RAM question for Giga-byte GA-5AX rev 2. (Update 7-9-06)

Discussion relating to Socket 7 hardware.
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

hello jim,

nice to see you back here!

you are right in both ways and you are in no way "acrimonious". you are just putting the old question on the table of what makes a benchmark tool a "good" one. but my conclusion regarding osmark is that it is the best (freely available) real-world benchmark.

a) it consists of many, very different, "applications", ranging from pure, stripped down bench-tools like integer/float calculation power, cpu's logic power (jumping, branching), plain memory throughput, 2D graphics output capabilities like lines, rectangles, filling etc. to highly complex cpu-graphics-mixed applications like the open-gl applications, working on pictures (color conversion, resizing, zooming) to "everyday" applications like zipping, wav-to-mp3 conversion, loading/displaying webpages and going back and forth in the history and so on. we are in the lucky positon to look at those sub-benches that interests us most. like for example if we are tweaking the memory settings via bios or wpcredit we can reassure ourselves if there is a "noticeable" effect in once identified appropriate real-world application (or a stripped down version of it) or can find out what applications are finally affected.

b) you get a result for each of those sub-benches! that is what every other real-world bench-tool I know of lacks of. they are only giving you aggregated numbers (like sysmark, winstone, winbench). you never know how they produced it. (I must admit that osmark should also give insight in the formula for the final "osmark" score ....)

c) osmark starts each of its sub-benches as an application with "normal" priority (seen from the windows core system). the longer the single tests run, the more often the system decides to give the "focus" to other applications. depending on what they are currently doing, they draw cpu-power from the currently running bench, induce the system to swap memory parts, start disk reading/writing, network activity and so on. this is reflected by varying results when you run the test several times. this is perfectly "legal" in a system with concurring tasks. and osmark measures performance by running "applications" in such a context. "everest" on the other hand does "meaningless" synthetic benching. for that purpose it "stops" somehow and to some degree other tasks while it is performing its tests. I can literally watch how the display of time (in the system tray; I have replaced the system time-task with one that displays seconds too) "freezes" and then suddenly jumps to the correct time. so when you record very different numbers by multiple runs then I would say that you are running one or more very "active" tasks in the background. I for myself have chosen to stop or deactivate all those noisy tasks when I run whatever benchtool (like e-mail-checkers, anti-virus programs, firewall etc.). that does not eliminate different results between multiple runs (there are still a dozen or more of sytem tasks you can not stop), but it frees the results of big outliners. up to 2 percent differences in the results are ok to be said of being "equal" because of the limits of the measurement tool and the environment.

d) the developpers of osmark then made a decision that has to be made when it comes to finally "determine" how fast the bench run. they could have choosen to take the lowest number, a mean, or whatever. they decided to take the highest number and I think it is ok, because they can perfectly say, look, in the best of all circumstances (system context wise) you can expect that performance. but I admit, that is a deliberate decision and everybody is free to take whatever one likes (because they tell you the numbers of all runs).

all this is not meant to give a reason to not use everest, sandra an alike. osmark plays in another league and complements the others.

and yes, you are right that osmark stresses the system that far that weaknesses (of electric circuitry of the mainboad, ram-stick quality, memory/bios settings, too high of fsb/cpu mhz setting, etc) will be brought to surface. I have experienced things like the program (sub bench) just does not start, aborts immediately, freezes itself, freezes osmark or even the system. one even gets a blue-screen or the system just restarts. well, something was pushed over its limits ..... but I never had any trouble with a corrupt system because of running osmark. and btw, if the system breaks it does not because of osmark, but because the system was not able to run it. to make a "picturesque" comparison: the steep hill is not responsible for a car's broken engine when the car is overloaded and the water in the cooling device was already at boiling point midway. osmark does not write to the system directory, does not alter the registry, does no dangerous tricks.

for safeguardng against such odds like unrecoverable system crashes (because of whatever made it happen) I would recommend a tool like "acronis partition image". it saves your system drive/partition (or whatever drive) into a single file. of course it is wise to choose another partition or drive for that backup purpose. in case of emergency you start your system from a bootable cd/dvd (which can be created from within the image-creation program). from there you then restore the image. depending on how big you had chosen your system partition (split system, programs and data to different partitions when you install first time) it takes just some minutes (on a 2gig sys-partition like mine) and you are back on track.

