VOGONS


Reply 120 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

My Cyrix Sx486SLC2-50MP finally arrived. It took about 7 weeks, but looks to be original product.

Cyrix_Cx486SLC2-50MP.JPG
Filename
Cyrix_Cx486SLC2-50MP.JPG
File size
76.97 KiB
Views
954 views
File license
CC-BY-4.0

I desoldered the SXLC2-40 and soldered on the SLC2-50.

Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP.JPG
Filename
Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP.JPG
File size
238.01 KiB
Views
954 views
File license
CC-BY-4.0

The SLC2 is indeed default in 2x mode. I ran it at 66.6 MHz. With all of L1's cache region set as non-cacheable, chkcpu shows the speed as 61 MHz. With the cache region inhibited, e.g. with cyrix.exe -i1, chkcpu reports 33 MHz. Obviously, chkcpu is mistaken in both situations. Nonetheless, the SLC2-50 works fine at 66.6 MHz, but it needs active cooling. Without a heatsink and fan at 66 MHz, the system will hang up once the CPU reaches around 60 C. So I will need to figure out some kind of solution that isn't unsightly.

Here's some numbers at 66.6 MHz with an ET4000AX.

Landmark v2: 166 MHz ALU and 45.3 MHz FPU
Cachechk: L1 = 44.3 MB/s and RAM = 11.5 MB/s
3Dbench = 18.1 fps
DOOM = 10389 realtics, or 7.19 fps

The system definitely feels chipper. Note that the FSB = CRYSTAL / 4, so FSB = 66.66 / 4 = 16.7 MHz. The Evergreen unit has a custom 2x PLL, making it 33.3 MHz, and the SLC2 has another 2x PLL, making the CPU run at 66.6 MHz. Is the ISA frequency the same as the FSB on these 286 boards, meaning that the ISA bus is running at 16.7 MHz? If so, can the Adaptec AHA-1542CP cope?

By way of comparison, when I had the SLC-33 soldered on, I recorded:

Landmark v2: 105 MHz ALU and 41.6 MHz FPU
Cachechk: L1 = 25.1 MB/s and RAM = 11.5 MB/s
3Dbench = 14.2 fps
DOOM = 12609 realtics, or 5.92 fps

I then tried to run the SLC2-50 at 70 MHz, but I was noticing some characters on the screen that should not be there, so I assume 66.6 Mhz is a realistic maximum speed.

SLC2_at_70MHz.JPG
Filename
SLC2_at_70MHz.JPG
File size
156.87 KiB
Views
954 views
File license
CC-BY-4.0

The other configuration I wanted to try was an overclocked SLC33. If the SLC25 can only do 33 MHz reliably, then I figure the SLC33 can only do 40 MHz. This would give a 20 Mhz FSB compared to the SLC66's 16.7 MHz, but I doubt would make for a faster system. Now if the SLC33 could run at 45 or 50 MHz MHz, it would probably be faster overall than the SLC66. Unfortunately, the effort and risk required to solder/desolder these QFP chips is the limiting factor for me and I won't be experimenting further in this capacity. If I had another Evergreen 486 SuperChip, I'd be more willing to experiment further as I have three NOS SLC33 chips.

Plan your life wisely, you'll be dead before you know it.

Reply 121 of 174, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

I never had much luck with 1542 cards above 12mhz, but the 1542cp would be your best bet since it was the final revision.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 122 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I decided to go with a 5-volt 25x25x7mm fan, a heatsink, and some thermal tape. I hope the thermal tape doesn't remove the Cyrix branding from the chip's surface if I need to remove it.

Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP_with_heatsink-fan_01.JPG
Filename
Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP_with_heatsink-fan_01.JPG
File size
357.45 KiB
Views
941 views
File license
CC-BY-4.0
Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP_with_heatsink-fan_02.JPG
Filename
Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP_with_heatsink-fan_02.JPG
File size
245.33 KiB
Views
941 views
File license
CC-BY-4.0
Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP_with_heatsink-fan_03.JPG
Filename
Evergreen_486_SuperChip_with_Cyrix_Cx486SLC2-50MP_with_heatsink-fan_03.JPG
File size
209.42 KiB
Views
941 views
File license
CC-BY-4.0

Plan your life wisely, you'll be dead before you know it.

