Well, I pin-pointed that weird SC2000 problem. It must be some random bug or hardware incompatibility since it only happens when the program is started in real mode. With EMM386 installed and the machine in V86 mode, SC2000 works fine. I still don't understand why it happens in Creative's diagnose.exe program. I might pull out the NiCE SuperEISA board again and see if the problem crops up there.

EDIT: The NiCE board seems to work fine, weird. Might want to try 8MB of memory in it though since I only put 4MB in.

My hunch somehow proved correct. Swapping the keyboard controller fixed the odd behavior. Now I just need to find another chip. Anyone got a spare i8042 compatible keyboard controller laying around? 😜

I'm running into some odd problems with memory speed on this board. For some reason, memory throughput numbers seem to drop a bit when L2 writeback cache is enabled in the BIOS. Below is my Speedsys and Cachechk results. Don't ask me why speedsys doesn't detect any L2 cache when the BIOS Memory Speed is set to "Fastest" (Both the Tyan and NiCE board does this). Anyone know what might cause these odd results?

L2 Writeback enabled:

1                Cache/Memory Benchmark
2                ÚÄÄÄÄÄÄÄÄÄÄÄÄÄÄÂÄÄÄÄÄÄÄÄÄÄÄÄÄÄÂÄÄÄÄÄÄÄÄÄÄÄÄÄÄÂÄÄÄÄÄÄÄÄÄÄÄÄÄÄ¿
3                ³     Read     ³     Write    ³     Move     ³    Average   ³
4ÚÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄ´
5³ Cache Level 1 ³    94.45 MB/s³    41.43 MB/s³    41.13 MB/s³    59.00 MB/s³ 
6³ Memory        ³    31.33 MB/s³    41.43 MB/s³    21.60 MB/s³    31.45 MB/s³ 
7ÀÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÙ
8 ÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄ

Cachechk results:

1 CACHECHK v4 2/7/96  Copyright (c) 1995 by Ray Van Tassle. (-h for help)
2 CMOS reports: conv_mem= 640K, ext_mem= 11,264K, Total RAM= 11,904K
3 "GenuineIntel"  486 DX4 Clocked at 99.8 MHz
4 Reading from memory.  
5 MegaByte#:      --------- Memory Access Block sizes (KB)-----
6       1    2    4    8   16   32   64  128  256  512 1024 2048 4096 <-- KB
7 0:   11   11   11   11   11   26   26   26   26   71   --   --   --    æs/KB
8 1:   11   11   11   11   11   26   26   26   26   71   71   71   71    æs/KB
9 2  3  4  5  6  7  8  9 10 11  <--- same as above.
10
11 Extra tests----
12 Wrt  16   16   16   16   16   16   16   16   16   16   16   16   16<-Write mem
13 This machine seems to have both L1 and L2 cache. [read]
14    L1 cache is  16KB -- 102.5 MB/s  10.2 ns/byte  (665%) (239%)   3.9 clks
15    L2 cache is 256KB --  42.8 MB/s  24.5 ns/byte  (277%) (100%)   9.3 clks
16    Main memory speed --  15.4 MB/s  68.1 ns/byte  (100%) [read]  25.9 clks
17    Effective RAM access time (read ) is 272ns (a RAM bank is 4 bytes wide). 
18    Effective RAM access time (write) is  61ns (a RAM bank is 4 bytes wide). 
19    "GenuineIntel"  486 DX4 Clocked at 99.8 MHz.  Cache ENABLED.
20 Options: -t0

L2 Writethrough enabled:

Cachechk results:

