more ram slows down the cpu?

[projectb]

hi...
i upgraded my pc (amd k6-2 400 / asus p5a mobo / voodoo banshee graca)from 128mb to 256mb ram... (128mb pc100 +128mb pc133)
after installing the new pc133 chicp the system started up lika always...

but when i benchmarked it with sandra99 pro
i got 100mips less than before!! my performance uder 3dmark99 pro got down 400-600 points, also...

is it possible that the ram has errors and the cpu is always busy to correct them?
(it is no-name ram...) or am i using a ba timing for the chips... at the moment i use 3-2-2 (the bios automatically uses this settings by speed detection...)

thanx 4 ur help!
<PB>

[Peter M]

No, it's just that you went beyond your mainboard's cacheability limit. The level of cache on your mainboard covers only 128 MBytes of RAM.

Your benchmark will be a tad slower when it happens to use the upper half of your 256 MBytes - but your applications will be a lot faster, provided they actually NEED more than 128 MBytes. L2-uncached RAM is orders of magnitude faster than HDD swapping ...

Regards, Peter

[Fingers]

Just like Peter said:

<A HREF="http://support.crucial.com/scripts/crucial.exe/solution?11=001128-0003&130=000975429440&14=&2715=&15=&2716=&57=faq&58=&2900=JqswLusgDW&25=-1" TARGET=_blank>Why did my computer slow down after I installed more RAM?
</A>

[Nighthawk]

I had a motherboard with the same chipset and noticed a slight slowdown when adding 128mb of RAM, but I don't think it's a caching issue.

I just checked on ALi's website for the northbridge re: caching, and came up with the following:
"Cacheable memory up to 512MB with 10-bit Tag SRAM when using 512KB L2 cache, 1GB when using 256KB L2 cache"

That doesn't make much sense that it can cache more with less L2, but that's straight from the text of ALi.
http://www.ali.com.tw/eng/product/core/m1541.htm

Anyway, I'd run a memory checking utility. Usually a memory error would cause a complete lockup and bluescreen, but it could be performing under spec.

Another thought: Your previous config of 3-2-2 could easily have changed to 3-3-3. I've seen some cas3 stuff at 3-3-3, even PC133... Run PC133 at 100 and it'll change to 2-3-3 (I think). Since it would detect the slowest on each, you'd be running at 3-3-3 instead of 3-2-2.

Edit-&gt; provided link to the ALi website and changed my memory advice (didn't notice some stuff you had provided)

[This message has been edited by Nighthawk (edited 02-19-2001).]

[Undertow]

I found that running my PC133 RAM at 100 allowed me to use 2-2-2, where as at 133 I couldn't boot at anything quicker than 3-3-3. What do you typically use your computer for?

[projectb]

hmm...
@peter: what about the article nighthawk brought up... it supports up to 512mb cacheable...

@nighthawk after installing the bios (rev 1007a) told me that it's using a 3-2-2 clocking by speed detection...

i bootet up with a 2-2-2 timing, too and there where no problems... should i use this for a better performance? (i'm getting nearly the same benchmark results as before... only the alu is 80mips slower...) using a faster timing doesn't heat up the chips more i think, am i right?

btw: ut takes bloody ages to load http://sysopt.earthweb.com/forum/wink.gif

cu!
&lt;PB&gt;

[Peter M]

The cache size vs. cacheable size thing comes from design choices in the chip. Intel's socket-7 chips did not change cacheable area with cache size; VIA's and SiS's offerings can cover more DRAM when you give them larger cache chips, and with ALi's it's the other way round.

But anyway, in most revisions of the 1541, the internal TAG RAM cannot be used, and an external 8-bit one must be on the mainboard. This limits cacheable area to 128 MBytes.

If you want to learn about your memory being OK or not, use memtest86. This is the most extensive and thorough memory test out there.
With that, you'll also see whether you can run CL2 timing (222) safely. Normally, PC133 memory should be able to manage that at 100 MHz.

Regards, Peter

[projectb]

hmm...

i've ran thememtst yesterday... with a 2-2-2 clocking... it reported no errors and all went fine...

@peter memtest reported that it found 256mb cacheable ram... (with 80.4mb/s data transfer...)

but over night lightwave crashed... hmm... maybe steamer went down http://sysopt.earthweb.com/forum/wink.gif i'll try tonight with a 3-2-2 clocking... and better steamer options http://sysopt.earthweb.com/forum/wink.gif

so far!
&lt;PB&gt;

[Peter M]

For a reliable cacheability check, use the CACHEMEM utility. It also gives more precise throughput figures, so you can work out what you actually gain from risking 222 timing instead of 322.

Regards, Peter

[projectb]

well thanx but here's another problem http://sysopt.earthweb.com/forum/wink.gif
i can't get a runnig link to the cachemem prog... anyone an idea where i can download it?

&lt;PB&gt;

[Buji]

http://www.fullon3d.com/temp/cachemem/

[projectb]

thanx... i'll try it today http://sysopt.earthweb.com/forum/wink.gif

&lt;PB&gt;

[LeaDxPainT]

Mixing pc100 and pc133 will cause your computer to slow down...try buying the same pc's for ram.
LeaDxPainT

[Fingers]

<A HREF="http://support.crucial.com/scripts/crucial.exe/solution?11=001204-0006&130=000975967499&14=&2715=&15=&2716=&57=faq&58=&2900=SKOuHOJdZ3&25=-1" TARGET=_blank> Should I buy a PC100 module or a PC133 module?