@stedman
I just want to repeat what jim is asking you: what about your 112fsb - osmark scores? I guess you ran into some trouble ;)
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Well Peter, Superpuppy 3 was able to run OS Mark just fine except when I tried to run at 5.5 x 112 w/CL2 RAM timings using RAM that is PC 133 CL3 & PC 100 CL2. That resulted in everything being screwed up including my backups, in another partition on another drive.

Superpuppy 2 was unable to run OS Mark at all using the mixed bag of junk ram I orriginally had installed. I now have PC 133 CL2 Hynix for it; but as you know it has other problems just now.

The problem really is that both machines had very complex OS - Software installations, that take weeks to get up and running right. Would have been nice to have a warning about OS Mark's tendancy to trash OS installs before running it; but I guess you didn't know how bad it can get.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
DasMan2

Post by DasMan2 »

Just to inform people on Socket 7 motherboards that supposively (offically) in their manual state support for being able to use SDRAM sticks of 512 mb size.

a P5SS-ML (Elite Group) motherboard (M-ATX) Chipset is a SiS 530 North & SiS 5595 Southbridge . Manual (page 4) clearly states able to use SDRAM non-buffered DIMM memory modules up to 512 in size ...three slots for a total of 1.5 GB. Page 24 gives more detailed info; 1st stick must go into DIMM 1(so onboard can share memory) and then the other can go into either slot 2 or 3 (hmm!) Again stating total capacity as 1.5 GB.

There is also support for EC or ECC sticks & Ultra DMA 33/66.

Maximum bus speed is 100mhz.

I have this motherboard with Win 98 S/E & AMD K6-2 475+ @ 5.5 x 100 . Runs well with a 7200 RPM 2 mb cache hard drive and I use a PCI video card that allows for better performance than using the onboard video.
Even with only 96 megs ram (3x 32mb sticks) it is still snappy.

Now sometime in the future I will try to get 256mb sticks in there to check out any performance boosts (noticed) Also a better PCI videocard would be good too. :P
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Since when is an oldtimer like you a guest? Should try to find Hynix "Low Power" PC133 CL2 for best performance. Though I did once see a guy selling PC150 RAM; but it was too tall to fit in my machines.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
DasMan2

Post by DasMan2 »

A little searching on the internet reveals a place called : www.memoryx.net

Seems to have a SDRAM 168 pin PC100 in sizes of 256 / 512 / 1 GB !

Also ECC or ECC Registered MemoryTen Lifetime warranty :shock:

PC100 / PC133 ,even low profile / PC150 , alot of different manufacturers listed : MICRON , TwinMOS , Adata , Kinston . MemoryTen , Lifetime warranty
Example: KVR100X64C2/512 as a P/N or ECS-XAD512 .
Cost listed is still high around $90 (US) plus shipping

There must be still alot of Socket 7 / super 7 Motherboards not even listed normally that have the ability to use large SDRAM of 512 for sure or 1 GB (maybe) :P
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

You are probably right about the mobos; but I would still prefer Hynix Low Power PC133 CL2. My 2 sticks of Hynix PC133 CL3 Low Power support a couple of tweaks that none of my other sticks support. - And they arn't bad tweaks either.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
User avatar
KGB
K6'er
Posts: 98
Joined: Mon Apr 07, 2003 7:55 pm

Post by KGB »

Quick recap fellas, been chugging along nicely on this vintage system. I "found" a Micron RAM module, a 256MB version, dual sided, PC133 3-3-3 markings on it.

Installed it into the 6'er and the system booted, counted all ram, started to desktop, win98 reported all memory present. RAM Banks 1a1b2a were technically full, equaling 768MB. Didn't run any tests, system felt the same. I have since removed the 512 module and benched everest under Win98. Results are a mixed bag (all averaged from 6 runs)
K6-2 AFX
110x5 @2.2V - Stock (No L2cache)
MR 362
MW 114
ML 234.2

110x5 @2.2V - WpCredit Tweaked (No L2 cache)
MR 389
MW 123
ML 210.9

105x5.5 @2.3 Stock (No L2 cache)
R 344
W 103
L 244.8

105x5.5 @2.2V Wpcrdit Tweaked (No L2 cache)
R 370
W 117
L 218.3

Tests were does under Windows 98se, K6Speed was not used, coincidentally I no longer have to manually turn on Write Allocation through K6Speed, a task I did earlier with the 512 MB module.

