The best remedy for bad news is often humour and I found this video on Hitler's response to the Bulldozer benchmarks really funny - http://www.youtube.com/watch?v=SArxcnpXStE.

Well as I'm still sat on my arse waiting for my SB EX boards to show up I thought I'd do something mildly constructive and sort out a Bulldozer spec.

My inital aim was to run a AMD cpu at 4.5GHZ as that's going to be my target for the Sandy Extremes next week. I'm using the FX8120 for benching for a number of reasons, the main one being that it seems to be the same chip as the FX8150 with a lower clock at stock. So I thought bang per buck wise if I'm going to run these things at 4.5GHZ we'll save £50 on the price and ramp up the cheaper chip would offer more value.

So started with a Gigabyte board which set up fine but hung everytime I put Cubase underload.
Moved onto an Asus that I'm still cursing about. Absolutely maddness trying to overclock it, with the normal tweaks causing it to do nothing but hang on every other reboot.

So MSI 970A C45. Plugged it in whacked in the settings and got into windows.... froze in testing, so rebooted ramped up the NB/SB voltages and voila a working bench. I could probably refine it for heat gains but as a rough and ready bench it'll do.

So open up Cubase with a DAWBench template and crack on :

64 - 98
128 -104
256 -144
512 -144

I retested each latency 3 times total with a reboot between each. The last two scores look well wrong unless the's some crazy bottleneck somewhere.

So I thought I'd de-clock it back to stock in case it was... you know... me being a numpty.

64 -54
128 -86
256 -96
512 -97

Apparently not then.

If anyone can explain those 256/512 scores I'm all ears.

However that aside I'm shocked. I honestly thought these would be the in the ballpark as the i7 950's and they are not remotely even close.



(I've not re-edited the graph yet, I'll do that when the SBEX scores come in, it's just there for reference along with the scores above)

In fact it's even beaten at stock by the older Hexcore Phanom 1090T chip from the last generation refresh, which whilst I'd seen it mentioned it would happen in some benchmarks in some other published test sessions, I'm still quite shocked to see it here.

So all in all? Poor value for money for audio. At stock it's no better than a Q6600 system for considerably more cash and overclocked it's still outperformed by a OC'd 2600k by about 250% for only a few hundred quid more.

TL;DR : It sucks more than even I suspected.
TL;DR for TL;DR : Meh.

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

Sounds like the chip might be at fault - take it back and exchange it for a 8150 - Often felt that the lower clocked / budget versions were done that way because they did not meet with top spec, tend to notice that when the production run starts out - then once things are going smoothly you can bargain hunt with opening cores etc - but for now take it back and get a 8150 - with water cooler would be nice - thought the opti OC for these beasties was 4.6

They have now overclocked this chip to 8.5 and are hoping to hit warp factor 9 with a change to the B3

Must away, have a AC30 that I've just fixed and it needs a damn good thrashing... or should that be trashing, either way I dare say it will be loud and enjoyable The joys of testing equipment

--------------------
www.dragonlogos.co.uk

Quote DragonLogos:

---

Sounds like the chip might be at fault - take it back and exchange it for a 8150 - Often felt that the lower clocked / budget versions were done that way because they did not meet with top spec, tend to notice that when the production run starts out - then once things are going smoothly you can bargain hunt with opening cores etc - but for now take it back and get a 8150 - with water cooler would be nice - thought the opti OC for these beasties was 4.6

---

Could you repeat that in English please?

Quote DragonLogos:

---

Sounds like the chip might be at fault - take it back and exchange it for a 8150 - Often felt that the lower clocked / budget versions were done that way because they did not meet with top spec, tend to notice that when the production run starts out - then once things are going smoothly you can bargain hunt with opening cores etc - but for now take it back and get a 8150 - with water cooler would be nice - thought the opti OC for these beasties was 4.6

---

Yeah, as it's a Hexcore through I was going put it up against my Hexcore Intel and that one has an optimum of 4.5Ghz on the SBEX platform. I don't expect the 8150 to get a better score as it appears to be the same chip binned differently but I'm willing to give it a go whenever AMD decides to acturly import some into England. I did suspect a faulty chip/board but I've tried multiples of each before recording the results and they all pulled the same figures.

