alexanrs wrote:

Low cacheable limits and apps that crap out without a pagefile do exist. I'd rather enable the "ConservativeSwapFile" option on SYSTEM.INI.

Please don't remind me about cachable limits -- I rock a TX board. I like your solution, btw. I think it would do very well.

One other thing which would be worth exploring on a high RAM spec Win98 build is the pagefile on a RAM disk.

So I've been doing more research and testing. I've installed Windows 95 OSR2 and gotten ATTO Disk Benchmark running and HD Tach. I also compared Winbench 96 running in Win95 compared to WFW and Win95 is a tad slower.

HD Tach and Speedsys both shows the AVA-2825 EIDE controller with the Samsung Drive (in PIO mode 4) at 12 MB/s linear read, and ATTO gives a 8MB/s write and 10MB/s read on all files above 256kb. Looking at the CPU power being used I think the limiting factor is the AMD X5-P75 133mhz. I tweaked some memory and system BIOS settings and got a boost. The only way to increase disk performance seems to either offload the CPU usage and/or increase the system speed. On a VLB 486 system there are only two ways to offload CPU usage, that is through DMA busmastering SCSI controllers or IDE cache controllers (which have their own CPU). From my previous benchmarking the DC4030VL-2 cache controller was hampered by the overhead and most likely the CPU on cache controller which couldn't keep up with the modern IDE harddrive, so my next text will be Winstone which tests CPU and disk performance at the same time in real life Windows applications. Here we'll see if the offloading of the main CPU have any effect and/or if it can weigh up for the decreased performance compared to a VLB EIDE controller.

I've read into contemporary reviews and articles in PC Mag and their findings is that software cache (SmartDrive) is almost just as quick as hardware cache. Software cache is directly accessible from the system memory, while the hardware cache on the controller have to go through the I/O bus. I think this is why these controllers weren't as popular back in the days, the price vs performance was just not worth it in many cases.

They also showed that VLB IDE have better performance over VLB SCSI (on contemporary harddrives) in DOS, they contributed this to the overhead on SCSI controllers since they need to translate between SCSI commands and DOS int13, while IDE doesn't need to do this. DOS doesnt take advantage of the special SCSI features like command queuing like a proper operating system does.

I tested my Connor drive against a 1992 Fujitsu 512MB SCSI drive I have laying around. They were about equal in performance. The Connor drive was faster on smaller files than the Fujitsu, but in linear read speed they benched at around 2MB/sec. In WinBench96 the Connor drive actually scored higher than the SCSI drive when it was connected to the VLB cache controller. Again, I need to connect a much more modern SCSI drive and bench it in Winstone to see if the offloading of the main CPU have any effect on overall performance.

I'll also bench a SSD drive with IDE to S-ATA converter. It's a bit fun to see what options you have and how they compare on pre-PCI systems. Not much research/benching have gone in to it and it also makes me value PCI alot more. It is alot quicker with the available controllers and give much more options.

After long $!"#$"#% I finally managed to get Tekram DC-680T VLB to work. Will do some tests tomorrow... but preliminary test with CF card and 4MB on-board cache shows little to no improvement 😐

kixs wrote:

After long $!"#$"#% I finally managed to get Tekram DC-680T VLB to work. Will do some tests tomorrow... but preliminary test with CF card and 4MB on-board cache shows little to no improvement 😐

Great of you to join in on this little project 😀 I do feel your pain. The troubles I've had to go through to get these VLB controllers working to the optimal in DOS, WfW and Win95 merits its own thread! It's actually been very helpful having the original documentation and disks to troubleshoot.

I am looking forward to your results with a modern SCSI drive on VLB vs. modern IDE on VLB.

I have been wondering if using an Ultra160 SCSI drive on my AHA-2842A would have any benefit over a generic IDE VLB controller with modern ATA-133 drive.

More work done, I tested out Winstone96 between the Samsung and Connor drive on the AVA-2825 controller.

Score was 43.5 to Samsung and 37.5 to Connor. Time to execute the whole bench took 31min 42sec on the Connor and 28min 13sec on the Samsung. I'm not sure WinStone is such a great benchmark for disk performance. It does play a part, but it's more centered around memory, graphics and CPU power.

I tried to switch back to the DC4030VL-2 IDE cache controller, but since I don't have the LBA BIOS for it, and it was never available online it will take me a bit more time to complete. When I installed Win95 on the Samsung drive some of the installation/boot files moved to sectors outside the 504MB limit on the 2GB partition. This means that unless I format the drive I can't boot from it with the DC4030VL-2. I don't want to do that, so instead I'll take another drive and setup Win95 on it within a 504MB partition and when it comes to the benching I'll just connect the Samsung as the slave drive.

