Do you own a Bulldozer CPU?

[Chiropteran]

I just bought a Phenom-II 1100T from Microcenter. Main reason was that the stock at Microcenter is going out too fast. FX will still be available if I decide to go with that in the future.

Thanks for all the help!

I'm sure it'll be a fine CPU for your needs, enjoy it.

 

[gmaster456]

It is not really "4 cores" though. More like 2 in terms of performance - maybe 2.5-3.

Think of it as 2 full cores, and two 80% cores.

 

[Vic Vega]

Having been underwhelmed by Bulldozer, I'd still like to see Anandtech revisit it when the MS patch is finalized and with refined drivers.

Don't bet on it. There's a reason BF3 benches were not included. Showing BD in any positive light is not on Anand's agenda.

 

[Idontcare]

Don't bet on it. There's a reason BF3 benches were not included. Showing BD in any positive light is not on Anand's agenda.

Anand's "agenda"

 

[soccerballtux]

finally a benchmark Bulldozer wins at!!!

 

[soccerballtux]

Anand's "agenda"

of all the reviewers I have to say Anandtech is definitely the kindest towards AMD.

 

[soccerballtux]

OP - These benchmarks are why i decided to try a 6100. My usage is highly tied to Excel so this is a good match for me - if you look at the rest of the review you will see that BD doesn't look so good in the other benches.

It is almost blasphemy to say anything positive about BD around here, but it does have a few areas that it does fine in.

link to benches .. http://www.techspot.com/review/452-amd-bulldozer-fx-cpus/page8.html

Encoding benches .. http://www.techspot.com/review/452-amd-bulldozer-fx-cpus/page9.html

who does serious excel though? If you need to operate on datasets large enough to the point that Excel is bogging down, you'd be working in Matlab anyways. I've always been mystified as to why Excel gets considered as a "benchmark". Any CPU out in the last 5 years is plenty for any and all office "productivity" that happens in Excel. IMO.

 

[soccerballtux]

Having been underwhelmed by Bulldozer, I'd still like to see Anandtech revisit it when the MS patch is finalized and with refined drivers.

I'd like to see what said patch does for Phenom2 X6's too...I want to say I saw some benches of people running it on a Phenom2 system and performance improved even more than it did for Bulldozer...which made the whole thing look a bit shady to me...

 

[Magic Carpet]

I'd like to see what said patch does for Phenom2 X6's too...I want to say I saw some benches of people running it on a Phenom2 system and performance improved even more than it did for Bulldozer...which made the whole thing look a bit shady to me...

What does that patch do, anyway?

2) What would it do to a non-HT cpu?

 

[Idontcare]

who does serious excel though? If you need to operate on datasets large enough to the point that Excel is bogging down, you'd be working in Matlab anyways. I've always been mystified as to why Excel gets considered as a "benchmark". Any CPU out in the last 5 years is plenty for any and all office "productivity" that happens in Excel. IMO.

Honestly though, reality is very different.

Both my wife and I worked on "serious excel" spreadsheets at TI and even though we should have been supported with a matlab license and install the IT dept wasn't about to roll that out to us.

In reality you work with the tools you have, not the tools you want, and sometimes you really do only have a hammer. Corporate america is not keen on licensing software beyond the basics, and for much of corporate america the basics pretty much ends with MS Office.

I work for myself now (business owner) so I use Origin and Mathematica because I can prioritize investing in my own productivity, but my wife works for one of the largest global chemical gas suppliers and she has to manage stuff in Excel spreadsheets while attempting database queries that can literally take up to 3 hrs at times before her "serious Excel" calcs are finished.

It sucks, but the IT dept is deadset against the authorizing the expenditure for licensing the tools needed to do the job more efficiently. They look at it at as "if we buy one, we'll need to buy 10,000 licenses for everyone else in that job capacity, or I can NOT buy those licenses and instead claim it as a "cost savings" I personally made for the company and then bag me a year-end bonus for myself"...its a very self-defeating system but I've seen it at every company I've worked for or worked with.

(not bagging on IT either, its not their fault their personal career path and compensation incentives are geared towards doing stupidly short-sighted stuff when it comes to enabling employee productivity, it all comes from the top when budgets and strategies are set)

 

[Idontcare]

I'd like to see what said patch does for Phenom2 X6's too...I want to say I saw some benches of people running it on a Phenom2 system and performance improved even more than it did for Bulldozer...which made the whole thing look a bit shady to me...

What does that patch do, anyway?

2) What would it do to a non-HT cpu?

Any "patch" or code that in any ways changes the scheduler's propensity to migrate threads (core hopping, etc) will result in a performance boost just from the lessened degree of cache thrashing and so on that goes on for all multi-core cpus.

CMT and HT microarchitecures ought to benefit even more so than your CMP microarchitectures, but it really should be a tide that raises all boats.

 

[Magic Carpet]

Any "patch" or code that in any ways changes the scheduler's propensity to migrate threads (core hopping, etc) will result in a performance boost just from the lessened degree of cache thrashing and so on that goes on for all multi-core cpus.

CMT and HT microarchitecures ought to benefit even more so than your CMP microarchitectures, but it really should be a tide that raises all boats.

EDIT:
It's pointless to discuss the patch when Microsoft pulled it.

 

[Idontcare]

EDIT:
It's pointless to discuss the patch when Microsoft pulled it.

LOL, ninja-edit there :ninja:

My understanding is that Microsoft and Intel coordinated the scheduler and OS to be logical vs. physical core "aware" and as such the scheduler can use the core/thread topology information to coordinate threads in an advantageous way.