Solution:

If your PC has a 133MHz front side bus (FSB), you will need PC133 SDRAM.

If your PC has a 100MHz FSB, you can use PC100 or PC133 SDRAM. All PCs that accept PC100 SDRAM will also accept PC133 SDRAM; however, your memory will only run as fast as the slowest "link" in your system. If you have a 100MHz FSB or any PC100 modules installed, any PC133 modules that you install will only operate at 100MHz.

PC133 SDRAM doesn't offer any immediate benefit over PC100 SDRAM if you have a 100MHz FSB. However, if you are planning to upgrade to a system with a 133MHz FSB in the future, you may be able to use the PC133 modules you purchase now in your future system. </A>

[projectb]

well now... i ran cachemem yesterday...
but can anyone please tell me the benefit (and information) i can get from this prog? http://sysopt.earthweb.com/forum/wink.gif

here are 2 links to my logfiles... http://www.geocities.com/projectb01/amd_k62-400at420_windows.txt

...
http://www.geocities.com/projectb01/amd_k62-400_dos.txt

i'd be thankfull if anyone could tell me if this ist too slow for a k6-2 400...

btw: why do i have l2 cache only when running dos mode?

thanx!

&lt;PB&gt;

i love geocities *grr* http://sysopt.earthweb.com/forum/wink.gif

[This message has been edited by projectb (edited 02-22-2001).]

[Peter M]

CACHEMEM, when started from Windows, did warn you that results are inaccurate, didn't it?

Your DOS results look perfectly fine for a super-7. Now your suspected cacheable RAM area limitation didn't show, so you need to use ctcm7 (from ftp.heise.de) to find out about that.

Regards, Peter

[Amorpheous]

Fingers, I tried using a stick of Nanya 128MB pc133 in my old Tyan 1590S 100mhz socket 7 motherboard and it fried dimm slot #1...so i think its safe to say not *all* motherboards that support 100mhz fsb will run pc133 sdram, at least not the Tyan 1590S http://sysopt.earthweb.com/forum/smile.gif




[This message has been edited by Amorpheous (edited 02-24-2001).]

[projectb]

well.. here's my ctcm run...


PROCESSOR AND CACHE INFO c't 01/00/ Andreas Stiller V1.7a

Processor Timing : am6k86
Processor CPUID : AuthenticAMD Typ=00 Fam=05 Mod=08 Rev=0C Feat=008021BF
Processor Name : AMD K6-2CXT /AMD-K6(tm) 3D processor, Feat:808029BF
AMD K6 config. : Write Allocation to 256 MByte
Write Combining LFB : disabled, BF-Pins=010 =&gt; Ratio= 4.0
Actual clock rate : 400.895 MHz, according to Pentium Timer:400.949 MHz
Primary Cache (L1) : 32 KByte,2way associative
Secondary Cache (L2): 512 KByte,direct mapped
Code Cache (L1) : 32 KByte,2way associative
Main memory : 256 MByte, no Memory holes found
Cacheable Area L1 : 256 MByte, no noncacheable Areas found
Cacheable Area L2 : 128 MByte, Cache Area &lt; Main memory!!

^-^-------- i think this is my problem right? http://sysopt.earthweb.com/forum/wink.gif

Write Strategy L1 : Write Back, Write Allocation, linear Fill
Write Strategy L2 : Write Back, Write Allocation, no L2 Flush (wbinvd)
Dirty Tag L2 : ok

Data Flow and Bus Performance (Main memory: 00120000h)

Transfer in 4 GByte Real Mode, no paging, via CPU Integer Unit
best rate for 32K MOVSD Cache/Page Hit : 21.8 æs =&gt;1505.5 MByte/s
medium rate for 32K MOVSD (Miss + Hit) : 168.5 æs =&gt; 194.4 MByte/s
medium rate for 32K MOVSD (L2 clean) : 201.9 æs =&gt; 162.3 MByte/s
medium rate for 32K MOVSD (L2 dirty) : 292.2 æs =&gt; 112.1 MByte/s
worst rate for 32K MOVSD (misses) : 385.5 æs =&gt; 85.0 MByte/s
via FPU 32K FMOVI (misses) : 389.2 æs =&gt; 84.2 MByte/s
via MMX 32K MMOVI (misses) : 382.4 æs =&gt; 85.7 MByte/s
Blocktransfer 4M MOVSD (misses) : 55.9 ms =&gt; 75.1 MByte/s

averaged at 512 KB L2-Cache /DOS (640K) : 168.3 æs =&gt; 194.7 MByte/s
averaged at 512 KB L2-Cache /Win (4M ) : 245.8 æs =&gt; 133.3 MByte/s


well thanx for all the helpers http://sysopt.earthweb.com/forum/wink.gif

btw: how are the other results? acceptable for a k6-2 400?

cu!
&lt;PB&gt;

[This message has been edited by projectb (edited 02-25-2001).]