Reply 123 of 174, by pshipkov

User metadata
Rank Oldbie
Rank
Oldbie

This looks great.
The few perf metrics you shared are in the ballpark of what i saw with SXL2 CPU at 66-70MHz.
How stable is the system at 66MHz ?
Will be interesting also to see Wolf3D test.
Also what is the disk i/o reported by CORETEST or other benchmarking utils ?

AHA-1542C# handles out of spec ISA frequencies quite well, but you need to counter that with lowering the speed setting from BIOS otherwise data corruption can occur.
Also, it exhibits write access problems with some of the DLC/SXL 386 upgrade CPUs.
It will be interesting to see if it work well with SLC processors and if bring any advantage over IDE.
These VLSI-200 based mobos handle IDE really well and outdo any ISA SCSI card i know of.
In other words their IDE perf is clock-to-clock faster than SCSI.

retro bits and bytes

Reply 124 of 174, by Sphere478

User metadata
Rank l33t++
Rank
l33t++

This is exactly the kind of stuff I’m here for. I love it! Good job!!!

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 125 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++
pshipkov wrote on 2022-02-06, 09:01:

How stable is the system at 66MHz ?

I've only run those DOS benchmarks and booted Windows 3.11. I still have a lot of work to do on the system, e.g. sound, VGA, memory expansion.

pshipkov wrote on 2022-02-06, 09:01:

Will be interesting also to see Wolf3D test. Also what is the disk i/o reported by CORETEST or other benchmarking utils ?

I remember our Wolf3D bench sets being very different. I'll have to use yours. I think I have it somewhere.

pshipkov wrote on 2022-02-06, 09:01:

AHA-1542C# handles out of spec ISA frequencies quite well, but you need to counter that with lowering the speed setting from BIOS otherwise data corruption can occur.

Lowering which speed settings and from which BIOS (MB or Adaptec)? MB options are nil, so I assume you are referring to the Adaptec SCSI DMA speed? Do you recall what's the max non-corruption producing DMA speed for 16.7 MHz ISA?

pshipkov wrote on 2022-02-06, 09:01:

Also, it exhibits write access problems with some of the DLC/SXL 386 upgrade CPUs.

If I recall, the corruption issue w/154X cards is due [exclusively?] to the SXL, with the DLC not being afflicted. Did your tests also show the DLC was an issue here?

pshipkov wrote on 2022-02-06, 09:01:

It will be interesting to see if it work well with SLC processors and if bring any advantage over IDE.
These VLSI-200 based mobos handle IDE really well and outdo any ISA SCSI card i know of.
In other words their IDE perf is clock-to-clock faster than SCSI.

I'm probably going to stick with SCSI throughout, although I should have UMC, Winbond, and Acer based ISA IDE cards.

One main issue with this board is the minimal number of 16-bit ISA slots. I might need to use an 8-bit sound card to free up some 16-bit slots. Can pretty much any 16-bit ISA sound card work in an 8-bit ISA slot, just not have access to IRQ's above 7?

Plan your life wisely, you'll be dead before you know it.

Reply 126 of 174, by pshipkov

User metadata
Rank Oldbie
Rank
Oldbie

Yes, what I meant was lowering AHA's DMA speed.
With 286 cpu all AHA-1542# controllers did well at 25mhz + 0-ws + 10ms/s DMA speed. After that dma speed has to go down.
SXL cpus result in jnherently unstable SCSI. Checked old notes about DLC + SCSI stability, they are conflicting. Some notes say that DLC+SCSI was stable, other say otherwise. That is on 286 hardware.

Had plenty of trouble with SXL/DLC+SCSI on 386 class rust. SCSI adapters didn't to well on highly optimized systems running at 40+ mhz.
But that's outside of the discussion here.

No idea if each and every sound card can work in 8-bit ISA slot. The common SB ones are fine with that.

retro bits and bytes

Reply 127 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

When you write "10ms/s", I assume you mean "10 MByte/s"? The default is 5.0 MB/s. If I recall, only the 10 MB/s setting had any performance benefit compared with sub-10MB/s settings. So far, every app I have opened in Win3.11 has been able to run, including IE5 w/google. I did a network transfer of 50 MB, no problem.

