: Blofelds DSP / L-Factor II / DAWbench DSP Benchmarks :.

[TAFKAT]

* Updated 16 March 2007 *

Stage I :

Credits and Thank You's : Here

Summary and Closing remarks for Stage I : Here

Stage II :

Updated Blofelds DSP Results - Start of L-Factor II Testing: Here

Tabulated Results / Graphs / Ongoing Project : Here




Hi All,

A lot of you would be aware that I have been working on a benchmark project with various end users, vendors , and hardware manufacturers over the last 4-5 months, well its time to bring the results to the general public, and also offer the test session to all that are interested.

Sorry this is going to be a bit long, but I feel it is important that I explain the history, methodology and inevitably the politics that have needed to be navigated in the process.

History , Hardware & Politics :

This project has been a work in progress since around the start of August, I decided to develop a suite of test sessions that were not based on the typical Dyno Run / CPU readout style of benchmarks that have been the norm over the last few years. Instead of initially opening the test sessions to an open forum I asked numerous vendors, end users, and audio manufacturers to be involved not only in helping develop the suite , but also to run the tests in a controlled environment so that the results would be as accurate as possible. The results would then be independently verified by all the respective participants , and the collective data would then be presented in a unified and qualified report. The sessions were then to be offered to the open forums, with a detailed , qualified and verified body of data so that an accurate comparable benchmark result could then be used as a clear reference for the end users to gauge their respective systems against.

As much as it disappoints me to say, the qualification idea was good in theory, but in the end it turned out to be near impossible to deliver in practice. This was primarily due to the politics involved in navigating certain obstacles thrown up when some unexpected variables were encountered with the chosen reference hardware. The project has been delayed numerous times waiting for the issue to be addressed, unfortunately as time dragged on, it was becoming more and more evident that the respective parties were not interested in acknowledging or addressing the issues being reported , and the longer it dragged on, the more reluctant a number of participants who were crucial in the qualifying process became in continuing. In short the qualification project was dead in the water for no other reason than the possible political fallout in presenting the data to an open forum.

I have ended the official qualification project, and will present the data as it unfolded. The session has definitely thrown up a large number of variables that can be explored , we have learned a great deal in the process , and I have no doubt that the collected and ongoing discussion can be a benefit to all.

Reference Audio Hardware :

It became evident very early on during the development stages that there was enormous variance in results at lower latencies using numerous audio cards that I and others had at our disposal , in that respect , the decision was made to restrict the official qualifying to 1 -2 high end professional audio cards. The 2 cards that were chosen were the RME HDSP Series and the Lynx 2 Series, it was expected that the variances between the 2 top end products would be negligible at best, and would be a good reference point as a lot of end users could qualify the results easily due to the saturation of RME products within the Nuendo community.

Unfortunately the variances were far more than expected, even to the point on the Intel DualCores systems that the Session was unable to be run with any consistency at any settings below 512 Samples on the RME cards. What initially started out as a benchmark to test outright system performance quickly became an exercise in testing ASIO driver efficiency, or lack there of. I will present what data I managed to collate with the assistance of numerous participants who graciously offered their time and energy :-)

Test Session and Methodology :

Blofelds DSP40 R2 : 40 Stereo Tracks of Audio, 40 4 band EQ's , 40 VST Dynamics with all componants active + Multiband Dynamics on Master Out.

40 Magneto's - 1 per track inactive to be turned on one at a time until session taps out.

You can download the test session here:

http://www.aavimt.com.au/downloads/bench.zip

File size is about 6.17 MB.- Most of that being a few Bounced Waves. Expanded it is over 130 MB.

There are also 25 stereo Sine Waves in the download, I'll explain below.


Test Preparation :

The Test Session was created using elements of the original Blofelds Return Demo.

I have modified the original Demo by first Bouncing all of the Guitar and Synth Tracks into respective stereo tracks. I have kept the Drum Tracks as they are. I did this to give a constant audio loading as the original track had short audio comps all over the place which would not give a constant loading. I have also created and added audio tracks containing full Sine Waves following the instructions given by Joe at UA on the forum, worked a treat.


1. Copy the complete Blofelds Return session folder off the Nuendo or SX 3.0 DVD to your Audio Hard Drive.

2. Unzip the downloaded file.

3. Copy all of the audio files - 2 x Bounced Waves + 25 Sine Waves - to the Audio directory of the Blofelds Return session folder.