The cooler I was using for benching was a Thermalright Archon so it's not a cooling problem as those things bench better than the Corsair H50 in group tests.

Quote DragonLogos:

---

They have now overclocked this chip to 8.5 and are hoping to hit warp factor 9 with a change to the B3

---

Yeah, I'm aware of that but Ghz mean nothing if the rest of the achitechture doesn't take advantage of it. You can stick a Porche engine in a 1973 Skoda body but it doesn't mean it's going to hit 200Mph and still be driveable!

Quote DragonLogos:

---

Must away, have a AC30 that I've just fixed and it needs a damn good thrashing... or should that be trashing, either way I dare say it will be loud and enjoyable The joys of testing equipment

---

That acturly sounds far more fun than another afternoon of benchmarking!

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

Quote Pete Kaine:

---

If anyone can explain those 256/512 scores I'm all ears.

---

Hey Pete,

Using the reference audio interface over the years, on this particular bench I found there was little scaling above 256 with the later chips, but the AMD numbers are suggesting that the architecture -whether the execution is Cache/Core bound in this particular test - is totally levelling out in this style of test.

It may be interesting to test with DAWbench VI and see if that shows the same behaviour.

In the end it doesn't really matter, BD is not going to miraculously change its spots !

V.C

--------------------
AAVIM Technology
DAWbench.com

> Sounds like the chip might be at fault

Yep, it's DEFINITELY the chip that's at fault...

There seems to be a lot of confusion in this thread. Bulldozer is a server architecture. Using it on the desktop is really a desperation move on the part of AMD, because they don't have the R&D resources to do anything else.

Server workloads tend to be mainly I/O bound and integer based. Most of the CPU's cycles are wasted waiting for a disk to seek, or worse, for packets over the network. Because of this, you simply want the maximum number of threads. Bulldozer is a relatively intelligent way of doing this. Although each 'core' is really only an integer unit with some L1 cache, you get most of the integer performance of a full core for a fraction of the silicon. Shared cache, long pipelines and poor scheduling increase the amount of lost cycles due to cache misses, but they're negligible compared to cycles lost waiting for I/O.

In spite of a much cruder basic microarchitecture, Bulldozer-based Opterons provide more threads for less money than comparable Xeons. AMD are still well behind when it comes to heat and power, but there are plenty of applications where an Opteron makes sense. Several new high performance clusters have been commissioned based on Bulldozer chips.

If you're paying attention, you'll have realised that the Bulldozer is optimised for a workload that is more or less the exact opposite of a modern DAW. We want to maximise floating point performance, and minimise wasted cycles to keep latency down. Unsurprisingly, the Bulldozer architecture does an absolutely miserable job at digital audio work - it only has four of the cores we really need and implements them inefficiently. Same goes for games, same goes for all sorts of desktop workloads - if you're not bound by integer performance, those extra cores are completely useless. You can see this clearly in the bipolar benchmarks - very quick for some tasks, miserably slow for others.

Quote TAFKAT:

---

Quote Pete Kaine:

---

If anyone can explain those 256/512 scores I'm all ears.

---

Hey Pete,

Using the reference audio interface over the years, on this particular bench I found there was little scaling above 256 with the later chips, but the AMD numbers are suggesting that the architecture -whether the execution is Cache/Core bound in this particular test - is totally levelling out in this style of test.

It may be interesting to test with DAWbench VI and see if that shows the same behaviour.

---

Hi Vin,

Tom pulled me on that yesterday and said the same thing. I seem to recall the same result in the Phenom testing now, but still it suprises me. DB VI test does seem like it might be fitting to see how it translates, and when I someone will pony me up a 8150 I may have to slap that on to see how those higher latencys translate.

Quote Dishpan:

---

> Sounds like the chip might be at fault

Yep, it's DEFINITELY the chip that's at fault...

---

ಠ_ಠ

Can't tell if serious, or if my sarcasam detector is in need of new batterys...

Anyhow on a related note Scott ADK posted up his BD benchmarks yesterday over at KVR in response to a thread I posted this lot in and they were pretty much identical, so I'm going to rule out muppetry, broken chips & cosmic flux at this point.

