AMD - Bulldozer Chitchat

[CAT-THE-FIFTH]

Looks like Intel don't feel too threatened

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

Its more like retailers which are probably doing this as there are gullible people who are willing to pay £20 to £30 more.

All the geek rage against Bulldozer has made retailers realise they can get away with charging more for the same processors they bought at lower prices. The geek rage has only served to push up prices.

You need to consider Intel Core i5 2500K and Core i7 2600K pricing this year was relative to Phenom II X6 CPUs which are still being made. If anything the FX8150 and FX8120 don't change this situation very much at all. The lower end Core i5 prices have not changed that much TBH only the K series.

Its like how many of the UK retailers are now pushing up prices of hard drives in anticipation of hard drive shortages.

[CAT-THE-FIFTH]

You can ask AMD questions about Bulldozer in this thread:

http://hardforum.com/showthread.php?t=1645024

[watercooled]

Any idea when someone will answer the posted questions?

[Biscuit]

i think it says they will publish the answers as an interview

[scaryjim]

OK, I was having a think yesterday. We all know that Bulldozer is dissappointing in lightly threaded tasks, and a monster in the right kind of parallel workloads.

I'm pretty sure there's more than one reason for this, so here's a couple of ideas that I'd like to throw open for debate. Feel free to tear me down

1) FPU Mis-scheduling.
Reading around it looks like the intended FPU operation is that when the FPU isn't processing 256bit AVX instructions, it's mean to be able to handle 2x 128bit FP instructions simultaneously. The benchmarks show pretty clearly that this isn't happening. So is this just a problem that the FPU scheduler isn't dispatching the second instruction when it's able to? If so, could that be fixed in a stepping (B3?!) and miraculously boost the FPU throughput by up to 100%?

2) Branch Misprediction
AMD's redone branch prediciton with bulldozer to have a "quick" and a "slow" predictor. The idea is that when the cores are lightly loaded the quick predictor will make sure there's something to work on so the core isn't sitting idle, but it's not as accurate: when the cores are heavily loaded the slow predictor will have time to put a much more accurate prediction into the queue. So what if the quick predictor is just really bad? Could the poor performance in lightly threaded loads be a case of the quick predictor pushing lots of bad predicitons into the queue so it keeps having to flush and restart? If the pipeline is particularly long (as rumoured) then (iirc from the p4 days) incorrect prediciton can have a hugely detrimental effect on performance...

[CAT-THE-FIFTH]

The TechReport has had a quick look at BD thread scheduling:

http://techreport.com/articles.x/21865

[DanceswithUnix]

OK, I was having a think yesterday. We all know that Bulldozer is dissappointing in lightly threaded tasks, and a monster in the right kind of parallel workloads.

I'm pretty sure there's more than one reason for this, so here's a couple of ideas that I'd like to throw open for debate. Feel free to tear me down

1) FPU Mis-scheduling.
Reading around it looks like the intended FPU operation is that when the FPU isn't processing 256bit AVX instructions, it's mean to be able to handle 2x 128bit FP instructions simultaneously. The benchmarks show pretty clearly that this isn't happening. So is this just a problem that the FPU scheduler isn't dispatching the second instruction when it's able to? If so, could that be fixed in a stepping (B3?!) and miraculously boost the FPU throughput by up to 100%?

2) Branch Misprediction
AMD's redone branch prediciton with bulldozer to have a "quick" and a "slow" predictor. The idea is that when the cores are lightly loaded the quick predictor will make sure there's something to work on so the core isn't sitting idle, but it's not as accurate: when the cores are heavily loaded the slow predictor will have time to put a much more accurate prediction into the queue. So what if the quick predictor is just really bad? Could the poor performance in lightly threaded loads be a case of the quick predictor pushing lots of bad predicitons into the queue so it keeps having to flush and restart? If the pipeline is particularly long (as rumoured) then (iirc from the p4 days) incorrect prediciton can have a hugely detrimental effect on performance...

It sounds like it can do 2x128 AVX instructions. Might be useful in the future, but not as useful right now as 2 SSE instructions.

As for branch mispredictions, that is the sort of thing that would get massive simulation efforts so I would be shocked if they got that very wrong. On the other hand, they will now have lots of live data and will no doubt apply some tuning to the predictors over time.

Re-spins in interesting. Traditionally they tried to change as little as possible in a minor step update, but if this is a synthesisable part as has been rumoured then more major changes in steppings might be possible?

That is kind of the point about Piledriver in the Llano replacement thought isn't it. Fusing the gpu means it has to be a complete new floorplan, hence they might as well use an updated design for the CPU. Am still surprised they can switch bulldozer over to the new design so fast though.

TSMC say you need to sell 30M chips to break even on a design. Don't think AMD made that many Bulldozers yet though

http://www.brightsideofnews.com/news...hcon-2011.aspx

[kalniel]

OK, I was having a think yesterday. We all know that Bulldozer is dissappointing in lightly threaded tasks, and a monster in the right kind of parallel workloads.

I'm pretty sure there's more than one reason for this, so here's a couple of ideas that I'd like to throw open for debate. Feel free to tear me down

I think there's a much simpler explanation:

Long pipeline.

This really hurts sequential tasks if your clockspeed isn't much higher to compensate, but has less effect on very parallel loads which can start a new independent thread.

[CAT-THE-FIFTH]

The FX8150 is tested under Ubuntu at Phoronix:

http://www.phoronix.com/scan.php?pag...ulldozer&num=1

[DanceswithUnix]

Think I'll give up with Phoronix.
Just spent 10 mins trying to find the relevant text for their BD compiler article amongst the noise on the page to learn that turning the optimiser off generates slow code. Wouldn't exactly be hard to come up with a genetic algorithm to hunt for the best (or close enough at least) optimiser settings for a task.
If only I had the time to set up my own Linux benchmarking site

[CAT-THE-FIFTH]

I was looking at this FX8150 review and noticed this:

http://translate.google.com/translat...en/index17.php

It will be interesting to see how BD and PD fares with better support for AVX and FMA.

[Noxvayl]

That is very interesting. Considering the massive price difference between the FX8150 and 990X those are incredible results even though they are highly specific.

[CAT-THE-FIFTH]

Some more info about Trinity:

http://www.donanimhaber.com/islemci/...da-her-sey.htm

http://www.xtremesystems.org/forums/...=1#post4998434

[watercooled]

You can ask AMD questions about Bulldozer in this thread:

http://hardforum.com/showthread.php?t=1645024

Any idea if the interview has been published yet?

[CAT-THE-FIFTH]

Any idea if the interview has been published yet?

http://www.hardocp.com/article/2011/...ozer_questions

"We are also working with Microsoft on a scheduler update for Windows 7 that will be available soon."

[Biscuit]

the reviewers conclusion sums it up for me:

From a "reviewer" stance, the answers have not done much to give me any hope that there are any immediate Bulldozer performance remedies we will see any time soon. The mention of an updated Windows 7 scheduler to be released is certainly nice to hear though. We know where Bulldozer sits on the desktop in terms of performance today, and that is firmly behind the processors from Intel's strongly competitive lineup. This is of course not to say it is a "bad" product, far from it, but certainly it is still a letdown in the eyes of the enthusiast. While some of the marketing talk here is fairly thick it will be interesting to see where Bulldozer lands as true multithreaded applications evolve along with the newer operating systems.

It is good that there is finally an official announcement of a patch which may improve the Win 7 Scheduler, cant see it making groundbreaking changes to anyones decision in buying a CPU though.