I still don't know when the Buslogic cache controller arrives in the mail, that card do have LBA support.

If anyone have a DC4030VL (of any variation) with the LBA BIOS upgrade, please make a BIOS dump and make it public!

feipoa wrote:

I am looking forward to your results with a modern SCSI drive on VLB vs. modern IDE on VLB.

I have been wondering if using an Ultra160 SCSI drive on my AHA-2842A would have any benefit over a generic IDE VLB controller with modern ATA-133 drive.

I don't have a conversion from 68pin to 50pin available, and the ones on Ebay atm is quite expensive. I've ordered the 80pin to 50pin converter though, crossing fingers it'll work. The fastest SCSI drive with 50pin that I have is a 2000's 9GB IBM Deskstar. I do have some 15k RPM Ultra320 and Ultra160 drives.

I made a mistake benching the SCSI drive earlier. AVA-2825 (with SCSI and EIDE) is PIO SCSI, not busmastering like the AHA-2842. Luckily I have that controller as well, so I'll swap it out for further benchmarking. Beside that I think the AVA-2825 is one very sweet controller, EIDE w/ ATAPI, SCSI and Floppy all in one VLB card.

I also have the 80-pin SCA to 50-pin Narrow, however it did not work with my SCA Ultra160/320 drives and the Adaptec AHA-2842. That adapter did work on the PCI-based 2940U2W and 2940W though. I am not sure what the problem was. To get a modern SCSI drive working with the AHA-2842, I had to use a 68-pin Ultra 160/320 drive with a 68-pin (male) to 50-pin (male). That particular adapter goes on the HDD side, however you might be able to use a 68-pin (female) to 50-pin (female) on the host controller side. I bought one of these on dealextreme about 2 months ago for approx. $6 shipped.

One thing you might want to add. Under Windows 3.1x, are you running with 32-bit disk and file access turned on? Just about every SCSI card had a driver available for it, and most of the IDE cards did. I have a VLB caching IDE card here, a DTC 2277A (actually made by ATronics), but its onboard ROM is pretty brain dead and doesn't support drives over 99 heads 😜, so needless to say it can't do LBA.

https://web.archive.org/web/19970503165025/ht … anuals/2277.htm

I also have a EISA Adaptec AHA-2970W, but no drives at the moment.

NJRoadfan wrote:

One thing you might want to add. Under Windows 3.1x, are you running with 32-bit disk and file access turned on? Just about every SCSI card had a driver available for it, and most of the IDE cards did. I have a VLB caching IDE card here, a DTC 2277A (actually made by ATronics), but its onboard ROM is pretty brain dead and doesn't support drives over 99 heads 😜, so needless to say it can't do LBA.

https://web.archive.org/web/19970503165025/ht … anuals/2277.htm

I also have a EISA Adaptec AHA-2970W, but no drives at the moment.

Yes, as mentioned in first post, 32bit disk and file access is activated (was a pain to get going on the DC4030VL-2) 😀 The DTC card looks like it has a BIOS upgrade available, so might be possible to get LBA on it.

Can you also run Checkit or Comptest HDD benchmark? My VLB controllers are almost never any faster than ISA ones 😕

kixs wrote:

Can you also run Checkit or Comptest HDD benchmark? My VLB controllers are almost never any faster than ISA ones 😕

Could you attach those to this thread or provide links?

feipoa wrote:

That particular adapter goes on the HDD side, however you might be able to use a 68-pin (female) to 50-pin (female) on the host controller side. I bought one of these on dealextreme about 2 months ago for approx. $6 shipped.

Thanks for the tip! I have a female to female adapter, but I feel stupid for not thinking it could be attached on the host controller side. I was so focused on attaching it to the harddrive.

vetz wrote:

kixs wrote:

Can you also run Checkit or Comptest HDD benchmark? My VLB controllers are almost never any faster than ISA ones 😕

Could you attach those to this thread or provide links?

In this thread you have a Benchmark pack ready for download with everything 😉
Performance comparison of CPU: 286-25 vs 386DX-25 vs 386SX-25

It's also nicely organized - checkit is under CPU, comptest is under RAM.

vetz wrote:

feipoa wrote:

That particular adapter goes on the HDD side, however you might be able to use a 68-pin (female) to 50-pin (female) on the host controller side. I bought one of these on dealextreme about 2 months ago for approx. $6 shipped.