Quote uphillbothways:

---

There seems to be a lot of confusion in this thread. Bulldozer is a server architecture. Using it on the desktop is really a desperation move on the part of AMD, because they don't have the R&D resources to do anything else.

---

In the same respect the Intel cpu range now is pretty much interchangeable with Xeon's being able to run on desktop boards with only the data exchange pipelines being being disabled. We're fast approaching the point where people cease to differentiate between server and desktop... as such AMD's move to make the most out of their very limited R&D budget would make sense.

Quote uphillbothways:

---

Server workloads tend to be mainly I/O bound and integer based. Most of the CPU's cycles are wasted waiting for a disk to seek, or worse, for packets over the network. Because of this, you simply want the maximum number of threads. Bulldozer is a relatively intelligent way of doing this. Although each 'core' is really only an integer unit with some L1 cache, you get most of the integer performance of a full core for a fraction of the silicon. Shared cache, long pipelines and poor scheduling increase the amount of lost cycles due to cache misses, but they're negligible compared to cycles lost waiting for I/O.

---

This is the performance problem on the desktop and it doesn't really add up. The bottleneck seems to be that physical cache being split between multiple cores and it's pretty much what is crippling these chips. For years AMD's been stood behind this point of "Physical cores = Faster Parallel Processing" which means you want to buy 8 slower AMD cores over 4 Faster Intel cores with hyper threading because you don't want to be using any of that virtual processing junk....

Except due to the cache throttling the bandwidth that virtual threading is looking like the far superior solution at this point in time.

Quote uphillbothways:

---

In spite of a much cruder basic microarchitecture, Bulldozer-based Opterons provide more threads for less money than comparable Xeons. AMD are still well behind when it comes to heat and power, but there are plenty of applications where an Opteron makes sense. Several new high performance clusters have been commissioned based on Bulldozer chips.

---

Low powered clusters for mail searvers, ftp's, database clients all make perfect sense for these chips in much the same way I'd spec an AMD if someone came to me with £400 for a SOHO machine. Intel is still the prefered option when going for raw power through (data processing rather than data servage) I tend to find.

Quote:

---

If you're paying attention, you'll have realised that the Bulldozer is optimised for a workload that is more or less the exact opposite of a modern DAW.

---

Although you'd have thought that the would be a way to optimize load balancing further than we already have to ensure as much of the available clock cycles were used by the competing plug in's. You'd have thought that DAW work with a few dozen different bits of code being executed constantly would have benefited heavily from this type of parallel processing more than most other software types.

Quote:

---

We want to maximise floating point performance, and minimise wasted cycles to keep latency down. Unsurprisingly, the Bulldozer architecture does an absolutely miserable job at digital audio work - it only has four of the cores we really need and implements them inefficiently. Same goes for games, same goes for all sorts of desktop workloads - if you're not bound by integer performance, those extra cores are completely useless. You can see this clearly in the bipolar benchmarks - very quick for some tasks, miserably slow for others.

---

Yeah, agreed. Just want to make it clear that I didn't set out to question any of the above on here originally. My only concern was "Should I be getting behind this kit as a good value solution for my end users" and the answer was no.

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

I am more interested in application specific benefits. Specifically the use of AVX FP SIMD instruction support in Sonar X1.

An AMD Bulldozer sees a huge advantage over non AVX CPUs with 24 bit audio in some scenarios (quite the opposite with 32 float though). See the AVX blurb here: http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,30 43-5.html

I will be getting my FX-8150 tonight, but wont be able to test it until next week (recording all weekend, paying customers no less).

I currently run a Phenom X6 1090T on a GIGABYTE GA-990FXA-UD7 with 16GB of Corsair 1866 DDR3 RAM,A 256GB Crucial M4 SSD and a 4x500GB Raid 0 on dedicated high end raid controller with an MSI GTX 570 handling video (Great for The Mercury playback engine in Premiere).

I wont have much time for a very thorough benchmark session. I will do a 2 runs at 256ms latency with Sonar X1 32bit and 64bit on the 1090, then 2 more with the 8150. I hope that AVX makes my upgrade worth it as I use 24/48 and Sonar's 64bit double precision engine. In any case, the 1090 will make a good upgrade for the tracking PC, I've hit the headroom of the X4 965 which currently resides there.

