nForce 590 SLI Intel Edition: NVIDIA prepares an Intel 975X Killer
by Gary Key on June 27, 2006 6:15 AM EST- Posted in
- CPUs
Media Encoding Performance
Our first test is quite easy - we take our original Office Space DVD and use AnyDVD Ripper to copy the full DVD to the hard drive without compression, thus providing an almost exact duplicate of the DVD. We then fired up Nero Recode 2, selected our Office Space copy on the hard drive, and performed a shrink operation to allow the entire movie along with extras to fit on a single 4.5GB DVD disc. We left all options on their defaults except we checked off the advanced analysis option. The scores reported are the time for the full encoding process and are represented in minutes, with lower numbers indicating better performance.
The results are very interesting as we did not expect the NVIDIA system to perform this well and to do so is very surprising considering the early build sample and limited BIOS functionality. In fact, we ran the test several times and verified our settings before accepting the test results. It appeared to us that the storage subsystem on the NVIDIA board made a difference, though as you'll see in a moment this is not the case with one of our audio benchmarks.
Audio Encoding Performance
While the media encoding prowess of the NVIDIA nForce 590 SLI was superb in our limited testing, we wanted to see how it faired on the audio side. Our audio test suite consists of Exact Audio Copy v095.b4, LAME 3.98a3, LAME MT3.97a-MS Compiler, and Nero Digital Audio. Our first task was to figure out what test CD to utilize. We needed one that contained a significant number of tracks and had over 600MB of data in order to properly stress our platforms. After rummaging around the lab once again and coming up empty handed we reverted to our INXS Greatest Hits CD. This one time '80s glory masterpiece contains 16 tracks totaling 606MB of songs that can make the sturdiest of optical drives whimper.
Our first test consists of utilizing Exact Audio Copy as the front end for different versions of LAME. We set up EAC for variable bit rate encoding, burst mode for extraction, use external program for compression, and to start the external compressor upon extraction (EAC will read the next track while LAME is working on the previous track, thus removing a potential bottleneck with the drive).
Our two versions of LAME consist of the recently released 3.98a3 and LAME MT 3.97a, which is a multi-threaded version of the LAME MP3 encoder. LAME MT was originally designed as a demonstration to show the advantages of multi-threading on the Pentium 4 with Hyper-Threading enabled. Instead of running multiple parallel threads, LAME MT generates the encoder's psycho-acoustic analysis function in a separate thread from the rest of the encoder using a simple linear pipeline. The results are presented in minutes/seconds for the encoding process, with lower numbers being better.
As in the media encoding section, the more intensive CPU and storage system tests seem to favor the NVIDIA platform, although very slightly. We ran these tests several times with the same results all being within the same percentage of each other. When utilizing LAMEMT we see upwards of a 35% improvement in encoding times. This once again shows the advantages of program when it is written to take advantage of multiple-core processors.
Our first test is quite easy - we take our original Office Space DVD and use AnyDVD Ripper to copy the full DVD to the hard drive without compression, thus providing an almost exact duplicate of the DVD. We then fired up Nero Recode 2, selected our Office Space copy on the hard drive, and performed a shrink operation to allow the entire movie along with extras to fit on a single 4.5GB DVD disc. We left all options on their defaults except we checked off the advanced analysis option. The scores reported are the time for the full encoding process and are represented in minutes, with lower numbers indicating better performance.
The results are very interesting as we did not expect the NVIDIA system to perform this well and to do so is very surprising considering the early build sample and limited BIOS functionality. In fact, we ran the test several times and verified our settings before accepting the test results. It appeared to us that the storage subsystem on the NVIDIA board made a difference, though as you'll see in a moment this is not the case with one of our audio benchmarks.
Audio Encoding Performance
While the media encoding prowess of the NVIDIA nForce 590 SLI was superb in our limited testing, we wanted to see how it faired on the audio side. Our audio test suite consists of Exact Audio Copy v095.b4, LAME 3.98a3, LAME MT3.97a-MS Compiler, and Nero Digital Audio. Our first task was to figure out what test CD to utilize. We needed one that contained a significant number of tracks and had over 600MB of data in order to properly stress our platforms. After rummaging around the lab once again and coming up empty handed we reverted to our INXS Greatest Hits CD. This one time '80s glory masterpiece contains 16 tracks totaling 606MB of songs that can make the sturdiest of optical drives whimper.