4. Copy the Blofelds DSP40-R2.npr file into the Blofelds session folder


To run the test :

Load the session, then while the session is playing, click on each channel as you scroll across the mixer panel , the channel strip will update to the new channel - Doh - then progressively turn on each Magneto until you hear any breakup. The number of Magnetos that you can successfully load is the figure we are after, not the actual CPU loading.

The Test Session will take about 15 Minutes per latency setting to complete.

Some Tips and interesting behavior that I experienced developing the test, and that you should be aware of, just in case you encounter the same anomalies.

1. Sometimes the session will overload or glitch on first playback. Hitting the 1 Key to reset to the left indicator clears the audio buffer and the session should then playback fine. You can then start loading the Magneto's.

2. When running the session, as you turn on each Magneto , you do get a slight glitch as Nuendo buffers the PDC into memory, you will see the HD jump, and it will slightly pause.. that’s fine. It may also drop the audio driver out occasionally. Just Hit Stop, Hit 1 to reset to the left indicator, and continue until you hear break up/crackles.. This behavior is worse in 3.1 than in 3.02 ... ?? !!

3. It has also been noted that results may vary on consecutive runs with certain audio cards. At each latency setting, it is required to run the session at least 3-4 times , each time exiting out of Nuendo and then re-starting the App, loading the session again, and running thru the test. Report any variance .

4. The same can also be applied with a re-start in between runs @ each latency setting.

5. When shifting the test to a different latency setting, always restart the system.

6. If you manage to exceed the 40 Magnetos that are preloaded, scroll back to the first track and just start doubling up on each track as needed.


Test Results : Intel : Un-qualified :


Some Initial results from the early development stage on an In House System:

------------------------------------------------------------------------------------------
Blofelds DSP 40 : Pentium D 830 @ 3.0 GHZ : Nuendo 3.02

Motherboard : ASUS P5WD2-P

Memory : 1 GB - XMS 4300 @ PC3200 - 400 MHZ : Standard Timings by SPD :

O.S : XP SP1A :

Echo Mia : Driver : 6.11b

512 Samples – N/A

256 Samples – Not a Hope I could get this to run reliably, I did manage to get it to start on a number of occasions , but it was basically tapped just loading it. Simply stopping and starting the playback would overload the system, and drop the audio driver out.. ;-(

-------------------------------------------------------------------------------------------------
Blofelds DSP 40 : Pentium D 830 @ 3.0 GHZ : Nuendo 3.02

Motherboard : ASUS P5WD2-P

Memory : 1 GB - XMS 4300 @ PC3200 - 400 MHZ : Standard Timings by SPD :

O.S : XP SP1A :

Lynx 2 : Driver : V2 Build 13e

First Run : 26 Clear - very slight crackle on 27 Magneto's : CPU load 90+
Slight break up on 28th , no CPU overload.

2nd Run : Basically identical 27 Magneto's slight crackle : CPU load 90+
Slight break up on 28 , no CPU overload. HD Overload ???

--------------------------------------------------------------------------------------------------------



More Results for the In-House 840 @ 3.2 and 3.85 GHZ, running both Lynx and RME HDSP.

-----------------------------------------------------------------------------------------------

Blofelds DSP 40R2 : Pentium D 840 @ 3.2 GHZ – FSB 800 : Nuendo 3.1

Motherboard : ASUS P5WD2-P

Memory : 1 GB - XMS 4300 @ PC3200 - 400 MHZ : Standard Timings by SPD :

O.S : XP SP1A :

RME : HDSP 3632 : Driver : 2.94 Rev 1, 2, 3

512 Samples - 32 Magneto's

256 Samples - 5-9 Magneto's - Not Consistent .


Lynx 2 : Driver : V2 Build 13e

512 Samples - 36 Magneto's

256 Samples - 30 Magneto's

-----------------------------------------------------------------------------------------------

Blofelds DSP 40R2 : Pentium D 840 @ 3.85 GHZ – FSB 960 : Nuendo 3.1

Motherboard : ASUS P5WD2-P

Memory : 1 GB - XMS 4300 @ PC3200 - 480 MHZ : Standard Timings by SPD :

O.S : XP SP1A :

RME : HDSP 3632 : Driver : 2.94 Rev 1, 2, 3

512 Samples - 35 Magneto's

256 Samples - Didn't even bother .. :-)


Lynx 2 : Driver : V2 Build 13e

512 Samples - 40 Magneto's

256 Samples - 30 Magneto's

--------------------------------------------------------------------------------------


O.K, the interesting thing is @ 256 Samples , I didn't achieve any better performance @ 3.85 GHZ in regards to the number of Magneto's which is hinting that despite the meters not overloading, I was suffering from headroom in either Memory/HD. Actually it didn't scale as I expected, and I'd suspect it maybe memory related as the PC4300 Corsair is way below what I should be using with the 960 FSB.

