AMD - Bulldozer Chitchat

[CAT-THE-FIFTH]

Comment on Bulldozer IPC by JF-AMD:
http://forums.anandtech.com/showpost...&postcount=106

It seems the different motherboards seems to show different results to the Asus one used in most reviews:

http://forums.extremeoverclocking.co...=358508&page=3

http://www.overclock.net/amd-cpus/11...board-may.html

[flearider]

anyone know where you can get a 8120 ?

[CAT-THE-FIFTH]

http://quinetiam.com/?p=2356

Edit!!

I saw it over on the Aria forums but having read through it, it looks like FUD.

[EvanJackPenn]

Hmmm... maybe Piledriver could be a sucess. If they iron out the kinks, add some performance, sort out the power and ship reviewers a working mobo, they may get a better response. Especially if this years games now scale well with multiple cores (e.g., Battlefield/Skyrim). If they launch it alongside the Radeon 7000s, it may be a hit with the gamers...

[Noxvayl]

The logic does make sense, scheduling for each integer core can't be helping in most applications. But then there has been loads of talk about a scheduling issue, I hope it turns out to be as big an issue as that guy made it out to be.

I'm more interested in the release of BF3 and Skyrim to see how well threaded they are and if Bulldozer works admirably with them.

[CAT-THE-FIFTH]

The logic does make sense, scheduling for each integer core can't be helping in most applications. But then there has been loads of talk about a scheduling issue, I hope it turns out to be as big an issue as that guy made it out to be.

I'm more interested in the release of BF3 and Skyrim to see how well threaded they are and if Bulldozer works admirably with them.

BF3 does seem to work fine with it:

http://www.hardocp.com/article/2011/...mance_review/2

http://www.kitguru.net/components/cp...990fxa-ud7/19/

[Biscuit]

Comment on Bulldozer IPC by JF-AMD:
http://forums.anandtech.com/showpost...&postcount=106

It seems the different motherboards seems to show different results to the Asus one used in most reviews:

http://forums.extremeoverclocking.co...=358508&page=3

http://www.overclock.net/amd-cpus/11...board-may.html

Wow that's interesting. I mean it is still distinctly average for the price but at least the performance is a bit more competitive. Lends itself to the theory that bios does make a big difference.

[Hixxy]

Piledriver will be the one to watch.

[DanceswithUnix]

Charlie has done a couple of post-mortem articles on Bulldozer on Semi-Accurate, but interestingly one of the comments to the article was to this pile of rumour mongering:

http://hardforum.com/showpost.php?p=...8&postcount=88

Wonder how much is true...

[CAT-THE-FIFTH]

It looks like a B3 revision is in the works:

http://www.planet3dnow.de/vbulletin/...d.php?t=399157

Look at page 27 of the document.

[scaryjim]

Some of the follow on comments from that are actually quite telling in terms of understanding how businesses work. Namely, "yes, the tools aren't as good as hand designing, but AMD don't have time to hand design every factor of the chip and publish a believable release schedule". Which makes sense to me.

I suspect, however, that there were problems with how the changes were implemented (again, not unusual in any size of business) and they lost a lot of expertise in the shake-down that could've been very valuable as an addition to the machine-design process. Problem with that, is persuading old-hand engineers to learn new tricks. It should be possible, but whether it's probable... that's another matter

[DanceswithUnix]

Well I remember talking to someone from ARM many years ago who said they went through that sort of pain when switching from hand layout to fully sythesisable, and basically enough performance that it took I think 6 months to wait for process improvements to get that back.

So not even sure it is learning new tricks. I was asked to switch from C programming to Java, I don't want my code performance to suck that badly, so I left the company. I am learning plenty of new tricks here, just much faster ones

[watercooled]

I was asked to switch from C programming to Java ... so I left the company.

I don't blame you! xD

[scaryjim]

It's definitely a balance. Last time I decided to change languages it was from VB to C# though, so I've got a different viewpoint

The bottom line is that you can manage those changes successfully, but only if you understand the changes from both sides of the fence. Obviously in both AMDs case, and yours, it wasn't well managed, as in both cases experienced staff were lost due to the change. One suspects that a gradual change could have been made, blending machine-designed and hand-designed components, or starting from a hand-designed component then machine optimising it, and using the expertise of the expert engineers to improve the machine design tools.

Well, that's how my company would approach it, anyway

[DanceswithUnix]

Looks like Intel don't feel too threatened

http://www.brightsideofnews.com/news...er-launch.aspx

[scaryjim]

If that's true then it's dangerously cocky of Intel - sure, they may have already hit large parts of the market they were hoping to keep safe from AMD, but pushing their prices up is only going to make FX seem more attractive - if the price difference between FX-8159 and i7 2600K has gone from, say 40 euro to 60 euro, BD starts looking a better buy...