VOGONS


First post, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I am tyring to get this system setup, with an AMD X5-160 ADZ at 40x4.

At 40 MHz FSB, setting the cache write cycle to 2 and DRAM Speed to Fastest gives really nice cachechk speeds:

L1 Read: 165 MB/s
L2 Read: 75 MB/s
RAM Read: 51.5 MB/s
RAM Write: 83.6 MB/s

However the system is not stable in Windows. I need to set the cache write cycle up to 3 for the system to be stable enough in Windows to run the Quake 2 benchmark, or play mp3's continually.

Setting cache write cycle to 3 greatly lowers the Cachechk speeds as follows:

L1 Read: 165 MB/s
L2 Read: 75 MB/s
RAM Read: 39.4 MB/s
RAM Write: 55.7 MB/s

Strangely enough, it doesn't alter the L1/L2 cache speeds at all, only the RAM speed, which is what puzzles me. In checking the various DRAM Speed options of Normal, Fast, Faster, and Fastest, nothing above Fast increase the DRAM speed when cache write cycle is set to 3. I can also run the system with cache write cycle set to 2, however I cannot go above DRAM Speed: Faster. 2 with Faster gives the same memory throughput as 3 with fast/faster/fastest. 2 with Fastest gives the 51.5 MB/s again, but not stable enough in Wndows, even with 5 V going to the CPU.

I first thought that the Cache Write Cycle was the first character in the typical 2-1-2 or 2-1-1-1 which you may be used to. On other motherboards, going from 2-1-2 to 3-1-2 does not change the memory throughput. However, the BIOS Companion mentions that the Cache Write Cycle: Affects the data hold time for writes to DRAM.

Can anyone else with a Zida Tomato 4DPS motherboard see if the maximum cachechk memory read speed is 39.4 MB/s when the cache is set to 3-1-2 and everything else in the BIOS's chipset settings are optimised? I am thinking I must have pooched something on this board. Much appreciated!

Here are all the chipset settings for this motherboard,

ISA Bus Clock: 7.159 MHz, 1/3, or 1/4 of PCI Clock
LBD# Sample Point: End of T3 or End of T2
Cache Write Cycle: 2 CCLK or 3 CCLK
Cache Burst Read Cycle: 1 CCLK or 2 CCLK
L2 Cache / DRAM Cycle WS: 2 CCLK or ?
DRAM RAS to CAS Delay: 2 CCLK or 3 CCLK
DRAM Write Cycle: 0 WS or 1 WS
DRAM Write CAS Paline: 2 CCLK or ?
DRAM CAS Precharge Time: 1 CCLK or?
DRAM RAS to MA Delay: 1 CCLK or 2 CCLK
DRAM Speed: Normal, Fast, Faster, Fastest
DRAM Slow Refresh: Enabled or Disabled

L2 Cache Policy: Write Back or Write Through
L2 Cache Tag Bits: 7 bits or 8 bits

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

Reply 1 of 31, by Mithloraite

User metadata
Rank Member
Rank
Member

oops, just stumbled over this message only now, sorry. I had a plan of getting this very board and got it, but only the 133 CPU was working. Tomato itself was dead. Bad china +__+ Fortunately I didn't pay more than the CPU price can go...

That topic of mine, on "Abysmal performance",
AMD x5 133 - the Mystery of Abysmal Performance

does have some bios cache & memory setting screens corresponding to the fastest Speed Sys memory score I have (for this SiS chipset).

So if your cache chips and the simm modules are guaranteed to be good, but they somehow don't work with these fast bios settings (that were listed by me) - the logical conclusion it's the board that spoils up things.

Besause that's the same SiS chipset... Besides Lucky Star I've tested A-Trend and it could perform _exactly_ at the same speeds.

So, if your cache and simms are good...

Reply 2 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++

That is the conclusion I came up with, the SiS chipset just can't handle the 40 MHz speed on the fastest memory/cache settings [edit, when using single-banked 512K cache].

Last edited by feipoa on 2017-10-11, 23:12. Edited 1 time in total.

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

Reply 3 of 31, by 5u3

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote:

Can anyone else with a Zida Tomato 4DPS motherboard see if the maximum cachechk memory read speed is 39.4 MB/s when the cache is set to 3-1-2 and everything else in the BIOS's chipset settings are optimised?

My board is an Asus PVI-486SP3, but it might interest you that it behaves in the exact same way as your Tomato board with these settings.
I only use WfW 3.11 for Windows, but running 40 MHz FSB at the fastest cache timings never was a problem for me. However, I had a hard time finding PCI cards able to cope with the PCI bus speed. Went through several S3 Virges and 3Com 905 cards until I found reliable ones.

Reply 4 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++
5u3 wrote:
feipoa wrote:

Can anyone else with a Zida Tomato 4DPS motherboard see if the maximum cachechk memory read speed is 39.4 MB/s when the cache is set to 3-1-2 and everything else in the BIOS's chipset settings are optimised?