Yes, I need to replace the 154X in my AMI Mark V Baby Screamer + SXL2-66 with a 152X card for those fringe cases of HDD corruption. It is not an easy task because that system multi-boots between Win3.11, w95, NT3.51, and NT4.

So far, my ESS 1868F is working fine in the 8-bit slot. Tried playing waves in Win3.11, DOOM in DOS w/both internal OPL3 clone and add-on wavetable card.

Plan your life wisely, you'll be dead before you know it.

Reply 128 of 174, by pshipkov

User metadata
Rank Oldbie
Rank
Oldbie

I meant 10mb/s.
Was typing on a tablet which is error prone for non-standard words.
Varying the scsi speed setting between 5-to-10 actually has measurable impact on i/o metrics.

Ok so far it looks like SLC is a great cpu upgrade path for 286 system, especially if SCSI is in the mix.
SXL climbs a bit higher but no SCSI.
DLC cannot go past 45MHz, so not as good as the other two.

retro bits and bytes

Reply 130 of 174, by Sphere478

User metadata
Rank l33t++
Rank
l33t++
pshipkov wrote on 2022-02-07, 02:03:
Sphere478 wrote on 2022-02-06, 10:02:

This is exactly the kind of stuff I’m here for. I love it! Good job!!!

that is evident from your K6 adventures

😁

So is this the fastest surface mount that will fit on this interposer board?

Sphere's PCB projects.
-
Sphere’s socket 5/7 cpu collection.
-
SUCCESSFUL K6-2+ to K6-3+ Full Cache Enable Mod
-
Tyan S1564S to S1564D single to dual processor conversion (also s1563 and s1562)

Reply 131 of 174, by pshipkov

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote on 2022-02-07, 00:25:

Yes, I need to replace the 154X in my AMI Mark V Baby Screamer + SXL2-66 with a 152X card for those fringe cases of HDD corruption. It is not an easy task because that system multi-boots between Win3.11, w95, NT3.51, and NT4.

That's not fun.
I would probably rush install the OSes to populate boot files, delete the win311, win95 and winnt dirs. Then copy your existing dirs there. Will cut the pain enormously.
But win95 and winnt on this machine ? Do you have an use for them, or for completeness ?

retro bits and bytes

Reply 132 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

The reason for W95 and NT4 on the SXL2-66 is 2-fold. First I wanted to see how well they would run and get a feel for their responsiveness. Secondly, and is what made me install these OSes from the onset, was the developers of HWINFO and CPU-Z were looking for some perf, troubleshooting, and log files with this CPU on these OSes.

Installing w95, nt3.51, and nt4 is too slow on an SXL2. The issue with swapping the SCSI card comes mainly from NT3.51 and NT4. I need to have the 152X's driver loaded by NT for the system to boot, so I plan on inserting the 152X with the 154X still installed so that I can set the 152X's driver to load at boot. Then hopefully, I can pull out the 154X and everything will work. We'll see. I remember trying this one evening when I was tired and ran into a snag, so I've been avoiding round two.

Concerning the DMA speed on the SCSI card, the options in MB/s are 3.3, 5.0, 6.7, 8.0, and 10.0. If I recall right, my past tests showed 3.3, 5.0, 6.7, and 8.0 MB/s options having roughly the same benchmark results. The only setting with any noteworthy benefit was the 10.0 MB/s setting. This probably is different for each system, so I cannot really generalise.

Plan your life wisely, you'll be dead before you know it.

Reply 133 of 174, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

I used a laptop machine years back with Win95a set up on a Cx486SLC-40 with 12MB installed and it was surprisingly okay. All the GUI stuff seemed to be as fast as 486es get, only when you needed horsepower like decompression, decryption etc that you noticed it. Certainly wasn't as "ploddy" as an Intel 486SL33 machine I've got, which you'll recall is actually a full DX with power saving, not an SX core. In fact the AST 486 dx4-75 laptops we had seemed doggier just getting around in windows, but the ooomph would have won out on the math intensive stuff.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 134 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I ran a few tests from pshipkov's util pack. Note that SCSI DMA left at 5.0 MB/s.