In either case, the RME's fared worse than the Lynx Cards again, even @ 512 samples..



O.K some other quick figures to digest using RME Fireface :.

------------------------------------------------------------------------------

Blofelds DSP 40 : Pentium D 830 @ 3.0 GHZ : Nuendo 3.02

Motherboard : ASUS P5WD2-P

Memory : 1 GB - XMS 4300 @ PC3200 - 400 MHZ : Standard Timings by SPD

O.S : XP SP1A :

Fireface 800 : Driver : ?

512 Samples : 30 Magneto's

256 Samples : 8 Magneto's

--------------------------------------------------------------------------

Blofelds DSP 40 : Pentium D 830 @ 3.75 GHZ Nuendo 3.02
C/O Shane Bushman : Category 5

Motherboard : ASUS P5WD2-P

Memory : 1GB PC6400 @ 833 MHZ : 5-5-5-10

O.S : XP SP1 :

Fireface 800 : Driver : ?

512 Samples : 42 Magneto's

256 Samples : 26 Magneto's

------------------------------------------------------------------------------------------

Now what is really interesting is that Shane’s overclocked system is performing so much better with the Fireface @ 256, also it is scaling a lot better with the overclock, the only other variable is that he does have some very high grade memory modules in the system which would account for some added headroom.. ? !!

That is hinting at some other system related variables that come into play , as I noted above, I have noticed as the session starts to get into the higher regions , the HD is working overtime loading all of the buffers for the PDC, and in 3.02 was actually overloading before the CPU clipped, interesting stuff, so depending on the buffer handling capability of the card, it felt like Nuendo would transfer all of the load onto the HD when the card couldn't handle the buffers on its own DSP. The rise in HD was quite dramatic. I noticed this markedly in 3.02, however in 3.1, they may have just changed the way the metering is displaying, as it feels like its doing the same thing, but I am not getting the same feedback visually...??
This is not in any way confirmed , just an observation.



Updated In House Pentium D 840 Figures with Faster Memory Modules :.

-----------------------------------------------------------------------------------
Blofelds DSP 40R2 : Pentium D 840 @ 3.2 GHZ : Nuendo 3.1

Motherboard : ASUS P5WD2-P

Memory : 1 GB - XMS 6400 @ PC5300 - 667 MHZ : Standard Timings by SPD :

O.S : XP SP1A :

Lynx 2 : Driver : V2 Build 13f

512 Samples - 38 Magneto's / slight crackle on 39

256 Samples - 32 Magneto's / slight crackle on 33

128 Samples - 19 Magneto's / No Crackle / drop out on 20

All sessions were saved at the loaded settings , all re-opened and played back fine, all results consistent across consecutive runs .


----------------------------------------------------------------------------

Blofelds DSP 40R2 : Pentium D 840 @ 3.85 GHZ : Nuendo 3.1

Using Standard Overclock Profile / 960 FSB - PC800 :

Motherboard : ASUS P5WD2-P

Memory : 1 GB - XMS 6400 @ PC6400 - 800 MHZ : Standard Timings by SPD :

O.S : XP SP1A :

Lynx 2 : Driver : V2 Build 13f

512 Samples - 52 Magneto's / slight crackle on 53

256 Samples - 45 Magneto's / slight crackle on 46

128 Samples - 31 Magneto's / No Crackle / drop out on 32

All sessions were saved at the loaded settings , all re-opened and played back fine, all results consistent across consecutive runs .

----------------------------------------------------------------------------------------------


I was right about the overclocked system hitting a memory bottleneck with the slower DDR 4300 RAM. The only change from the original testing is the new XMS 6400 DDR2 RAM, the difference at the overclocked speed is staggering. I even managed a few more Magneto's at the standard clocking :

It is just proving that overall performance is a combination and fine tuning of all subsystems .



Some Independent Pentium D 820 Results using various audio cards :.