> Can't tell if serious, or if my sarcasam detector is in need of new batterys...

Definitely change those batteries :-)

Quote trailmixxx:

---

I am more interested in application specific benefits. Specifically the use of AVX FP SIMD instruction support in Sonar X1.

An AMD Bulldozer sees a huge advantage over non AVX CPUs with 24 bit audio in some scenarios (quite the opposite with 32 float though). See the AVX blurb here: http://www.tomshardware.com/reviews/fx-8150-zambezi-bulldozer-990fx,30 43-5.html

I will be getting my FX-8150 tonight, but wont be able to test it until next week (recording all weekend, paying customers no less).

I currently run a Phenom X6 1090T on a GIGABYTE GA-990FXA-UD7 with 16GB of Corsair 1866 DDR3 RAM,A 256GB Crucial M4 SSD and a 4x500GB Raid 0 on dedicated high end raid controller with an MSI GTX 570 handling video (Great for The Mercury playback engine in Premiere).

I wont have much time for a very thorough benchmark session. I will do a 2 runs at 256ms latency with Sonar X1 32bit and 64bit on the 1090, then 2 more with the 8150. I hope that AVX makes my upgrade worth it as I use 24/48 and Sonar's 64bit double precision engine. In any case, the 1090 will make a good upgrade for the tracking PC, I've hit the headroom of the X4 965 which currently resides there.

---

Those links dont work ... and the bulldozer kinda sucks ... but good luck.

broken link... lets have a look, ahhh there is a space in the address - easy enough to spot and fix

Just pop it into a box to stop the overflow

The 8150 Link

this is were it went wrong

bulldozer-990fx,3043-5.html

bulldozer-990fx,30 43-5.html


--------------------
www.dragonlogos.co.uk

Quote:

---

don't expect the 8150 to get a better score as it appears to be the same chip binned differently but I'm willing to give it a go whenever AMD decides to acturly import some into England

---

Was supposed to get a 8150 three weeks ago but there is no stock to be had, thought maybe the overclockers are buying them out or something

--------------------
www.dragonlogos.co.uk

There has been a fair amount of talk about how much better it might work out with a OS that handles multi-core such as Windows 8 - Multi-core seems to be the way that things are going, and for music applications it makes good sense... does not look great on benchmarks, put the following into a search engine

i7 2600k vs 980x

There are lots of questions raised about the how the six core 980X seems to miss out

Will things get better with the right OS - well you don't have to wait for Windows 8 to pop along to answer that question, Linux can give a fair idea now

AMD FX-8150 Bulldozer On Ubuntu Linux

Quote:

---

With this just being the first article of a series looking at the AMD FX-8150 "Bulldozer" under Linux, more conclusions will be drawn later. However, as these results show, under Linux the AMD FX-8150 is a competitive product to the Intel Core i5 2500K when dealing with multi-threaded workloads. For single-threaded work and other select tasks, the Bulldozer performance is disappointing.

Fortunately, for Linux users, most open-source software is well multi-threaded. If you are running Gentoo, Arch, or just doing a lot of compiling in general, the AMD Bulldozer CPUs should be able to prove their value very well. Beyond that, with open-source software that you may be building, GCC and Open64 already have Bulldozer (version 1) optimizations in place

---

--------------------
www.dragonlogos.co.uk

Quote DragonLogos:

---

does not look great on benchmarks, put the following into a search engine

i7 2600k vs 980x

There are lots of questions raised about the how the six core 980X seems to miss out

---

The 980x through is from the previous generation of chips, and advances in architecture both on and off the chip explain the 2600k performance. The extremes launched this week are the same generation as the 2600 range so that explains it to some extent. Agreed that 980x isn't fully used by a lot of software, in the same way that both Bulldozer and the SBEX's are likely not to be and once more we're waiting for programmers to play catch up... always the case through!

Quote DragonLogos:

---

With this just being the first article of a series looking at the AMD FX-8150 "Bulldozer" under Linux, more conclusions will be drawn later. However, as these results show, under Linux the AMD FX-8150 is a competitive product to the Intel Core i5 2500K when dealing with multi-threaded workloads. For single-threaded work and other select tasks, the Bulldozer performance is disappointing.

