VOGONS


Cachable Memory Info

Topic actions

First post, by alfonso

User metadata
Rank Newbie
Rank
Newbie

Hi all!

New lurker to Vogons here. Thank you for all the great posts and info shared by everyone here. I have benefited greatly from the discussions and answers shared here.

I recently became the proud owner of an IBM Aptiva machine being thrown away. It currently has the following specs:
V70MA motherboard
64MB PC100 RAM (1xDIMM)
K6-2 350mhz CPU
ATI Radeon 7000 64MB (PCI)
Windows 98

I have seen discussions about "cachable memory" with respect to the motherboard documentation and according to those docs I should only use 128mb PC100 RAM total to be in the "cachable"/max performance configuration. I am upgrading to a K6-2+ 450 CPU (ACZ) which will have its own 128kb L2 cache. For this processor, how do I go about finding the "maximum cachable memory" size? I ask because I would like to understand how the cache configuration ties into this performance-specific issue. How does 128kb L2 cache map to some amount of "cachable" memory? Can anyone take me through the factors/relevant math?

Would I benefit most from going up to 128mb PC100 Ram? 256mb? 512mb?

Thanks!

Reply 1 of 9, by Matth79

User metadata
Rank Oldbie
Rank
Oldbie

Supports 2x 128MB max, looks like all cached
Not listed for K6-2+ support, only lists voltages down to 2.2V, though there are 5 voltage select switches. No K6-2+ mod BIOS found, you may be out of luck for K6-2+ support

Reply 2 of 9, by alfonso

User metadata
Rank Newbie
Rank
Newbie

Thanks for telling me about that. Is there any risk in trying the CPU in this MOBO?

Reply 3 of 9, by Matth79

User metadata
Rank Oldbie
Rank
Oldbie

2.2V will run it hot, right on the edge of tolerable, (actually, they can take 2.2V with good heatsinking when going for high overclocks), the other option, with no CPU installed, is to probe the result of undocumented voltage switch settings
you might find a 2.0V, even a 2.1 would be more acceptable

Reply 4 of 9, by dionb

User metadata
Rank l33t++
Rank
l33t++

Cacheable area depends on the memory controller, which is in the motherboard chipset in boards from this era. That's an ALi Aladdin V. That's an interesting case as the answer depends on the exact chip revision. Early revisions A-F can cache max 128MB, G or H revision can cache 512MB. So check the revision of the M1542 chip on your board. It's the last letter of the second-last line of text on the chip.

If you use a CPU with its own L2 cache, it contains its own controller and tag RAM, so it's completely independent from whatever the motherboard does, whose cache then becomes L3 cache. The K6-3 and K62+/3+ can cache up to a theoretical 4GB of RAM. The L3 cache is of minimal impact on performance, but still the highest performance will be where it also can cache everything, so 128MB (if you have an old revision Aladdin V) - unless your software actually needs more memory; in that case you'll be thrashing to disk if you don't have enough. That is much, much slower than even completely uncached memory. So avoid thrashing at any cost.

Realistically speaking, Win98 on a CPU in this speed won't normally be running any software that needs more than 128MB, in fact 64MB was common in computers in those days. Win9x doesn't really benefit from huge amount of extra RAM the way NT-based Windows do.

Regarding voltage, 2.2V is pretty high for K6plus. The board has five voltage jumpers, so there should be 2^5=32 settings, whereas only seven are documented. Looking at the photo of the V7MA on TheRetroWeb I can't make out the voltage regulator, but the very similar V72LA has a Semtech SC1152CSW voltage regulator. Its datasheet is here. It indeed has a 5 bit VID selector with the combinations listed on page 5. Comparing that to the jumper settings, "0" in the VID corresponds to the switch set to "On" and "1" in the VID is "Off". Also the patterns match if reading VID 4-3-2-1-0 corresponding to SW2/4-5-6-7-8, so I'm guessing the V70MA has the same regulator or one with the same operating logic.

In that case, for nominal 2.0V operation, you'd want to set SW2/4-5-6-7-8 to On-On-On-On-Off

However voltage is only part of the challenge to get K6plus CPUs properly supported. BIOS also needs to cooperate and requires patching if not officially supported. You can find a list of supported boards here. No Acer V70MA I'm afraid. Exactly what to expect if you try it anyway is anyone's guess, particularly as this is an OEM motherboard - but at 2.0V it won't harm CPU or motherboard to try.

Reply 5 of 9, by alfonso

User metadata
Rank Newbie
Rank
Newbie

That's a lot of good information. Thank you for your post!

Reply 6 of 9, by alfonso

User metadata
Rank Newbie
Rank
Newbie

Given all that, I think I will avoid tinkering with the K6-2+ and avoid potentially damaging it. The CPU that arrived is an "ACZM" K6-2+ rather than the advertized "ACZ". Even on the forums here I cannot get a clear take on what the "M" connotes. It may be due for a return given the misrepresented Ebay listing.