---------------------------------------------------------------------------------------
Blofelds DSP 40R2 : Pentium D 820 @ 2.8 GHZ : Nuendo 3.02
C/O Serge @ AudioOz :

Motherboard : ASUS P5LD2-D

Memory : 1 GB - XMS 5400 @ PC5400 - 667 MHZ : Standard Timings by SPD :

O.S : XP SP2 :

RME 9636 : Driver ?

512 Samples – 22-23 Magneto's

256 Samples – Nil – Would not run session at all
-------------------------------------------------------------------------------------------

Blofelds DSP 40R2 : Pentium D 820 @ 2.8 GHZ : Nuendo 3.02

Motherboard : ASUS P5LD2-D

Memory : 1 GB - XMS 5400 @ PC5400 - 667 MHZ : Standard Timings by SPD :

O.S : XP SP2 :

ESI Julia : Driver ?

512 Samples – 27 Magneto's

256 Samples – 22 Magneto's

-----------------------------------------------------------------------------------------------

No further comment required :-)





Test Results : AMD : Un-qualified :

---------------------------------------------------------------------------------------
Blofelds DSP 40 : AMD X2 4200 @ 2.2 GHZ : SX 3.1
C/O franYo :

Motherboard : ASUS A8N-E

Memory : 2 GB – PC3200 , 1T :

O.S : XP SP2 :

EMU 1820m : Driver : EMU ASIO v1.80

440 Samples – 33 Magneto's

308 Samples – 25 Magneto’s

---------------------------------------------------------------------------------------------

Blofelds DSP 40 : AMD X2 4200 @ 2.6 GHZ : SX 3.1
C/O franYo :

Motherboard : ASUS A8N-E

Memory : 2 GB – PC3200 , 2T :

O.S : XP SP2 :

EMU 1820m : Driver : EMU ASIO v1.80

440 Samples – 45 Magneto's

308 Samples – 37 Magneto’s

-----------------------------------------------------------------------

Blofelds DSP 40R2 : AMD X2 4200 @ 2.2 GHZ : SX 3.1
C/O franYo :

Motherboard : ASUS A8N-E

Memory : 2 GB – PC3200 , 1T :

O.S : XP SP2 :

PreSonus FireBOX : Driver : v1.20

512 Samples – 37 Magneto's

256 Samples – 32 Magneto’s

--------------------------------------------------------

Blofelds DSP 40R2 : AMD X2 4200 @ 2.6 GHZ : SX 3.1
C/O franYo :

Motherboard : ASUS A8N-E

Memory : 2 GB – PC3200 , 2T :

O.S : XP SP2 :

PreSonus FireBOX : Driver : 1.20

512 Samples – 49 Magneto's

256 Samples – 42 Magneto’s

---------------------------------------------------------------

Blofelds DSP 40R2 : AMD X2 4200 @ 2.2 GHZ : SX 3.1
C/O franYo :

Motherboard : ASUS A8N-E

Memory : 2 GB – PC3200 , 1T :

O.S : XP SP2 :

Terratec EWX 24/96 : Driver : 5.0.2000.128

512 Samples – 42 Magneto's

256 Samples – 34 Magneto’s

----------------------------------------------------------------------

Blofelds DSP 40R2 : AMD X2 4200 @ 2.6 GHZ : SX 3.1
C/O franYo :

Motherboard : ASUS A8N-E

Memory : 2 GB – PC3200 , 2T :

O.S : XP SP2 :

Terratec EWX 24/96 : Driver : 5.0.2000.128

512 Samples – 54 Magneto's

256 Samples – 46 Magneto’s

------------------------------------------------------------------------------------------------------

Some interesting and varied results depending on the respective Audio Cards/ Drivers being used.


The LEX Factor :

O.K, now we come to LEX’s contribution, I asked Lex onboard to do some RME / X2 testing while the project was sitting idle waiting for an official response to the issues being reported, and let me tell you, his testing thru up so many variables that my head is still spinning…, amount of Ram installed, memory timings, whether the system was rebooted between session runs, inconsistencies with consecutive runs with and without reboots, etc, etc. The results I am presenting are just a summary of the report he sent me, and I am sure he will go into greater detail if requested.