Thanks for the tip! I have a female to female adapter, but I feel stupid for not thinking it could be attached on the host controller side. I was so focused on attaching it to the harddrive.

Looking forward to your analysis. I ordered the female-to-female but haven't tested it in this configuration yet.

feipoa wrote:

Looking forward to your analysis. I ordered the female-to-female but haven't tested it in this configuration yet.

The female to female adapter worked!

Primlinary tests seems to show that IDE on modern ATA-133 drive with a VLB controller that supports EIDE and PIO Mode 4 is the fastest. I benched with a Seagate Cheetah 15K.3 73GB U320 68pin SCSI Hard Drive ST373453LW on the Adaptec AHA-2842VL SCSI host controller. In a proper system with a good controller this drive is faster and more modern than the Samsung IDE drive.

WinBench96: 1520
(Score for the Samsung IDE drive on AVA-2825 EIDE VLB was 1840)

L1 Write-Back was turned on in BIOS settings.

I'll see if I can tweak the settings a bit, but the main limitation from benches in ATTO and SiSoftSandra 99 seems to be the 10MB/sec transfer limit on SCSI-2, which it is very close to hitting (8-9MB/sec). Burst rates are 18MB/sec (max theoretical is 40mb/sec on the VLB SCSI). The SCSI drive scored better on seekrate and smaller files, but I think that mainly comes from the 15k RPM. On larger files the transfer rate on the VLB IDE was just so much better. If you get an even quicker IDE drive, like a SSD or SD-card I'd recon the differences will be even larger. When I'm done with all the testing I'll sum it all up in the first post.

I also did some benchmarks on the DC4030VL-2 VLB IDE cache controller using a Western Digital 6.4GB IDE drive. It scored 718 in WinBench96 on plain VLB PIOMODE 2 and 834 using the DC4030VL-2 with 16MB hardware cache. So this seems to start confirming that using VLB cache controllers is only beneficial on time period hard drives. If you use a modern drive it will cut maximum performance because it cant keep up.

Those results a certainly disappointing from my perspective. I had already setup my VLB system using the AHA-2842. However, I wonder how your AVA-2825 compares to the AHA-2842 with heavy CPU load when the HDD is accessing.

Could you also compare your AVA-2825 / ATA-133 results to a standard Winbond VLB I/O card? I wonder if your AVA-2825 has the edge on account of its design.

feipoa wrote:

Those results a certainly disappointing from my perspective. I had already setup my VLB system using the AHA-2842. However, I wonder how your AVA-2825 compares to the AHA-2842 with heavy CPU load when the HDD is accessing.

I wouldn't worry too much. The differences aren't that huge, and there are some situations were a SCSI setup would make more sense. I wonder how a SCSI cache card like the Tekram DC-880 would perform. I believe it would most likely be slower on modern drives, but faster on contemporary ones, just like VLB IDE caching controllers. I'd love to get my hands on one in the future.

I tweaked the BIOS settings and I got a score of 1650 with the SCSI setup in WinBench96. For Winstone96 I actually improved the score to 44.4 and reduced completion time by 30 seconds on the SCSI drive compared to the IDE. Again, the SCSI drive is technical superior (15k RPM, 8mb cache) to the IDE drive, so I'll need to up the drive quality on the IDE side (newer drive, SSD, SD-card).

feipoa wrote:

Could you also compare your AVA-2825 / ATA-133 results to a standard Winbond VLB I/O card? I wonder if your AVA-2825 has the edge on account of its design.

Yes, that is the plan. Hopefully it'll activate PIO MODE 4 this time, but it was good to know that unless you actually run your hard drive in PIO MODE 4, the VLB transfer bus speed increase alone offers little performance increase compared to plain ISA In Winbench96. There was a 30 points difference (1280 vs 1250).

Which settings did you tweak in the AHA-2842's BIOS to increase your benchmark results?

feipoa wrote:

Which settings did you tweak in the AHA-2842's BIOS to increase your benchmark results?

I switched to manual settings which allowed me to set DRAM to 0WS, and I also set burst and refresh settings to enabled. I've also switched the graphics card to a Diamond Stealth 64 Vision964 2MB VRAM card and reduced resolution to 640x480 while running Winstone96. This improved the score and completion time. My goal is to make the graphics a less variable in the result.

In other news I've gotten a Tekram DC-680C VLB IDE cache controller on the way in the mail. So I should really be able to make a comparison now, even amongst cache controllers 😀

Those sound like motherboard BIOS settings, not SCSI BIOS settings.