Wolf_286.exe TIMEDEMO = between 22.1 - 22.3 fps (no sound), or between 20.9 - 21.0 fps (with sound)

Speedsys score = 3.36. I need more RAM to run more tests.

3DSpace = 42%

DMA Speed = 1040K/s

CABT = 4.67 seconds

ATFLOAT = 5.5

Memspeed, 512KB, =8000 KB/s

Waitstat, clock=72.5 MHz, 11 waitstates, perf=50.2

AGSI reports 66.5 MHz. So far the only program to get the speed right on.

CoreTest 2.7, block size 64KB, ===================> 2965.3 KB/s
(what's the fastest CoreTest speed you've recorded on the ISA bus?)

DiskMeter 1.3, block size 64KB
Random =========> 3333.3 KB/s
Sequential ========> 3375.5 KB/s
Buffered =========> 3382.6 KB/s
Max seek =========> 0.3 ms
Perf index ========> 1249
(what's the fastest DiskMeter speed you've recorded on the ISA bus?)

Maxspeed (not in pshipkov's pack), which uses some fancy timing algorithm to measure the speed, reports between 64-69 MHz

Plan your life wisely, you'll be dead before you know it.

Reply 135 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I tried to adjust the Adaptec DMA speed from 5.0 MB/s to 8.0 MB/s, but the SCSI card was, upon reboot, unable to find the HDD and CD-ROM during bus scan. So I set it to 6.7 MB/s and re-ran CoreTest and DiskMeter .

CoreTest
5.0 MB/s DMA --------> 6.7 MB/s DMA
2965.3 KB/s ----------> 3765 KB/s

DiskMeter

5.0 MB/s DMA --------> 6.7 MB/s DMA
3333.3 KB/s ----------> 3397 KB/s, RANDOM
3375.5 KB/s ----------> 3448 KB/s, SEQUENTIAL
3382.6 KB/s ----------> 3911 KB/s, BUFFERED

Looks like the buffered result is the one which has the benefit.

I'm a bit nervous to leave the DMA speed at 6.7 MB/s, given that 8.0 MB/s didn't work at all.

Plan your life wisely, you'll be dead before you know it.

Reply 136 of 174, by maxtherabbit

User metadata
Rank l33t
Rank
l33t
feipoa wrote on 2022-02-07, 11:38:
I tried to adjust the Adaptec DMA speed from 5.0 MB/s to 8.0 MB/s, but the SCSI card was, upon reboot, unable to find the HDD an […]
Show full quote

I tried to adjust the Adaptec DMA speed from 5.0 MB/s to 8.0 MB/s, but the SCSI card was, upon reboot, unable to find the HDD and CD-ROM during bus scan. So I set it to 6.7 MB/s and re-ran CoreTest and DiskMeter .

CoreTest
5.0 MB/s DMA --------> 6.7 MB/s DMA
2965.3 KB/s ----------> 3765 KB/s

DiskMeter

5.0 MB/s DMA --------> 6.7 MB/s DMA
3333.3 KB/s ----------> 3397 KB/s, RANDOM
3375.5 KB/s ----------> 3448 KB/s, SEQUENTIAL
3382.6 KB/s ----------> 3911 KB/s, BUFFERED

Looks like the buffered result is the one which has the benefit.

I'm a bit nervous to leave the DMA speed at 6.7 MB/s, given that 8.0 MB/s didn't work at all.

Adaptec has a DMA test in ROM you can use to safely test the higher speeds without risking data corruption. Open DOS DEBUG and go:

g=c800:9

replace c800 with the base address of your ROM if different

Reply 137 of 174, by pshipkov

User metadata
Rank Oldbie
Rank
Oldbie

The best i was able to get on an entirely long-term stable 286 hardware path is posted here.
In short:
IDE - 4Mb/s
Wolf3D (no sound) - 22.9 fps
Superscape - 12.8 fps

with SXL2 cpu (post linked above in thread, i believe):
IDE - 1.5 - 2Mb (depends which test you ask) <- meh, apparently SCSI helps a lot in this case, but with Write access failures = no go
Wolf3D - 29.8 fps
Superscape - 18.1 fps
Doom - 8.4 fps
PCP Bench - 4.1 fps
Since then i was able to go to ~64 mhz which further bumps perf, but didn't capture/share the results.

