focus my time on servers, client overclocking will not be supported because it takes efforts away from making virtualization and cloud computing better.


Q. If AMD does beat intel in xyz it is the end of the road


A. OK, this is not a question, but enough people make statements like this. AMD is a 40 year old company, not likely.


Q. Is IPC higher on bulldozer? All I care about is IPC.


A. IPC is simply a measurement. What if IPC was 2X what it is today, but clock speed was 500MHz. Is that what you want? You are getting double IPC, right? IPC is only one measure. The people that are telling you IPC is the only thing that matters have an agenda. Taking only one measurement, out of context, is like trying to say that a person's weight is all that matters. I weigh 195. Does that make me fat? Does that make me skinny? It is impossible to say unless you know my height. IPC is like weight - it tells you something in context to other factors, but is meaningless on its own.


Q. What about single threaded performance?


A. See above. Also, if all you care about is single threaded performance, might I recommend a lovely, inexpensive single core processor for your system?

Link to comment
Share on other sites

One note to JF-AMD's post, though. A question I've brought up when Sandy Bridge was almost ready to launch and when Dr.Who? (Intel) informed the folks at XS that the real performance of SB was not known yet.


Both basically argued that any performance result prior to the retail launch cannot be taken seriously because a lack of final 'everything' (as JF states: final production silicon, processor microcode, system BIOS, OS optimizations, drivers). The problem is that media receive test samples prior to the launch to have their articles up by the time the product is in the stores. A consequence of the logic as stated by JF is, in other words, that no launch article is presenting accurate performance results OR that leaked results do indicate real performance levels.


I assume the latter is true, obviously. The only question that remains is how much tme prior to the launch can final performance be measured by the reviewers that test the product. One week, two weeks?

Link to comment
Share on other sites

  • Crew

About that Massman, we use to wait weeks before the BIOS are good (it'd way better now but remember x58? ) so even with a specific time and latest silicon / version etc... there will be some change.


I assume that a es or silicon at the same revision of the retail part will perform the same. But that is true thann.. if silicon prior launch are not representative... none of the "on launched reviews" are true... or 100%accurate :) don't think so ...


About the leaked result etc, we Knox that ocing capabilities are highly impacted depending if its a pre batch silicon or the retail one... but on IPC there can't be like 10/15% difference.... the thin is you can only estimate thing ...

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Create New...