• Please review our updated Terms and Rules here

5x86 upgrade benchmarks worse than DX2-66? (TopBench)

J

jmetal88

Veteran Member
Joined
Jul 23, 2010
Messages
878
Location
Derby, KS
I decided to run a few benchmark tests on my 486 machine today, and to my surprise, when I run Trixter's TopBench utility, my machine has a better score with the original DX2 installed than it does when I install a 5x86 upgrade that should be clocked at 133 MHz. Another oddity, however, is that the machine appears more stable with the 5x86 installed than it does with the original DX2 installed (with the DX2, I got a runtime error the first time I started TopBench, and that has never happened with the 5x86). Does anyone know why this might be?

Incidentally, I also tried the 5x86 locked to a 3x multiplier (instead of 4x) with the motherboard set to 40MHz. This resulted in a similar instability to the original DX2 (runtime error when I first tried to start TopBench), but still worse benchmark scores after I did get it started.

EDIT: When I use -r to get TopBench to return the raw speed in MHz, it does say I'm running at just slightly above 133MHz. I wonder if it's a cache thing?

EDIT 2: System performance is much, much worse when I turn cache off in the BIOS. TopBench says I match a fast 286 or a slow 386 with the cache off. With the cache on, it says I match a 486 DX2 system, with the 5x86 installed. With the DX2, it still says I match a DX2 system, but it's one that was put into the database with a better score.
 
Last edited:
So, I was reading the 'speed issues' thread, and someone recommended running CACHECHK. I tried it and it appears to be indicating that only my CPU's on board cache is functioning. The motherboard has a socketed/upgradeable cache, and I'm pretty sure the chips are real cache chips, so it either isn't set up right or one or more of the chips is damaged.

Upon inspection, jumpers are set correctly for 128k cache, and 128k is installed. BUT, it says there's supposed to be an 8k chip in the tag socket, and I have a 32k chip there. So could it be that having a too large tag chip is causing it not to be addressed correctly? Maybe someone was planning a 256k upgrade that they never finished, as for 256k it IS supposed to be a 32k tag chip. Interestingly, information I've found on this board online does say it uses a 32k tag chip for 128k of cache, which contradicts the silkscreen on the board.

EDIT: It's getting too late for me to want to try right now, but I'm wondering if the BIOS is disabling external cache for the AMD processor, and that's why the DX2 scores better on the benchmark. That seems like that'd be a weird thing to do, but this motherboard has some other odd things going on due to its all-in-one design anyway.
 
Can You tell us what part number is written on the cache chips and tag chip, and how many of each?

I say this because there are people selling s3 pci video cards on ebay as 512k when they are 1mb, also 1mb when they are 2mb, and this is because
they don't know how to read the part number correctly. 512k x 16bit is 1mb, 512k x 4bit is 256k.
 
The correct TAG SRAM is required for motherboard cache to function properly. Using motherboard cache without a TAG SRAM or with a TAG SRAM of the wrong size usually causes severe performance degradation.

The CPU uses comparator logic in the TAG chip to match cache entries in the motherboard SRAM. If it doesn't have this, then things slow way down due to the CPU having to go out to the main system memory for whatever it was looking for.

It may speed your system up if you disable motherboard cache completely so the CPU doesn't have to waste time trying to search through it.
 
GiGaBiTe said:
The correct TAG SRAM is required for motherboard cache to function properly. Using motherboard cache without a TAG SRAM or with a TAG SRAM of the wrong size usually causes severe performance degradation.

The CPU uses comparator logic in the TAG chip to match cache entries in the motherboard SRAM. If it doesn't have this, then things slow way down due to the CPU having to go out to the main system memory for whatever it was looking for.

It may speed your system up if you disable motherboard cache completely so the CPU doesn't have to waste time trying to search through it.
Click to expand...

Well, do you think the silkscreen is right when it says 8k, or that the multiple pages on the internet saying 32k are correct? I'm not sure how to disable the motherboard cache by itself. There's one cache option in the BIOS and it seems to disable the CPU cache as well, based on the results I've seen so far. Do you think I could just remove the cache jumpers? It's an ACER V10 motherboard, by the way.
 
Imperious said:
Can You tell us what part number is written on the cache chips and tag chip, and how many of each?

I say this because there are people selling s3 pci video cards on ebay as 512k when they are 1mb, also 1mb when they are 2mb, and this is because
they don't know how to read the part number correctly. 512k x 16bit is 1mb, 512k x 4bit is 256k.
Click to expand...

The cache chips and the tag SRAM chip are all the same part number: UM61256FK-15. Should be 256k x 8bit, or 32k. There are four in sockets marked 'BANK 0' and one by itself above the main cache banks. There's also a CY7C166-15PC chip alongside the tag chip that's a 16kx4 SRAM. The socket the silkscreen references as 8k or 32k is U59, and that 16kx4 SRAM looks to be installed in U78, if I'm reading the numbering correctly.
 
If you have a regular SRAM in the TAG SRAM socket, that explains your problem. TAG SRAM has additional comparator logic in addition to having SRAM. Using a regular SRAM in a TAG socket won't work.
 
Well, the stuff that's in there is what was on the board when I got it, so if it's set up wrong, it's been set up wrong for a long time. How can I tell if it's supposed to have regular SRAM or a special TAG SRAM? There's not a lot of information on this board out there. All the board says is '8k' and '32k'.
 
May have found the problem. One of the cache chips has 3 or 4 mangled pins, and it wasn't evident until I took them all out to try re-seating them. I will try to see if they can be straightened. If not, I might be breaking them off and soldering on some bits of wire to try to work around the problem (I only want to buy new parts as a last resort).

EDIT: Edited wrong post.
 
Last edited:
Alright, CACHECHK successfully finds and tests the L2 cache now that I unbent the pins on that chip. It actually looked like that chip had actually been soldered in to something previously. The bent pins all had little balls on the ends of them. I wonder why? None of the other cache chips were like that. I'll run the benchmarks again now and see how they do.

EDIT: Yup, benchmark is much better now. Actually matches another 5x86 machine in the database. Seems like I might have been reading the original benchmark scores backwards, too, as I was assuming lower was better, but now it's obvious that in the overall rank a higher number is actually better (the DX2 was about 10 lower in overall rank than the 5x86 when I did the tests yesterday. I probably don't need to run the DX2 tests again today now that I realize a higher overall rank number is better).
 
Last edited:
Okay, the improvement I'm most excited about? I can set my graphics card to 72Hz refresh now without causing random hang-ups in Windows!

I seriously had no idea why it was having trouble with that in the first place, but now I guess it kind of makes sense that the cache could have been the problem.
 
I seem to recall there was also some utility to enable the L1 cache on the CPU...? I remember doing something like that back in the day when I was running a 5x86-133.

Wesley
 
You must log in or register to reply here.
Back
Top