---

The most disappointing thing there is the comment that it's a competitive product to the 2500k. It's the same under windows as well... and really shouldn't be. It should really be up there with the 2600k if it's going to hit the market in the postion it needs too. The 2500k chip is viewed as the gamers choice which due to it's multicore nature the Bulldozer clearly isn't.

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

Quote DragonLogos:

---

There has been a fair amount of talk about how much better it might work out with a OS that handles multi-core such as Windows 8 - Multi-core seems to be the way that things are going, and for music applications it makes good sense... does not look great on benchmarks, put the following into a search engine

i7 2600k vs 980x

There are lots of questions raised about the how the six core 980X seems to miss out

Will things get better with the right OS - well you don't have to wait for Windows 8 to pop along to answer that question, Linux can give a fair idea now

AMD FX-8150 Bulldozer On Ubuntu Linux

Quote:

---

With this just being the first article of a series looking at the AMD FX-8150 "Bulldozer" under Linux, more conclusions will be drawn later. However, as these results show, under Linux the AMD FX-8150 is a competitive product to the Intel Core i5 2500K when dealing with multi-threaded workloads. For single-threaded work and other select tasks, the Bulldozer performance is disappointing.

Fortunately, for Linux users, most open-source software is well multi-threaded. If you are running Gentoo, Arch, or just doing a lot of compiling in general, the AMD Bulldozer CPUs should be able to prove their value very well. Beyond that, with open-source software that you may be building, GCC and Open64 already have Bulldozer (version 1) optimizations in place

---

---

Hey that's great Dragon, so its absolute crap in single threaded apps, and in the best case scenario on some obscure operating system it runs almost as fast as the budget low end intel i5 ...

1400 thousand AMD workers will agree its been a total disaster

http://www.forbes.com/2011/11/04/amd-announces-1400-job-cuts-marketnewsvid eo.html

PCs are over anyway, its going to end up as a fight between Intel and ARM a few years down the road...

Any fool can criticize, condemn and complain and most fools do - Benjamin Franklin

--------------------
www.dragonlogos.co.uk

We don't like their sound; also Mr Epstein guitar music is on the way out – Decca's rejection The Beatles

--------------------
www.dragonlogos.co.uk

All truth passes through three stages. First, it is ridiculed. Second, it is violently opposed. Third, it is accepted as being self-evident

Talent hits a target no one else can hit; Genius hits a target no one else can see

Arthur Schopenhauer

--------------------
www.dragonlogos.co.uk

Never give in, never, never, never, never. In anything great or small, large or petty, never give in except to convictions of honour and good sense. Never yield to force; never yield to the apparently overwhelming might of the enemy - Sir Winston Churchill

--------------------
www.dragonlogos.co.uk

Quote:

---

The 980x through is from the previous generation of chips, and advances in architecture both on and off the chip explain the 2600k performance

---

I wonder if anyone could give a reason as to why the 980X still has such a high retail price then? - perhaps there is no competition in Intel vs Intel

--------------------
www.dragonlogos.co.uk

Quote DragonLogos:

---

Never give in, never, never, never, never. In anything great or small, large or petty, never give in except to convictions of honour and good sense. Never yield to force; never yield to the apparently overwhelming might of the enemy - Sir Winston Churchill

---

Dragon you really have gone batshit crazy this time. Its time for AMD fanboys to be quiet. Look at the results of the DAW performance tests. Its over.

Quote DragonLogos:

---

I wonder if anyone could give a reason as to why the 980X still has such a high retail price then? - perhaps there is no competition in Intel vs Intel

---

Intel stubboness more like. High end chips never go below certain price points they just drop off the retail list.

High end entry - £200 - £300
Lower Mid £400 - £500
Upper Mid £550 - £700
top £800 - £1000

During the lift cycle they don't drop below the lower points, they just replace them to make them appear to be better value (i.e. 980X - 990X) by cite'in bin improvements or minor revision changes.

Always been this way even when AMD was in front.