1 CACHECHK v4 2/7/96  Copyright (c) 1995 by Ray Van Tassle. (-h for help)
2 CMOS reports: conv_mem= 640K, ext_mem= 11,264K, Total RAM= 11,904K
3 "GenuineIntel"  486 DX4 Clocked at 99.8 MHz
4 Reading from memory.  
5 MegaByte#:      --------- Memory Access Block sizes (KB)-----
6       1    2    4    8   16   32   64  128  256  512 1024 2048 4096 <-- KB
7 0:   11   11   11   11   11   26   26   26   26   30   --   --   --    æs/KB
8 1:   11   11   11   11   11   26   26   26   26   30   30   30   30    æs/KB
9 2  3  4  5  6  7  8  9 10 11  <--- same as above.
10
11 Extra tests----
12 Wrt  24   24   24   24   24   24   24   24   24   24   24   24   24<-Write mem
13 This machine seems to have both L1 and L2 cache. [read]
14    L1 cache is  16KB -- 102.5 MB/s  10.2 ns/byte  (280%) (238%)   3.9 clks
15    L2 cache is 256KB --  42.9 MB/s  24.5 ns/byte  (117%) (100%)   9.3 clks
16    Main memory speed --  36.6 MB/s  28.6 ns/byte  (100%) [read]  10.9 clks
17    Effective RAM access time (read ) is 114ns (a RAM bank is 4 bytes wide). 
18    Effective RAM access time (write) is  92ns (a RAM bank is 4 bytes wide). 
19    "GenuineIntel"  486 DX4 Clocked at 99.8 MHz.  Cache ENABLED.
20 Options: -t0

1                Cache/Memory Benchmark
2                ÚÄÄÄÄÄÄÄÄÄÄÄÄÄÄÂÄÄÄÄÄÄÄÄÄÄÄÄÄÄÂÄÄÄÄÄÄÄÄÄÄÄÄÄÄÂÄÄÄÄÄÄÄÄÄÄÄÄÄÄ¿
3                ³     Read     ³     Write    ³     Move     ³    Average   ³
4ÚÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄÅÄÄÄÄÄÄÄÄÄÄÄÄÄÄ´
5³ Cache Level 1 ³    94.45 MB/s³    41.49 MB/s³    41.19 MB/s³    59.04 MB/s³ 
6³ Memory        ³    33.51 MB/s³    41.49 MB/s³    22.60 MB/s³    32.53 MB/s³ 
7ÀÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÄÄÄÄÄÄÄÙ
8 ÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄÄ

NJRoadfan wrote:

I'm running into some odd problems with memory speed on this board. For some reason, memory throughput numbers seem to drop a bit when L2 writeback cache is enabled in the BIOS. Below is my Speedsys and Cachechk results. Don't ask me why speedsys doesn't detect any L2 cache when the BIOS Memory Speed is set to "Fastest" (Both the Tyan and NiCE board does this). Anyone know what might cause these odd results?

I've noticed all sorts of weird things with cache and memory timings on SiS 486 boards. I have never found concrete answers to them, but I suspect the chipset (or BIOS) does some auto manipulation of the timings when the user manually sets speed combinations higher than the system can run.

The L2 is being detected in the graph, it's just not being reported for some strange reason. I wonder why it shows up on my board but not on yours. Maybe it's a chipset revision related issue? I'll have to test your writeback/writethrough issue too. I seem to recall getting a performance boost though. Does the BIOS have an option for tag length? If so, when it writeback mode make sure it's set to 7+1 rather than 8+0.

Anonymous Coward wrote:

The L2 is being detected in the graph, it's just not being reported for some strange reason. I wonder why it shows up on my board but not on yours. Maybe it's a chipset revision related issue? I'll have to test your writeback/writethrough issue too. I seem to recall getting a performance boost though. Does the BIOS have an option for tag length? If so, when it writeback mode make sure it's set to 7+1 rather than 8+0.

No option for tag length on either board. Looking at the physical chips, the Tyan looks like it was produced a few months after the NiCE. Its BIOS is dated 6/1993 with chip date codes from late 1993, while the NiCE appears to be from early 1993. The cache tag setup is different between the boards, the NiCE uses 3 tag chips, while the Tyan only has space for 2 (256k only uses one tag chip on that board).

Have you tried adding the second Tag chip on your Tyan board? On second thought maybe it won't matter. I consulted with TH99, and it seems the second tag is only needed if you use 1024kb cache.

What date is given in your BIOS splash screen? Mine is 6/6/92.

Anonymous Coward wrote:

Have you tried adding the second Tag chip on your Tyan board? On second thought maybe it won't matter. I consulted with TH99, and it seems the second tag is only needed if you use 1024kb cache.

What date is given in your BIOS splash screen? Mine is 6/6/92.