-------------------------------------------------------------------------------

Blofelds DSP 40R2: AMD X2 4200 @ 2.2 GHZ: Nuendo 3.2
C/O LEX:

Motherboard : Asus A8V Deluxe Revision 2 : BIOS 1014

Memory: 2 GB – PC3200: 2-3-3-6 1T

O.S: XP SP2:

RME HDSP 9652: Driver: 2.94 (2)

512 Samples – 37 Magneto's

256 Samples – 22 - 30 Magneto’s : Not Consistent

------------------------------------------------------------------------------

Blofelds DSP 40R2 : AMD X2 4200 @ 2.6 GHZ : Nuendo 3.2
C/O LEX :

Motherboard : Asus A8V Deluxe Revision 2 : BIOS 1014

Memory : 2 GB – PC3200 : 2-3-3-6 1T

O.S : XP SP2 :

RME HDSP 9652: Driver: 2.94 (2)

512 Samples –N/A

256 Samples – 37-39 Magneto’s

-----------------------------------------------------------------------------------


Blofelds DSP 40R2 : AMD X2 4200 @ 2.2 GHZ : Nuendo 3.2
C/O LEX :

Motherboard : Asus A8V Deluxe Revision 2 : BIOS 1014

Memory : 4 GB – PC3200 : 2-3-3-6 2T

O.S : XP SP2 :

RME HDSP 9652: Driver: 2.94 (2)

512 Samples – N/A

256 Samples – 14 Magneto’s

--------------------------------------------------------------------------------------

Blofelds DSP 40R2 : AMD X2 4200 @ 2.6 GHZ : Nuendo 3.2
C/O LEX :

Motherboard : Asus A8V Deluxe Revision 2 : BIOS 1014

Memory : 4 GB – PC3200 : 2-3-3-6 2T

O.S : XP SP2 :

RME HDSP 9652: Driver: 2.94 (2)

512 Samples – N/A

256 Samples – 27 Magneto’s

-------------------------------------------------------------------------------

Lex also reported that the consistency worsened the longer the system was running , reporting variance @ 256 anywhere from 4- 18 Magnetos running at the standard Clock settings, to 27 – 36 Magnetos @ 2.6 GHZ . The results are all over the place , and adding to the inconsistencies, the sessions if saved at the loaded settings , would not playback when re-opened without unloading a large number of Magneto’s , something that was also reported by Franjo with all of the cards he tested.

Definitely some talking points to explore.

O.K , just to finish off , a little icing on the cake .

The following Dual DualCore Opti results are simply astounding , and show perfectly how Nuendo can scale reliably with multiple cores if all other system variables are in check …

--------------------------------------------------------------------------------

Blofelds DSP 40R2 : Dual DC Opteron 280 - 2.4 GHZ : Nuendo 3.1
C/O Paul Erlandson @ Lynx.

Motherboard : Tyan 2885

Memory : 4 GB – PC3200

O.S : XP SP2 :

Lynx AES16 : Driver : V2 Build 13f

512 Samples – N/A

256 Samples - 96 Magneto's. * Dropouts during Magneto activation started appearing after 75

128 Samples - 65 Magneto's. * A small number of dropouts in the low 20s, more in the 50s

64 Samples - 43 Magneto's. * Dropouts after 43
* Drop Outs are referring to reported samples being dropped in the Lynx Playback Mixer - not audible dropouts.
All sessions were saved at the loaded settings , all re-opened and played back fine, all results consistent across consecutive runs .

-------------------------------------------------------------------------------

Shite… Say No More … :-)


O.K , for those that are still with me.. I thank you for your perseverance.. ;-)

These results are in no way qualified or conclusive, I have reported as the chips fell so to speak.., its been a long 4 months of trying to co-ordinate this to try and present a unified and qualified data base, unfortunately the politics got in the way, so now its open for all to contribute, analyze and hopefully further develop the methodology employed. There is definitely quite a few interesting areas of discussion to explore.

So with that, I close my chapter on this project, and hand it to all of you

Have Fun

Cheers

Vin Curigliano
AAVIM Technology

[BrianT]

Thanks for all your hard work on this.

It will take a couple of read throughs to absorb.
_________________
Regards,
Brian T

[ten]

Nice one, some good info there...cheers Taffy :)

ten
_________________
- Abit IP35-E - Intel Q6600@3.6ghz
- 4gig ddr2@1200 - 2x 150GB Raptors
- Albatron 8800GT - Dell 3007-HC @ 2560x1600
- Lynx2-C - N4.1/R2.x - XPsp2/vist64 - Adam A7

[Bernard Focquet]

Good job !
Thanks for sharing this.