For what it is worth, I have successfully upgraded the ram to 128mb PC 100 (2 x 64mb). Given these specs, do you have any recommendations for the next best upgrade to make? My ideal target is to run Half-Life 1 at some tolerable frame rate. Running Half-Life Uplink, I see choppy FPS except when standing in corners of levels. It might be 10-15 fps for general gameplay. It does not seem to matter whether I run Half-Life in 320x240 or 640x480 resolutions. Is there any part of this setup that might be obviously bottle-necking a better frame rate?

I have seen the numerous posts talking about socket 7 vs super socket 7 and ultimately it is unclear to me what really distinguishes one from another as far as CPUs go. It seems like the "+" suffix goes hand-in-hand with the super socket 7, but I am not clear if that is a complete definition. If I assume that my board only supports the non-super socket 7 chips, is the best I can do a K6-3 400 AHX or a K6-2 500 AFX/K6-2 550 AGR?

Reply 7 of 9, by dionb

User metadata
Rank l33t++
Rank
l33t++

We used to run Half Life 1 in the office on awful Compaq i810 systems with Celeron 433. You had to dial down graphics settings, but even with integrated VGA it was pretty playable. You have a far better GPU and higher memory bandwidth (as you're not sharing it with an IGP) -and resolution doesn't seem to matter so it's not GPU-limited.. That means CPU is the prime suspect. Half Life is pretty FPU-intensive and that's where (in that generation) Intel really shone and AMD was playing bad catchup.

If you want to improve performance with that build, try the K6-2+ after all. If you set it to 2.0V you're not risking any damage. To be sure, check the voltage regulator chip on your board. If it's the SC1152CSW, everything I posted above is correct. If it's not, post what it is and I can look up settings. TBH given the match with documented settings for higher voltage I'd be pretty confident regardless. It's the little black IC between the end of the DIMM slots and two MOSFETs with heatsink.

As for "Super" Socket 7, it's purely a marketing term so means whatever the marketeer using it wants it to mean. The only technical spec is Intel's Socket 7 specification. "Super" Socket 7 usually refers to 100MHz FSB and AGP (either a slot or integrated VGA using an internal AGP interface). Your board fulfils that, as does any K6-2 CPU with 100MHz FSB.
K6plus was never any part of official specifications (not even marketing ones) as they were never meant for retail use in desktop systems but purely intended as mobile CPUs. Unfortunately for AMD and fortunately for us, they didn't sell very well in their intended role so large numbers were left available at cheap prices for enthousiasts.

Given your board is an OEM board, documentation is scarce - just one manual with no explicit CPU support info (and no updates past its original date). Electrically it will support all So7 CPUs up to K6-3+. In terms of BIOS support, the newest BIOS is from late October 1998, but that's a Danish-language IBM OEM BIOS... in any event, that predates the 1999 K6-3 CPUs and 2000 K6plus CPUs. So regular CXT 100MHz FSB K6-2 CPUs are the newest that are definitely supported, anything newer might work, might not due to refusal to run unknown CPUs or due to missing microcode. In general, K6-3 will usually work wherever K6-2 does, so long as the 2.4V requirement can be met. K6plus tends to need dedicated BIOS support, but sometimes might work acceptibly.

If you want a real high-end late Socket 7 system you'd be better off with a proper retail board with BIOS support into the 2000s, either by the vendor or 3rd party patches.

Reply 8 of 9, by Matth79

User metadata
Rank Oldbie
Rank
Oldbie

Reading the SC1152CSW table, it looks like all off may also configure 2.0V
As for AZCM, it looks like ALL K6-2+ were officially mobile chips
Wonder if this can patch the BIOS, if you can get the voltage selected https://www.rom.by/articles/BP/index_english.htm

Reply 9 of 9, by alfonso

User metadata
Rank Newbie
Rank
Newbie

dionb, your enthusiasm inspired me to go for it. Given your comments, I am inclined to agree that the CPU is likely the bottleneck.

Upon inspection, the chip is the "SC1152CS" where you said. After double-checking your data sheet work, I gave it a try. Unfortunately, no post. I tried changing the multiplier between x3.5 and x4.5 also to no avail. I noticed that there is a special jumper "JPX1" which is for selecting between "AMD K6S-300" and "Other CPUs" for "CPU Type". In all combinations of multiplier and "CPU Type" I got no post.

I reinstalled the existing chip, restored all existing switch/jumper settings and got it to post. I then gave it a +50mhz overclock. The PC boots and Half-Life Uplink seems to play slightly smoother. I then tried to go up to 450mhz with/without raising the voltage to 2.3v. No post for either. I have reverted to the +50mhz overclock with the original 2.2v power. After 5-10 minutes Half-Life Uplink reports an illegal instruction, which "may" be because this overclock isn't stable. Hard to say.

I raised HL:U to 1280x960 and can say that the frame rate does not deteriorate in any way I could notice. It definitely seems to be CPU-bound as an issue. If K6-2 pluses are out of scope for now, are there any economical CPU upgrades you can recommend?

Matth79, Not sure if I am brave enough to try applying BIOS modding tools at the moment but appreciate the link.