Our first test consists of utilizing Exact Audio Copy as the front end for different versions of LAME. We set up EAC for variable bit rate encoding, burst mode for extraction, use external program for compression, and to start the external compressor upon extraction (EAC will read the next track while LAME is working on the previous track, thus removing a potential bottleneck with the drive).
Our two versions of LAME consist of the recently released 3.98a3 and LAME MT 3.97a, which is a multi-threaded version of the LAME MP3 encoder. LAME MT was originally designed as a demonstration to show the advantages of multi-threading on the Pentium 4 with Hyper-Threading enabled. Instead of running multiple parallel threads, LAME MT generates the encoder's psycho-acoustic analysis function in a separate thread from the rest of the encoder using a simple linear pipeline. The results are presented in minutes/seconds for the encoding process, with lower numbers being better.
As in the media encoding section, the more intensive CPU and storage system tests seem to favor the NVIDIA platform, although very slightly. We ran these tests several times with the same results all being within the same percentage of each other. When utilizing LAMEMT we see upwards of a 35% improvement in encoding times. This once again shows the advantages of program when it is written to take advantage of multiple-core processors.
Facebook
Twitter
RSS
37 Comments
View All Comments
Frumious1 - Wednesday, June 28, 2006 - link
The problem is determining whether theres really a problem or the reviewers just need to learn how to use the motherboard and BIOS options properly. Wes, Gary, and most of the rest of the AnandTech crew seem to know how to really get results out of motherboards. I saw an article a few days ago that was an absolute joke when it comes to OCing. They took a 4200+ and were crowing about a 240 MHz HTT bus overclock or something. I don't think they ever even tried the other memory ratios.Anyway, registry corruption? Yeah, I've lost the registry a few times on Intel and AMD systems when pushing the OC a bit too far in the wrong way. Bad memory timings for an OC can be just as harmful as a bad CPU or chipset OC - probably even worse. OC'ing is not really that easy if you don't know what you're doing. Too many people want to just increase everything 25% and then tehy wonder why the system won't POST.
Someone on AT did some OCing articles last year about the topic that really provided some good details, showing 2.6 GHz or so with an X2 3800+ using everything from POS value RAM up through top quality TCCD and CH5 modules. Probably took a hell of a long time to complete all the testing as well! If you want to do a motherboard review right and you want to look at overclocking, you simply can't do that without spending a good month or more with the board. Sometimes a seemingly small change will stabilize what appeared to be a hopeless OC.
Bottom line: if you want SLI (and honestly, for dual GPUs it's far better than CrossFire right now - maybe not in performance, but the ATI CF drivers are still crap!) for Conroe, you'll need an nVidia chipset. Unless they suddenly get with the program and start supporting SLI on any dual X16 slot board? God, wouldn't that be nice? Stupid political bullshit... from nVidia and ATI!
Frumious1 - Wednesday, June 28, 2006 - link
Dammit... I did it again and used an H in brackets... which turns on white text for reasons unknown. Repost so people can read the text without highlighing:--------------
The problem is determining whether theres really a problem or the reviewers just need to learn how to use the motherboard and BIOS options properly. Wes, Gary, and most of the rest of the AnandTech crew seem to know how to really get results out of motherboards. I saw a HardOCP article a few days ago that was an absolute joke when it comes to OCing. They took a 4200+ and were crowing about a 240 MHz HTT bus overclock or something. I don't think they ever even tried the other memory ratios.
Anyway, registry corruption? Yeah, I've lost the registry a few times on Intel and AMD systems when pushing the OC a bit too far in the wrong way. Bad memory timings for an OC can be just as harmful as a bad CPU or chipset OC - probably even worse. OC'ing is not really that easy if you don't know what you're doing. Too many people want to just increase everything 25% and then tehy wonder why the system won't POST.