Before the 9 series I always told people to ignore the extremes due to inherent poor value. Even now I fail to be convinced with anything above the £500 point normally, and then only in a few exceptional situations depending upon the end user requirements.

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

So as I sit here on my 1055T based system, with a relatively modern AM3 but not AM3+ mobo, what should I be aiming for on my next build, guys?

Presumably an Intel but which one is oging to gie me the best bang for the buck including a mobo that does pci, pci-e and has decent enough onboard graphics for DAW use and of course sata 3 & USB 3 (just in case)

I'm not kidding, I really would like suggestions for my next self-build.

--------------------
Me? But I`m such a loveable old bugger!

2600 chip on a 1155 board of some description.... whichever one ticks all of your requirements.

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

Thanks, Pete.
To be honest I am more than happy with the performance I am getting from my current AMD system but it IS nice to have an idea of where to go next.

I am getting silly low latency and very smooth performance using reaper and an RME card at present, so I suspect it will be a while before I need anything faster.

Might have to spring for something better than the onboard GFX on my M4A89GTD-Pro-USB3 board though...

And in the meantime I can be entertained by all you bleeding-edge nutters stretching the envelope for me ahead of time!




Blimey! Now I remember why I bought what I did.
Best price I could find on an i7 2600 3.4 was about the same as I paid for my AND cpu AND the latest (at the time) mobo for it.

--------------------
Me? But I`m such a loveable old bugger!

Edited by IvanSC (24/11/11 01:29 PM)

Quote:

---

So started with a Gigabyte board which set up fine but hung everytime I put Cubase underload.

---

Peter I wonder if you could perhaps do a service for me.... and try a Benchmark with Fruity Loops Studio, this might put the ICC (Intel Compiler) issue to rest, FL Studio uses Delphi

Knee deep in Video footage this side

--------------------
www.dragonlogos.co.uk

Been discussing this with Vin and having done some more research on it the certainly could be some skew although seeing as the vast majority uses the Intel compiler anyhow and AMD hasn't updated theirs in years it's a bit of a moot point.

However in an attempt to get to the bottom of it all we've had a chat with Justin over at Cockos about it and he's recompiled reacomp for us using a whole host of different compiler revisions as the plug in build should show up the differences. He sent me over a half dozen a few weeks back, but the initial batch wouldn't run on my configuration, so another set of recompiles later I have a working folder to get through. I'm on it today so should have some findings in the next 24hrs.

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

Thanks Peter - will be interesting to see, we only managed to get two four core BDs and doubt if I will have time to even get near one as there is not a Mobo for it (have yet to say what I want and they have yet to say what there is) The Op that has the 4x is raving about it (in a good way)but what for I know not as this is coming in via a third party

--------------------
www.dragonlogos.co.uk

Quote DragonLogos:

---

Peter I wonder if you could perhaps do a service for me.... and try a Benchmark with Fruity Loops Studio, this might put the ICC (Intel Compiler) issue to rest, FL Studio uses Delphi

---

Allow me to step in here,

Why in Gods name would you suggest FL for Multiprocessor scaling test when the developer , Didier Dambrin, still maintains the same stance he has held for years , that multicore processors are not beneficial for DAW usage.

Name one other DAW host that uses Delphi , or even one major plugin or virtual instrument developer for that matter !

We have a whole range of compiled versions , some of the non ICC's are causing huge issues in Cubase , so even if one of the alternate compilers were to deliver a performance boost, its a bloody minefield how they will react in the various DAW environments. Good luck getting the developers to head into those waters.

I already know what the results are , but I'll let Pete take it from here as he is in the front line doing the testing on the BD , but needless to say the conspiracy theorists are not going to be happy... :-)

V.C

--------------------
AAVIM Technology
DAWbench.com

Excuse the delay, but I did have to carry out some additional testing over my inital set.

So I did comparisons with 10 different builds done with various compiler options as supplied by Justin over at Cockos. The more recent compiler builds done this year underperformed in each test so he passed me the 2008 builds using the older versions of the various compilers.

The one Justin himself suspected might come out well was the one that won the group test group performance wise and gave around a 15% - 20% performance improvement over the original DAWBench standard revision on the AMD rig.