Cheers,
Bernard
_________________
Nuendo v5.0, v4.x, v3.x, v2.x, v1.x
PC A:XP Pro SP3/ GA-P35-DS3P QX6850, 4G DDR3/2xRME Digiface /Powercore PCI & Firewire /MCU Pro
PC B:XP Pro SP3/ S2885 2xOpteron 250, 4G DDR /2xRME Digiface /Powercore PCI & Firewire /MCU Pro

[Simon L]

Great work. Dual Opti results are astonishing. I can also confirm that Lynx gives more stablility at lower latencies. The problem for us is that we need to use our UADs and pocos and all the latest systems necessitate the use of magmas to facilitate this.
_________________
Nuendo 3.01, XPSP2, Supermicro H8DCE dual opteron 280, 4 gig DDR ECC Patriot 400 ram2 SATA WD Raptor 150, 180gig IDE, Lynx 2, 3 UAD, 1 Terabyte sample drives all WD, 1 poco pci, magma chassis

www.silmedia.com
www.myspace.com/silmedia

[vinark]

Thanks Vin!
Vincent
_________________
C4 4.5.2
Asus P5Q @ 425FSB
Intel Q9550 @3.6ghz
8gb Corsair XMS2 6400C4@1066C5
RME hdsp9652 with AEB4-O
2Xuad1
Liquidmix
XPpro64

[psvennevig]

Have you loopback checked the real latencies of Lynx 2 and RME ?
Is 256 samples with Lynx 256 samples? Is it with RME ??

Earlier tests with Lynx showed that there are extra buffers in the driver giving higher latencies.
256 samples were close to 340, etc.
RME has in all my calcs. been at 256 when checking loopback. 64 is 64 etc.

It of course possible to run more plugs at 340 samples latency (if that is the correct number) than on 256 samples.

It's been at least a years ince I did these tests, so Lynx may of course have improved their ASIO driver.

Pål
_________________
Nuendo 5.0.1, WK-Audio ID Control, RME HDSP MADIe, RME ADI-6432, 1xADI-8 QS, MBP i7, 8GB, SSD and MAC pro 8-cores Nehalem 12GB RAM 2x2.26GHz.

[Phil Fee]

Blofelds DSP 40R2 : AMD X2 4400 @ 2.2 GHZ : Cubase SX 3.1.1

Edit with corrections and additions:

Motherboard : K8NS Ultra 939

Test 1: @ 2T command rate

Memory : 2 GB – mixed pairs PC3200/PC3500 (4 sticks, all C2 Corsair)
Running at 200 MHz and 2-3-3-6 2T

O.S : XP SP1 :

RME HDSP: Driver: 2.94

1024 Samples - 42 Magnetos

512 Samples – 38 Magnetos

256 Samples – 14 Magnetos

(NB Earlier posted version of Magneto count at 2T was in error - 1 or 2 instances too high)

Note: Tuning RAM from CAS 2.5 to CAS 2 (1 parameter alterred - so 2.5 3 4 7 [=default] changed to 2 3 4 7) made a major improvement in the number of Magnetos. Changing the rest of the parameters ( 2 3 3 6) did not add a single Magneto at 512 samples.
This system would not boot at 1T at stock RAM voltage and 2 pairs of DIMMs (ie no RAM overvoltage in bios).

Retest with 2 sticks removed and 1T command rate:
Memory : 1 GB – PC3500 (1 pair C2 Corsair)
Running at 200 MHz and 2-3-3-6 @ 1T

O.S : XP SP1 :

RME HDSP: Driver: 2.94

1024 Samples - 42 Magnetos

512 Samples – 39 Magnetos

256 Samples – 28 Magnetos

Retest (22 March O6) with Hammerfall DSP driver version 3.0 (beta 4)

4 sticks 512 MB DDR 3200 Command Rate 2T

O.S : XP SP1 :

RME HDSP: Driver: 3.00 beta 4

512 Samples – 43 Magnetos (was 38 with driver 2.94 2 GB RAM @ 2T)

256 Samples – 35 Magnetos (was 14 with driver 2.94 2GB RAM @ 2T)

128 Samples -- 4 or 5 Magnetos.

[vinark]

[Phil Fee]

