AMD - Bulldozer Chitchat

[Biscuit]

i dont really understand the point in that?

[CAT-THE-FIFTH]

i dont really understand the point in that?

It seems the CPU cooler is too big and obstructs the first two RAM slots:

http://www.freeimagehosting.net/newuploads/58c92.jpg

[Terbinator]

4.89 in CB11.5.

[Biscuit]

It seems the CPU cooler is too big and obstructs the first two RAM slots:

http://www.freeimagehosting.net/newuploads/58c92.jpg

i meant the live stream... so pointless.

[CAT-THE-FIFTH]

It looks like there is supposedly is an issue with prefetch and AMD is working on a patch:

http://www.xtremesystems.org/forums/...=1#post4969164

I suspect that decent reviews sites like Anandtech would know about this. Even if they could put a review by the 12th in place without it they would tell us.

[Biscuit]

I suppose you are referring to this post?

Actually, we already have such an issue known for Bulldozer, and NO bench-marked system has the patch installed!

The shared L1 cache is causing cross invalidations across threads so that the prefetch data is incorrect in too many cases and data must be fetched again. The fix is a "simple" memory alignment and (possible)tagging system in the kernel of Windows/Linux.

I reviewed the code for the Linux patch and was astonished by just how little I know of the Linux kernel... lol! In any event, it could easily cost 10% in terms of single threaded performance, possibly more than double that in multi-threaded loads on the same module due to the increased contention and randomness of accesses.

Not sure if ordained reviewers have been given access to the MS patch, but I'd imagine (and hope) so! Last I saw, the Linux kernel patch was still being worked on by AMD (publicly) and Linus was showing some distaste for the method used to address the issue. One comment questioned the performance cost but had received no replies... but you don't go re-working kernel memory mapping for anything less than 5-10%... just not worth it!

In which case that makes sense. Its only afew days till release and people are seriously getting their panties in a twist i really dont get it. Get some sleep and continue with your life for heavens sake...

[CAT-THE-FIFTH]

I might have mentioned this before in the thread but AMD release some Linux patches during the summer. I wonder if this was to address the issue??

It looks like this information from last month was true:

http://semiaccurate.com/forums/showp...&postcount=813

TBH,I don't see why AMD just did not wait a few more months until they launched Bulldozer??

[Biscuit]

I might have mentioned this before in the thread but AMD release some Linux patches during the summer. I wonder if this was to address the issue??

It looks like this information from last month was true:

http://semiaccurate.com/forums/showp...&postcount=813

TBH,I don't see why AMD just did not wait a few more months until they launched Bulldozer??

I couldnt make sense of that link, the translation is pants.

[CAT-THE-FIFTH]

I couldnt make sense of that link, the translation is pants.

I was reading through the thread,and from the original post it had been inferred that OS support was the main issue. I will see what the reviews from the decent site say this week.

[DanceswithUnix]

Charlie has been writing up what he is allowed to tell us about BD so far:

http://semiaccurate.com/2011/10/10/e...-architecture/

[HalloweenJack]

it seems that the kernal needs a patch to address a few issues - and so far NO ONE has that patch from MS - reading XS , its more to do with the fact windows favours HT over real cores and the entire new architecture causing anything up to 20% performance loss.

[kalniel]

it seems that the kernal needs a patch to address a few issues - and so far NO ONE has that patch from MS - reading XS , its more to do with the fact windows favours HT over real cores and the entire new architecture causing anything up to 20% performance loss.

Windows favours HT cores.. that is in the traditional sense of the word favours, ie, rests. That shouldn't be an issue for bulldozer because there are no HT cores. However, the split int/fp scheduling might be a bit confusing. But XS is talking about shared L1 cache problems - which is to be expected because Windows not had to deal with that before. It's almost like instead of having an 8 core chip, it's actually more like having a 4S dual core setup.

[Terbinator]

4S = 4 Socket/Processors?

[kalniel]

pics and benches of the 8120:
http://www.overclock.net/amd-general...s-benches.html


edit: terb, yes. There's a tight link between the two cores in a module, including shared cache, that needs to be taken account of. Then there's further links between the modules.

[CAT-THE-FIFTH]

Another 24 hours or so until official benchmarks !! Will this be D-Day for Bulldozer or a Dunkirk??



OR



??

[shaithis]

I ain't waiting for benchies.....I'm waiting to order one!!!