If 8Mb/s SCSI DMA speed appears to work - i would go with it.

@maxtherabbit
So what happens after set that flag ?
Can you explain ?

retro bits and bytes

Reply 138 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Actually, it is only the 6.7 MB/s and slower DMA setting would boot. 8.0 MB/s was a no go.

I was asked to show my 286 build for this Evergreen 486 SuperChip, so here's some photos.

Here's the case. I'm fairly fussy about matching all the shades of beige. I'm also particular about not leavaing the case badge location empty, so I modified the Cyrix 386-to-486 logo to make it a 286-to-486, then added the SLC2 component to it.

01-Case-1.JPG
Filename
01-Case-1.JPG
File size
402.77 KiB
Views
767 views
File license
CC-BY-4.0
02-Case-2.JPG
Filename
02-Case-2.JPG
File size
164.47 KiB
Views
767 views
File license
CC-BY-4.0
03-Inside_the_case_1.JPG
Filename
03-Inside_the_case_1.JPG
File size
1.06 MiB
Views
767 views
File license
CC-BY-4.0
04-Inside_the_case_3.JPG
Filename
04-Inside_the_case_3.JPG
File size
474.4 KiB
Views
767 views
File license
CC-BY-4.0
05-Testbed_1.JPG
Filename
05-Testbed_1.JPG
File size
602.26 KiB
Views
767 views
File license
CC-BY-4.0
Last edited by feipoa on 2022-02-25, 13:19. Edited 2 times in total.

Plan your life wisely, you'll be dead before you know it.

Reply 139 of 174, by feipoa

User metadata
Rank l33t++
Rank
l33t++

An image of the motherboard and the various expansion cards I'm using. The memory expansion card adds 3 MB, but as will be shown later, the speed in this 3 MB space is slower than the mainboard's memory. Adding 70 ns 256Kx1 DRAM doesn't help any compared to 120 ns. The RAMDAC on this ET4000AX card is pretty slick - crystal clear 800x600x15bit and x16bit. I've set the system to 15bit.

Although I'm using a PS/2 mouse on a modified keyboard controller, I decided to add an RS-232 card just in case I wanted to test some serial mice. I've also added the REEL Magic Maxima card, hoping to test some REEL Magic games and MPEG-1 decoding at some point in the future. It comes with a pass-through cable, which doesn't show any noticeable noise up to 800x600. At 1024x768, you start to see it. I add ethernet to all my systems for file convenience. I find the Etherlink III 3C509B works best on slower 286 and 386 class hardware.

06-Motherboard_MB-1216V.JPG
Filename
06-Motherboard_MB-1216V.JPG
File size
1.73 MiB
Views
764 views
File license
CC-BY-4.0
07-Cards_Everex_EV-159_RAM_3000_Deluxe.JPG
Filename
07-Cards_Everex_EV-159_RAM_3000_Deluxe.JPG
File size
1.62 MiB
Views
764 views
File license
CC-BY-4.0
08-Cards_Cardinal_ET4000AX_graphics.JPG
Filename
08-Cards_Cardinal_ET4000AX_graphics.JPG
File size
273.74 KiB
Views
764 views
File license
CC-BY-4.0
09-Cards_ESS_1868_and_REEL_Magic_Sigma_Designs.JPG
Filename
09-Cards_ESS_1868_and_REEL_Magic_Sigma_Designs.JPG
File size
590.08 KiB
Views
764 views
File license
CC-BY-4.0
10-Cards_AHA-1542CP_Etherlink-III_and_Startech_serial.JPG
Filename
10-Cards_AHA-1542CP_Etherlink-III_and_Startech_serial.JPG
File size
841.4 KiB
Views
764 views
File license
CC-BY-4.0

BUILD PHOTOS CONTINUED ON THE NEXT PAGE

Last edited by feipoa on 2022-02-26, 10:04. Edited 2 times in total.

Plan your life wisely, you'll be dead before you know it.