My board is an Asus PVI-486SP3, but it might interest you that it behaves in the exact same way as your Tomato board with these settings.
I only use WfW 3.11 for Windows, but running 40 MHz FSB at the fastest cache timings never was a problem for me. However, I had a hard time finding PCI cards able to cope with the PCI bus speed. Went through several S3 Virges and 3Com 905 cards until I found reliable ones.

So you get these cachechk results,
L1 Read: 165 MB/s
L2 Read: 75 MB/s
RAM Read: 51.5 MB/s
RAM Write: 83.6 MB/s

on the fastest settings with an AMD X5-160? Or do you get,

RAM Read: 39.4 MB/s
RAM Write: 55.7 MB/s
?

You can check cachechk with write using cachechk -w. In general, I use cachechk -w -d -t4 for write, and cachechk -d -t4 for read.

Do you have 256K double banked, or 512K single bank installed? I've noticed that 486 motherboards with 256K double banked can handle faster timings at 40 MHz.

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

Reply 5 of 31, by 5u3

User metadata
Rank Oldbie
Rank
Oldbie

Ah, now I've tried the -w parameter, the results become even stranger. 😊
Here is what I get:

Fastest timings:
RAM Read: 51.5 MB/s
RAM Write: 83.6 MB/s

Fastest timings, but with cache write cycle set to 3:
RAM Read: 39.4 MB/s
RAM Write: 83.6 MB/s

The cache chips are 32Kx8, 15ns.

Reply 6 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I am assuming you have 8 pieces of 32Kx8, not 4 pieces. Double-banked cache seems to tolerate faster settings.

Do you have 4 pieces of 128Kx8 for 512KB cache to test?

I found that increasing the CPU voltage helped somehwat with the stability of the system at 40 MHz and 512KB of cache, however Windows would eventually crash. Slowing down the cache write cycle from 2 to 3 made it stable, however slower. I think the trick to keep the speed would be to use double-banked cache, which is what you have done, but it decreases your cacheable memory range.

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

Reply 7 of 31, by 5u3

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote:

I am assuming you have 8 pieces of 32Kx8, not 4 pieces. Double-banked cache seems to tolerate faster settings.

Yes, my current board has eight 32Kx8 28-pin chips, however I also have a backup board (same model, newer revision) which came with four 64Kx8 32-pin chips. Never had any problems with these at 40 MHz as well.

feipoa wrote:

Do you have 4 pieces of 128Kx8 for 512KB cache to test?

Years ago I got some 128Kx8 chips from china, but they turned out to be defective or fake.

feipoa wrote:

I found that increasing the CPU voltage helped somehwat with the stability of the system at 40 MHz and 512KB of cache, however Windows would eventually crash.

I routinely run an Intel DX4 at 120 MHz and the AMD X5 at 160 MHz, both wouldn't be stable without raising the CPU voltage to 3.6V. The iDX4 also has problems with 80 (2x40) MHz at stock voltage.

feipoa wrote:

I think the trick to keep the speed would be to use double-banked cache, which is what you have done, but it decreases your cacheable memory range.

I've got 64MB RAM installed on my machine, where the upper half is hosting a RAM disk by default. This is OK with Win3.11, but I can imagine it doesn't go down so well with Win9x.

Reply 8 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++
5u3 wrote:
feipoa wrote:

I found that increasing the CPU voltage helped somehwat with the stability of the system at 40 MHz and 512KB of cache, however Windows would eventually crash.

I routinely run an Intel DX4 at 120 MHz and the AMD X5 at 160 MHz, both wouldn't be stable without raising the CPU voltage to 3.6V. The iDX4 also has problems with 80 (2x40) MHz at stock voltage..

With ever increasing voltages to the X5, the system became more stable at 40 MHz and cache write cycle set to 2. I took it up to 4 V, which seemed fine initially, but still crashed after 5 min. of mp3 playback in Win98SE.

Setting the cache write cycle back to 3 and 3.6 V on the CPU made the system very stable. So I have concluded (for now) the issue is with using single banked cache of 512K and a 40 MHz bus. I will experiment with other cache configurations at a later time. For a final setup system, I don't really want to run less than 512K of L2 cache.

Do you happen to have a photo of the 128k cache pieces you ordered from China? The 10 ns stuff I ordered from China earlier this year worked, even at a FSB of 83 MHz on a 430TX board as the TAG RAM (32k). But some small percent were DOA.

Does anyone else have a system setup with an AMD X5-160 using the fastest possible BIOS timings and 512K of single-banked cache on either a UMC or SiS motherboard?

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

Reply 9 of 31, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

I sold him the cache a few years back. I don't know why his modules were faulty. Mine work just fine. Should be the ISSI 15ns parts. I had my wife handle the shipping. Perhaps they were not handled properly and they were damaged from static discharge.