The issue with Bulldozer is that you want the threads piled onto modules first so as to increase the chances of the clockspeed of the module itself being turbo-boosted.

So rather than distributing one thread per module (less resource sharing within the module, should be faster) and operating all modules at the stock clock, the idea is to intentionally gang the threads on a given module (reducing performance because of resource sharing) but enabling the turbo-clocks to kick in and, hopefully, elevating performance above and beyond the penalties that come from CMT style resource sharing.

 

[Magic Carpet]

LOL, ninja-edit there :ninja:

My understanding is that Microsoft and Intel coordinated the scheduler and OS to be logical vs. physical core "aware" and as such the scheduler can use the core/thread topology information to coordinate threads in an advantageous way.

The issue with Bulldozer is that you want the threads piled onto modules first so as to increase the chances of the clockspeed of the module itself being turbo-boosted.

So rather than distributing one thread per module (less resource sharing within the module, should be faster) and operating all modules at the stock clock, the idea is to intentionally gang the threads on a given module (reducing performance because of resource sharing) but enabling the turbo-clocks to kick in and, hopefully, elevating performance above and beyond the penalties that come from CMT style resource sharing.

Thanks for the time writing this all up. Appreciate that. Wish I had as much patience on the last day of 2011

I had initially thought that patch enabled core parking on Windows 7 but after installing it... my intentionally registry entries had been left unchanged, so there must be something else. Core parking makes use of Turbo mode as much as possible so this is a must for all Turbo enabled / power gating processors. I suppose Microsoft is still fiddling with it finding out best settings. On paper it looks very easy, yet they pulled the patch with some vague explanation:

"there are actually two updates needed for AMD Bulldozer CPU architecture. Microsoft posted just the first patch and we do not believe users would benefit in any way from it. The patch was originally scheduled for the first quarter 2012 and then the users will see tangible performance benefits when using Windows 7 and Windows Server 2008 R2 operating systems."

And what about Windows 8? Essentially it features the same Task Scheduler but w/ core parking enabled out of the box. No performance issues?

Amusing :whiste:

EDIT:
nevermind, Windows 8 hasn't hit the shelves yet :ninja:

 

[soccerballtux]

LOL, ninja-edit there :ninja:

My understanding is that Microsoft and Intel coordinated the scheduler and OS to be logical vs. physical core "aware" and as such the scheduler can use the core/thread topology information to coordinate threads in an advantageous way.

The issue with Bulldozer is that you want the threads piled onto modules first so as to increase the chances of the clockspeed of the module itself being turbo-boosted.

So rather than distributing one thread per module (less resource sharing within the module, should be faster) and operating all modules at the stock clock, the idea is to intentionally gang the threads on a given module (reducing performance because of resource sharing) but enabling the turbo-clocks to kick in and, hopefully, elevating performance above and beyond the penalties that come from CMT style resource sharing.

wait, I thought that point was you wanted them spread out so that you don't have resource contention over the 256-bit FPU?

 

[soccerballtux]

Honestly though, reality is very different.

Both my wife and I worked on "serious excel" spreadsheets at TI and even though we should have been supported with a matlab license and install the IT dept wasn't about to roll that out to us.

In reality you work with the tools you have, not the tools you want, and sometimes you really do only have a hammer. Corporate america is not keen on licensing software beyond the basics, and for much of corporate america the basics pretty much ends with MS Office.

I work for myself now (business owner) so I use Origin and Mathematica because I can prioritize investing in my own productivity, but my wife works for one of the largest global chemical gas suppliers and she has to manage stuff in Excel spreadsheets while attempting database queries that can literally take up to 3 hrs at times before her "serious Excel" calcs are finished.

It sucks, but the IT dept is deadset against the authorizing the expenditure for licensing the tools needed to do the job more efficiently. They look at it at as "if we buy one, we'll need to buy 10,000 licenses for everyone else in that job capacity, or I can NOT buy those licenses and instead claim it as a "cost savings" I personally made for the company and then bag me a year-end bonus for myself"...its a very self-defeating system but I've seen it at every company I've worked for or worked with.

(not bagging on IT either, its not their fault their personal career path and compensation incentives are geared towards doing stupidly short-sighted stuff when it comes to enabling employee productivity, it all comes from the top when budgets and strategies are set)

My faith in my future with corporate is now 100% more jaded.

What work do you do? I didn't think ATF paid its mods?

 

[soccerballtux]

Any "patch" or code that in any ways changes the scheduler's propensity to migrate threads (core hopping, etc) will result in a performance boost just from the lessened degree of cache thrashing and so on that goes on for all multi-core cpus.

CMT and HT microarchitecures ought to benefit even more so than your CMP microarchitectures, but it really should be a tide that raises all boats.

Hm hadn't thought about the core hopping.
I was of the opinion that they added in some performance stuff they've thought of over the last few years and were saving for Windows 8, to help poor AMD out cause they certainly need it. Perhaps that's a bit given to conspiracy...

 

[LoneNinja]

wait, I thought that point was you wanted them spread out so that you don't have resource contention over the 256-bit FPU?

Spreading the threads out prevents the modules from being able to turbo since turbo core works by module rather than by core. It seems with aggressive turbo speeds, the increased clock speed improved performance better than preventing resource sharing within a module. I wonder how this will effect performance when turbo core is disabled as it'll increase resource sharing on the module level. Either way I don't believe the 10-20% performance increase rumors that are all over the web right now, there is no evidence to back any claims exceeding 10% best case scenario.