Someone on AT did some OCing articles last year about the topic that really provided some good details, showing 2.6 GHz or so with an X2 3800+ using everything from POS value RAM up through top quality TCCD and CH5 modules. Probably took a hell of a long time to complete all the testing as well! If you want to do a motherboard review right and you want to look at overclocking, you simply can't do that without spending a good month or more with the board. Sometimes a seemingly small change will stabilize what appeared to be a hopeless OC.
Bottom line: if you want SLI (and honestly, for dual GPUs it's far better than CrossFire right now - maybe not in performance, but the ATI CF drivers are still crap!) for Conroe, you'll need an nVidia chipset. Unless they suddenly get with the program and start supporting SLI on any dual X16 slot board? God, wouldn't that be nice? Stupid political bullshit... from nVidia and ATI!
Anemone - Wednesday, June 28, 2006 - link
You certainly make good points. However one review of "good" and many, many of "bad" doesn't lead me to think the "good" just knew what they were doing. In fact it's quite the opposite. It leads me to think they didn't dig deep enough.Gary Key - Friday, June 30, 2006 - link
I will take a different path on this one. We were allowed a first look at the board with Conroe back in early May, in fact if we could have stayed an extra day we would have had significant hands-on time with the setup. We were also one of the first sites to receive a board from NVIDIA for the specific purpose of testing the board to provide specific feedback regarding Conroe compatibility and performance. I easily have over 200 hours of test time on this board along with a couple of pages of issues/improvements/suggestions we would like to see before the design goes into production. As far as not digging deep enough, I doubt we would have had this early of an opportunity if it were not for our work (and that of several AT readers) with NVIDIA over the past several months in assisting them and the board suppliers to get their Intel performance up to speed. We are still very disappointed with the FSB overclocking results with the NVIDIA Intel designs but our initial board had no issue running up to 304FSB with an early Conroe sample. I am personally disappointed with the entire FSB issue since last fall as I had a couple of boards that easily did over 320FSB only to see this capability whacked when the product was released.
Yes, the board will use the nF4 SLI SPP for the "Northbridge" but it is now at a C1 stepping after several months of tuning due to the issues found last fall in the first release. Are we disappointed that we will not get the newer C51XE SPP, yes, but the time to develop it along with the switch to a single dual x16 chipset this winter made it impractical for NVIDIA at this time. The good news is the NV590SLI boards should cost around $150 at launch with a feature set that will not be matched by Intel or ATI along with using the new MCP55PXE so drive and network performance is greatly improved from a stability viewpoint.
We did not post the actual memory scores as we are waiting on a new board revision and production level bios before making any final statements on this subject and FSB overclocking. However, even with the 805D the base unbuffered Sandra scores were about 2% better than the i975x. When overclocked, this margin flipped in favor of the Intel board. The margin was even greater during our Conroe testing. I am still concerned with the FSB overclocking capability. I stated this at the end of the article, it is a concern and will remain a concern until we see production level boards. I think it will improve compared to today's products but I doubt we will see anything near what the i975x and now P965 chipsets will be capable of in the high end boards from Asus, DFI, Abit, MSI, and Gigabyte. However, getting over 300FSB is a requirement we have placed on NVIDIA at this time. It will be interesting to see if they can get there now.
We appreciate the comments and please keep them coming. Our final review on the reference board will be available shortly and we should have boards from DFI and Asus around Core 2 Duo launch time. However, we do not expect ATI review samples until sometime in August along with some interesting information about their design choice that will be discussed at that time.
mino - Tuesday, June 27, 2006 - link
Almost every p$ chipset capable of 10066 FSB since i865 tiems DOES support Conroes. Just crippled Intel 915/925 series do NOT bute even this is caused by intel marketing decision not the capability of chipset design on itself.What is most important is the board/VRM support. Otherwise every not-crippled chipset should work.
mino - Tuesday, June 27, 2006 - link
P4 meant :)mino - Tuesday, June 27, 2006 - link
hell, I should get some sleep apparently ;-\