It has been a while since I benched this machine, and the numbers have never been consistent with anything. Looking through my previous posts, I have been rather lax about which operating system I used for benching. I do believe the higher benchmarks I received were under Windows 2000. Mainly because it is rather difficult to get these numbers under Win98.
-K6-2 550 Gigagyte GA-5AX (5.2) 1x512MB GeForce 2 Ti500 64MB
-Pentium III 933 Asus P3C-L 2x128MB GeForce 4 Ti4200 128MB DDR
-Pentium III 750 Asus P3B-F ATi Rage128 GL 32MB
-Celeron 1.2 Asus TUSI-M 2x128MB
-Pentium IV 1.5 DELL 2350 1x128MB
-Athlon 900 Asus K7M 1x256MB 2x128MB GeForce 3 64MB
-AthlonXP-M 2400+ Asus K7V880 2x512MB HIS Radeon HD3850 512MB DDR
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

I have noticed the same thing. I get better numbers under XP than I do under 98. Find it odd that you are getting better numbers w/ 5x110 than you are w/ 5.5x105. Probably because the cache is off..
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
User avatar
KGB
K6'er
Posts: 98
Joined: Mon Apr 07, 2003 7:55 pm

Post by KGB »

I spent the morning optimizing the system with 768MB of RAM, @ stock the system ram very sluggish. (Again these are averaged of runs)

110x5 (Stock) No L2 cache
R 371
W 84
L 224.8

110x5 (Tweaked with Wpcredit) No L2 cache
R 387
W 123
L 210.4

The "new" 256MB Module sticker is as follows: MT16LSDT3264AG-1333E3 PC133-333-542A

The chip markings are: MT48LC16M8A2TG-75E

Remarkable that; I get the same results with two different models. Jim, I'm contemplating whether to install Windows XP on this machine now, since I have enough ram, and Windows 98se is good but, I seemed to have gotten jaded with WinXP and have little in me to continue with Win98 (except for legacy games, which I play on the thinkpad anyway) The only downside is that 2K & XP require lots of L2 cache consumption. A feature I cannot boast. :(
-K6-2 550 Gigagyte GA-5AX (5.2) 1x512MB GeForce 2 Ti500 64MB
-Pentium III 933 Asus P3C-L 2x128MB GeForce 4 Ti4200 128MB DDR
-Pentium III 750 Asus P3B-F ATi Rage128 GL 32MB
-Celeron 1.2 Asus TUSI-M 2x128MB
-Pentium IV 1.5 DELL 2350 1x128MB
-Athlon 900 Asus K7M 1x256MB 2x128MB GeForce 3 64MB
-AthlonXP-M 2400+ Asus K7V880 2x512MB HIS Radeon HD3850 512MB DDR
Jim
K6'er Elite
Posts: 1745
Joined: Wed Jan 21, 2004 7:10 pm
Location: Toronto

Post by Jim »

Your Memory Read numbers are excellent as Super 7s go; but the Memory Write numbers are attrocious. The memory that you are using may support tight timings at 105; but I doubt if it will at 110 (Actually 112 on ASUS boards, though I don't know about Gigabyte). Correct me if I am wrong, you are running a K6-2, non plus, that right? That may explain the Memory Write numbers.
Superpuppy 3
K6-3+ 450 ACZ (6x100)
DFI K6BV3+/66 Rev B2 (2 Meg) w/ 2x28mm Chipset Fans
2x256 Meg PC 133 Hynix SDRAM
1x 20G Maxtor (7200)
2x 80G Maxtor (7200) Ducted w/ 2x486 Fans Mount
52/24/52/16 LG CDR/RW/DVD
8/4/3/12/24/16/32 LG Super Multi
ATI 9000 aiw Radeon AGP
SB Audigy 1 MP3 Sound
CMD 649 IDE Controller
NEC USB 2 Card
RicardoMTPX
Newbie K6'er
Posts: 4
Joined: Sun Jun 25, 2006 11:06 pm

Post by RicardoMTPX »

The setting "Host DRAM Read Command Mode" on GA-5AX controls how the onboard cache and memory controller work. When this setting is set to "Sync", the cache hit detection is carried out first. If the cache miss is detected, then memory cotroller is accessed to retrieve the data.
When this setting is set to "Bypass", the cache hit detection and memory access are done at the same time. If a cache hit is detected, memory access is terminated.