Both boards have the 6/6/1992 AMI Hi-Flex core. The 1993 date on the Tyan appears above the memory test count along with the board's BIOS revision. My NiCE board displays revision 1.2.1 on the same line, but with no date.

I probably already asked, but I forgot the answer. Have you tried an AMD 5x86-133 yet?

Anonymous Coward wrote:

I probably already asked, but I forgot the answer. Have you tried an AMD 5x86-133 yet?

I don't have one on hand.... yet.

I seem to recall a lot of older boards had compatibility issues with the IntelDX4, even with a VRM. The AMD chips were said to work better with such boards.

I do have a few DX2/66 CPUs laying around. Right now I'm waiting for my 16MB SIMM to return from RMA.

Can you do some benchmarks with the DX/2 66 to see if the memory scores exhibit the same problems?

What kind of DX/2 is it? A 5V model with WT cache would be preferable.

A plain old stack of Intel DX2/66s, nothing special. They all came with the EISA boards (some with their warranty void if removed stickers in tact!) so I doubt they are write back chips. Heck, I didn't even know they made write back ones.

Intel's version was probably only used by OEMs, but the AMD and Cyrix writeback models were fairly common.

My Nice Super EISA board is now non-functional. I've exhausted pretty much all possibilities, and I can't revive it. I now have a Tyan S1437.

I am not impressed with the Tyan for a couple of reasons. At 33MHz, I am unable to get "FASTEST" memory settings going unless I set the cache timings to their slowest settings, which makes it slower than "FASTER" with optimal settings. I am using a 10ns tag and 15ns cache. I will try different chips later.

The DRAM interleave option in the BIOS appears to be useless. I'm not really sure what's going on. Enabling or disabling it has no effect whatsoever on memory scores, whereas on my Super EISA I could see a dramatic difference between the two. There are a few undocumented jumeprs and an empty socket on the Tyan. Maybe something else needs to be configured to get it working?

Overall build quality on the Super EISA seems to be higher overall. The only thing the Tyan has going for it is that it uses standard tag RAM chips, which doesn't help me much since memory speed is so much slower anyway.

I ran both Cyrix and AMD 5x86 chips in my tests.

L2 Writeback cache on this board appears to be somehow broken. At least according to speedsys it is slowing down memory substatially while not improving the L2 cache bandwidth.

I was able to get DRAM interleave working by disabling the WB cache. Not sure why having WB enabled was causing this. It's rather odd. But that's okay, because I set a record of 51MB/sec uncached DRAM running at 40MHz in "FASTER" mode. That's insanely good. I will try replacing the SRAM chips eventually to see if "FASTEST" is possible.

Of other note, there is an empty socket next to the cache chips. It's 22 pin. There are two 28-pin sockets for TAG ram nearby as well. I suspected that the 22-pin socket might be for a WB dirty bit since I've seen this on a number of other boards. I had a 15ns 64kx1 chip floating around so I popped it in to see if it could fix WB mode. It didn't explode, but nothing happened. It would be nice to have the manual.

51 MB/s for the DRAM - Is that the combined Speedsys speed for main memory, or just the read speed? Is that speed for the AMD X5-160 or Cyrix 5x86-120? Is the system stable using these CMOS settings? Have you been able to install Win9x without issue? What about running Quake timedemo? What is the reported speed when you disable interleave mode?

51mb/sec is the combined read/write/move speed for main memory. With interleave off I believe it was 36mb/sec. I haven't tested the system for hours on end, but I ran DOOM and GTA1 for a while. I suspect the system should be pretty stable though, because I'm using less agressive memory timings than on my Super EISA. The results were obtained with a Cyrix 5x86-120 with enhancements turned on. Combined speed with AMD 5x86 at 160 and 133 are both about 48MB/sec.

At the moment the system is kind of in pieces because I'm attempting to upgrade the cache. There are problems though. It is not letting me upgrade to 512kb or 1024kb with cache that is known to be good.

What are the symptoms? The system doesn't detect the proper amount and shows 0 cache, or 256 kb cache? Or the system does not turn on? Or turns on and beeps? Did you tune down the CMOS timings?