"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 10 of 31, by 5u3

User metadata
Rank Oldbie
Rank
Oldbie

Yep, the chips are ISSI 15ns, and to be fair, I can't really determine if they're defective, because I only have those two PVI-486SP3 boards to test them in, and maybe it's just an incompatibility with the board. Or the chips went through X-ray at customs or something funny like that. In the end I decided that 64MB cacheable area just wasn't worth the trouble. 😉

Reply 11 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++

The PVI-486SP3 board I have works fine with four ISSI 128kx8 pieces. If you have more than 4 pieces, you could try to single out the bad piece. What were your symptoms of fairure? I've acquired 128k8 chips from two different sources in the past. One shipment was 15 ns, the other 10 ns. The 10 ns chips were 10% DOA, while the 15 ns chips were 5% DOA. I have not had any trouble with the working chips.

Sometimes the failure is obvious, like when just one cache module is burning hot. Sometimes POST reports bad cache, while at other times, there are no obvious signs of failure except that Windows was unstable. The latter case is the most difficult to pin down, but I eventually solved it by plugging each module into an SRAM tester. It only takes one bad pin for Windows to show instability.

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

Reply 12 of 31, by 5u3

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote:

If you have more than 4 pieces, you could try to single out the bad piece. What were your symptoms of failure?

I got more than ten chips, and yes, I did spend a rainy afternoon swapping them through my boards. The symtoms were either no L2 cache being reported at all or the system crashing during POST. I also ruled out mechanical problems with the 32-pin sockets, since 64Kx8 chips worked fine in the same sockets. Back then, I didn't know anyone who had successfully upgraded their PVI-486SP3 to 512 KB cache, so I assumed the board was simply not up to the job.

Reply 13 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++

What revision of the PVI-486SP3 do you have? I beleive mine is rev. 1.8. Even with the cache swap around approach, there were still a few bad pieces I could only identify with the SRAM tester.

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

Reply 15 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++

This is somewhat of a different topic from the OP, but I didn't think it warrented a new thread. The question is, has anyone had success getting a POD working in a Zida Tomato 4DPS board?

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

Reply 17 of 31, by feipoa

User metadata
Rank l33t++
Rank
l33t++
5u3 wrote:

I had a hard time finding PCI cards able to cope with the PCI bus speed. Went through several S3 Virges and 3Com 905 cards until I found reliable ones.

I have recently been doing some testing of PCI NIC's on a SiS 496/497-based motherboard in WinNT 4.0. I tested 3 different PCI network cards, and in every case, the system would freeze upon a file transfer after 30-90 seconds or so. 33 MHz or 40 MHz PCI bus had the same result. The solution for reliable LAN tranfers was to set the cache write cycle from 2 to 3. This slows down the DRAM read speed by a good 25% though. The system was completely stable with the cache write cycle set to 2 just so long as LAN transfers were not initiated. Small file size LAN tranfers were OK, but anything prolonged caused a hang up. I am left wondering if perhaps there is a SiS PCI implementation problem related to PCI bus mastering and NICs.

I tried the system with an Adaptec Ultra Wide 2 SCSI adapter, a Promise Ultra100, and a Promise SATA0150, all of which were stable enough to install WinNT and W9x, but when it came to LAN transfers, the system froze unless the cache write cycle was set to 3.

A 25% loss in memory read speed was too much for me. I was able to leave the cache write cycle set to 2 and do all the LAN tranfers I wanted if I used an ISA NIC.

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

Reply 18 of 31, by smeezekitty

User metadata
Rank Oldbie
Rank
Oldbie

I know my CPU is quite a bit slower than a 5x86 but I am not able to get near your cache and memory speeds.

feipoa wrote:
I have recently been doing some testing of PCI NIC's on a SiS 496/497-based motherboard in WinNT 4.0. I tested 3 different PCI […]
Show full quote
5u3 wrote:

I had a hard time finding PCI cards able to cope with the PCI bus speed. Went through several S3 Virges and 3Com 905 cards until I found reliable ones.

I have recently been doing some testing of PCI NIC's on a SiS 496/497-based motherboard in WinNT 4.0. I tested 3 different PCI network cards, and in every case, the system would freeze upon a file transfer after 30-90 seconds or so. 33 MHz or 40 MHz PCI bus had the same result. The solution for reliable LAN tranfers was to set the cache write cycle from 2 to 3. This slows down the DRAM read speed by a good 25% though. The system was completely stable with the cache write cycle set to 2 just so long as LAN transfers were not initiated. Small file size LAN tranfers were OK, but anything prolonged caused a hang up. I am left wondering if perhaps there is a SiS PCI implementation problem related to PCI bus mastering and NICs.

I tried the system with an Adaptec Ultra Wide 2 SCSI adapter, a Promise Ultra100, and a Promise SATA0150, all of which were stable enough to install WinNT and W9x, but when it came to LAN transfers, the system froze unless the cache write cycle was set to 3.

A 25% loss in memory read speed was too much for me. I was able to leave the cache write cycle set to 2 and do all the LAN tranfers I wanted if I used an ISA NIC.

Now you are going to make me need to grab a PCI NIC and try it on my board

Reply 19 of 31, by idspispopd

User metadata
Rank Oldbie
Rank
Oldbie

Interesting. Do you know which chips those network cards use? Perhaps some other chip that doesn't use bus mastering would work better?