DAWBench Build

64 - 76
128 - 85
256 - 93

Vc2005 SSE2

64 - 96
128 - 103
256 - 112

So the is some legs in it. However the Vc2005 SSE2 build would not run in Cubase (all testing had to be redone in reaper) and lost all performance advantages on a Intel based setup.

So a couple of things to take away from this.

Yes another compiler may/does favour AMD builds. But... (and it's a huge one) unless plugin builders are going to use the compiler in question it's a moot point. If they chose to go with it they would have to do Intel and AMD specific builds or they alienate both Intel users and Cubase users.

All of the other builds (the Intel bias'd one's included) showed at least a 20% performance drop against the standard plug revision so the current packaged one would seem to still be both the most compatible solution as well as the best performing.

What we have proved is that perhaps it would be interesting if developers did offer AMD optimized builds of plugs as well but that would take a harrasment from the public so start the e-mail campaigns guys.

The other point of note is that a 2500 based Intel machine which can be purchased for a few hundred quid less still beats the optimized AMD score by another 40% across the board leaving the BD based machines still poor value for money performance wise for pure audio usage.

--------------------
ScanProAudio & 3XS Audio Systems
ScanProAudio Blog

Quote:

---

standard plug revision

---

intel compiler
hosts:
Cubase/Nuendo - icc 12(AVX for Intel, SSE2 for AMD),Reaper - icc 10.1( SSE 4.2 for Intel, x87 for АMD).
plugins:
Softube (SSE 4.2 for Intel, SSE2 for АMD)
SSL (SSE2 for AMD)
Voxengo (SSE2 for AMD)
XLN Audio (SSE2)
Camel Audio (SSE2)
BBE
Cockos reaplugins (x87 for АMD)
Non intel:
big list :
Studio One,Sonar - VS2010 compiler (AVX optimizanion for Intel and AMD)...,
>80% PC VST plugins - Microsoft compiler (oldVS6,VS2005 VS2008,VS2010) and <20% - icc,delphi,mingw etc.
Microsoft compiler bad,but very popular

Pete summarised pretty well but let me add some final points.

This was a huge undertaking by Justin Frankel who went above and beyond to allow us to get an angle on what variables were in play here , 10 new compiled versions above the original, across 2 specific builds of the plugin.

As Pete noted , the 2011 builds of the plugins all failed to get anywhere near the original build results, so simply changing the compiler to one not using the ICC does not in anyway guarantee better performance on the AMD chips. So to be clear, using the specific alternate compiler that showed the better result on the 2008 build of the plugin, did not translate on the 2011 build , both which totally collapsed in Cubase but not in Reaper, which opens a huge can of worms for developers.

Now even with all of the cards stacked in favour of the AMD , the new architecture was still demolished comparatively by a 3 year old i7 930, not to mention the current Intel chips , so no amount of polishing will change the fact that the AMD Bulldozers have fallen into a large hole in regards to DAW performance.

We went thru this exercise because we were called out at another forum by a member of the AMDian lobby who basically accused me of purposely loading the deck in favour of Intel , which is ridiculous as by the time the DAWbench Universal Suite was developed the Intel v AMD platform wars were well and truly done and dusted after core 2 came out. Unlike others who choose to sit back and fire shots from a comfortable distance and froth on about Intels anti competitive conspiracy's, we knuckled down and applied valuable time and energy to get a clearer answer.

In closing, I am still working with Justin on a final build for the 2012 Suite as we are still navigating another unrelated issue with the plugin in some DAW hosts , which does not effect performance but can cause crashing with the GUI being called up. Will I be using an alternate compiler, well No , its a total and utter waste of time to compromise the stability of the test and the validity of the past results to cater for a small but overly vocal minority.

Placing that all into perspective , what developer would compromise their code base to try and ring a small % performance improvement while at the same time open themselves up to other variables , for a chip that is failing miserably performance wise ?

Quick edit.

Just missed that my accuser has navigated to this thread as well... :-)

@ q_h

Simply using ICC does not guarantee better performance on the AMD unless you have an alternate ICC build to prove other wise , this is something we discovered in our testing, also with the major Hosts using the ICC compiler , thats another huge hurdle to navigate.