[vinark]

as far as I know it´s a chipset limitation 2 pieces is 1T 4 pieces 2T. Could give you an Idea if upgrading to 2X1gb is interesting.
_________________
C4 4.5.2
Asus P5Q @ 425FSB
Intel Q9550 @3.6ghz
8gb Corsair XMS2 6400C4@1066C5
RME hdsp9652 with AEB4-O
2Xuad1
Liquidmix
XPpro64

[Phil Fee]

[franYo]

I can confirm that 1T makes a considerable difference, especially at 256 samples. At 512 it doesn't matter that much. It is actually the additional RAM bandwidth (in my experience 1T gets you approximately 20% of additional bandwidth) that makes the difference. Similar result can be obtained by overclocking just RAM. Previously I ran the Gigabyte GA-K8N Ultra-9 board which didn't run very stable at 1T, but it ran perfectly well with FSB set to 245 and CPU multiplier at 9X. This gave me approximately the same bandwidth as if 1T were enabled (at the same clock speed) and the performance was very similar too. In my experience, on AMD systems, you need about 5000 MB/sec to be able to get good numbers with this particular test at 256 samples.

Franjo
_________________
ASUS A8N-E / AMD Athlon 64 X2 4200+ / SuperTalent 2GB PC3200/ nVIDIA GeForce 6600 / PreSonus FireBOX on PCI-e FW card/ 2xUAD-1 /
Windows XP Home SP2
Cubase 4.0.1.2074

[BIG8]

[Beijing office chiming in]

Great work, TAFKAT,

[franYo]

[BIG8]

[Over Compressed]

[tamasdragon]

I ran a fast test on my setup (see in my sig.)
buffer at 512 with multiface
9 magnetos switched on
cpu 90+ percent

I think it's not too bad from a single core prescott. Someone should send the results to rme, then they may update their drivers for more a efficient one.
Regards Tamas Dragon
_________________
http://www.digisample.org

[tamasdragon]

Ah, I almost forgot it. This test was taken on a machine that has internet connection enabled. Actually I write these words on this.
Regards Tamas Dragon
_________________
http://www.digisample.org

[franYo]

[h4nc0]

Just did a quick run. 38 magnetos @ 128 buffer. CPU meter about 90-93%.

X2 4400+ @ 2.7Ghz (11x246) 2x 1GB Samsung @ 3-4-4-8-1T, RME Fireface 800

Not so sure if I am doing this right. ^^ IE with this forum + Outlook Express + MSN Messenger running while doing the test.

[LEX]

Okay all, thanks for the input but obviously tou are not reading the original post.

RME is coming up with differences when you get higher CPU's.
READ Taf's "LEX factor" portion.

I have Christmas party hang over right now, so I'l get into it later.

LEX
_________________
N3.2 XPProSP2/AMD4800X2@2.6/A8V DLX
4gRam/2xRME 9652 HDSP/UAD
G750 2x20"wide 1x17" video/MOTU XT/MCU/Xkeys
3 x AMD X2 4200/2gigs/MOL
G5 2.7/3gigs RAM/RME9652
1 Giga3/2gigs RAM P4-3ghz - 9652
1 Giga2/1Gig RAM /P3 1ghz - WCPCI

[TAFKAT]

[TAFKAT]

[franYo]

Hey Vin,

There are actually two monologues, an a1 and an embracer loaded and turned on, however, no MIDI is routed to them so I guess their contribution to the overall CPU consumption is close to zero and they can be safely ignored.

As for reported latencies, with my Terratec card SX reports 5.8ms and with FireBOX 6.5ms. I have no idea if this is accurate and why wouldn't it be. The added 0.8ms on FireBOX I have always attributed to the additional "safety" buffer all FW devices are supposed to be employing, but now I see both Lynx's and RME's reported latencies are even higher? Hmmm...

About saving/reopening the session: At 6ms I can only save/reopen with 17 Magnetos on using FireBOX (21 with Terratec), and get perfect playback. From 8ms and up it works fine. With FireBOX, at 8ms (reported 8.526ms by SX) I can save the project with 40 Magnetos on and it'll open and play back with no problem. That's with the system overclocked to 2.6 GHz, of course.

Franjo
_________________
ASUS A8N-E / AMD Athlon 64 X2 4200+ / SuperTalent 2GB PC3200/ nVIDIA GeForce 6600 / PreSonus FireBOX on PCI-e FW card/ 2xUAD-1 /
Windows XP Home SP2
Cubase 4.0.1.2074