Setting "Host DRAM Read Command Mode" to "Bypass" helps reducing the latency, but I also discovered that the stability with this setting set to "bypass" depends on memory stick on GA-5AX. Some can have that enabled, some can't. Asus P5A has this setting always set to "bypass" and hidden from user selection. Gigabyte said there isn't any GA-5AX carrying revision G of the Aladdin V, but looks like my one has a revision G chipset on it....

All 3 chipset venders - VIA, SiS, ALi, their products all have this setting.... but the name might be different.

512MB single stick memory module - It appears that VIA's and ALI's products have preliminary support for 256Mbit memory devices .... the proof is, in the PCR file, there are entries saying "Row X have 256Mbit memory device", and X varies rom 0 - 7. Memory makers often use 16 of 256Mbit IC devices to build a single stick of 512MB RAM. But I would expect such memory to have a better chance to work with ALi chipset as ALi's memory controller is better.
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

@ricardo

does it just "look like" your mainboard features the "G"chipset revision or what does the print on the ALI 1541-chip say (last letter on second last line)?
RicardoMTPX
Newbie K6'er
Posts: 4
Joined: Sun Jun 25, 2006 11:06 pm

Post by RicardoMTPX »

DonPedro wrote:@ricardo

does it just "look like" your mainboard features the "G"chipset revision or what does the print on the ALI 1541-chip say (last letter on second last line)?
The chip marking has "G" at the end of second last line. I am not sure if it is revision G chipset. However this mainboard still has a tag SRAM on it.

On GA-5AX, I discovered there were a few BIOS options that was hidden. You can download GA-5AX BIOS, and the modbin for Award BIOS v4, then you can see the hidden options and even enable them. The hidden ones I can recall is "L2 Tag Width: 8bit/10bit", and "ACPI Function: Enable/Disable".
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

@ricardo

when the chip is marked "G" you definitely got the g-revision! :)
that is very interesting! I've never heard of any ga-5ax using that chip revision ....

what is the revision of your mainboard?

regarding the internal/external tagram: please use the tool wpcredit (you can find it in the download section here at k6plus) to verfiy how your bios set up the cache handling.

I am running a ga-5ax rev 5.2 mainboard. it features the "E" revision chip. accordingly internal tagram is disabled and external is enabled. I have included two screenshots of the respective register settings.
Attachments
internal tagram disabled
internal tagram disabled
ga-5ax-ali5-int-tag-disabled.png (15.21 KiB) Viewed 14292 times
external tag-ram enabled
external tag-ram enabled
ga-5ax-ali5-ext-tag-enabled.png (15.14 KiB) Viewed 14292 times
DonPedro
K6'er Elite
Posts: 578
Joined: Wed Jul 27, 2005 2:11 pm

Post by DonPedro »

@KGB

something seems to be wrong with your setup. your everest numbers are in my opinion too low for a fsb of 110. especially your everest write scores should be much higher.

my tests for a k6-2-550 (fsb 100), ga-5ax rev 5.2, cache disabled, give the following results (everest is of version 2.20.405)

wpcredits not applied, 3x128 ram (2-2-4-5-8 )
read 356
write 104
latency 246

wpcredits applied, 3x128 ram (2-2-2-4-7) gives
read 367
write 145/136/128/120/138/146/139/136/126/133/143
latency 210

I don't know what made the write scores jump around here, but on average the score is 130+.


wpcredits applied, 1x256 ram (2-2-2-4-7):
read 368
write 138
latency 211


anyway, I think your very low write score of 84 at 110 fsb is due to write allocation not enabled. I had such an experience when I moved from 256 ram to 768 ram and my everest write scores went down to 60/70 something. I was able to set it straight with k6speed.
Post Reply