Its interesting that you note SONAR and StudioOne using an alternate compiler, SONAR is the worst DAW host I have tested over the years at multiprocessor scaling, across a whole range of plugins. StudioOne, I already knew that one as I am in contact witht he developers, well that performs better but of course that may now vary depending on which plugin is used.. :-(

So 80% of VST plugins use a better compiler you say, well name the majors and I'll weigh up some candidates which are worth looking at for an added session to the suite.


V.C

Edited by TAFKAT (16/12/11 08:51 PM)

Just noticed a typo.. :-(

@ q_h,

That should read..

"Simply *not* using ICC does not guarantee better performance on the AMD unless you have an alternate ICC build to prove other wise , this is something we discovered in our testing, also with the major Hosts using the ICC compiler , thats another huge hurdle to navigate."

Man I hate that time bomb on editing posts here...



--------------------
AAVIM Technology
DAWbench.com

TAFKAT,

Quote:

---

So 80% of VST plugins use a better compiler you say, well name the majors and I'll weigh up some candidates which are worth looking at for an added session to the suite.

---

Nebula,AcquaVox
freeverb3 bundle -freeverb3_vst-slow-AVX-doubleprecision

-
http://support.microsoft.com/kb/2592546/en-us



more info:
http://www.brightsideofnews.com/news/2011/12/16/microsoft-releases-amd-bul ldozer-patch-by-mistake2c-incomplete-download.aspx

I have already given Aquavox a runup , it is not suitable as it will not work as an incremental loading plugin in the same manner as the other reference plugins.

Nubula is great but it is not what I consider a conventional plugin platform and requires a different style of benching methodology which I am already in discussion with the developers of Nebula to try and jointly develop for 2012.

I'll checkout Freeverb.

Got any others ?

BTW: What does Cinebench have to do with audio application ?

Lets stay on topic eh !

Are you trying to imply that all audio applications are somehow skewed against Bulldozer because they do not recognise the 4 x Dualcore modules as 8 physical cores, if so, lets see the evidence.

Re the Windows patch for the task scheduler, lets see what it actually achieves , it will need to make an improvement in the vicinity of several hundred % to get anywhere near the same ballpark as the current Intel architecture in Audio application. If it improves the current situation even enough so that the BD architecture matches the last AMD , well then that is at least something, but nothing that will change the overall comparative performance significantly enough when placed in context.

I have some further insight why the current architecture is falling over itself in audio application even worse then the previous AMD architecture , directly from a DSP developer , so why don't you enlighten us on why you believe that is so and we can compare notes... ;-)

V.

--------------------
AAVIM Technology
DAWbench.com

Quote q_h:

---

TAFKAT,

Quote:

---

So 80% of VST plugins use a better compiler you say, well name the majors and I'll weigh up some candidates which are worth looking at for an added session to the suite.

---

Nebula,AcquaVox
freeverb3 bundle -freeverb3_vst-slow-AVX-doubleprecision

-
http://support.microsoft.com/kb/2592546/en-us



more info:
http://www.brightsideofnews.com/news/2011/12/16/microsoft-releases-amd-bul ldozer-patch-by-mistake2c-incomplete-download.aspx

---

That's great, now we have established AMD Bulldozers are an extremely poor choice for Windows 7 and DAW usage. As AMD are drastically cutting back their workforce and Bulldozer was the great white hope, it seems that all meaningful competition to desktop Intel chips is currently dead at this time.

Quote:

---

"Simply *not* using ICC does not guarantee better performance on the AMD unless you have an alternate ICC build to prove other wise

---

Surely you mean a alternative compiler

--------------------
www.dragonlogos.co.uk

Thanks Peter, have a few results and findings this side just need to sit down for a while join the dots and pool the paper work, with a bit of luck should have a few days off to do that

Quote:

---

What we have proved is that perhaps it would be interesting if developers did offer AMD optimized builds of plugs as well but that would take a harrasment from the public so start the e-mail campaigns guys

---

As part of the Intel / AMD Anti Trust ruling Intel has to fund any developers that want to redo their code, so no one is asking them to do it for free... It would also be nice if all the other people that have put in hard work on this issue could get something out of it

--------------------
www.dragonlogos.co.uk