anyone remeber an article about ram adressing over 1 gig

pinki

Member
Mar 23, 2003
25
0
0
Some time ago ( a year maybe?) i read this really great article on unknown website with benchmarks concerning what happens when you use more than 1 gig of ram with intel and amd processors. the contents of the article are vague but i remeber that the p4 was adversly affected in about half the benchmarks (i know sandra mem benches were among them) and the AMD machine was effected the other half of the time.

so im looking for this article because i cant remeber if the benchmark was done on an amd 64/opteron or if it was done on a barton core or older athlon xp.

the findings were quite clear that after you add more than 1 gig of ram to a system it would cut the bus width in half on a p4 to 32 bits and though many things werent effected some benchmarks were drastically effected. Though i dont remember about the amd machine in spesific i do remember that it was effected similiarly but in different benchmarks and that the method of adressing more than a gig of memory was different.

what i want is a link to the article, not a discussion on y 32 bit processors can physically adress 4 gigs of ram cause i know that they can just not at full speed depending on the processor.

i found out this info while searching for many hours for this lost article that discusses similiar issues with the pentium 2 processors and how when using more than 512 megs of ram the L1 cache and L2 cache would have to be disabled to adress that much and there fore performance dropped so much that it was never worth it.

thank you for your time and expertice
Pinki
 

mechBgon

Super Moderator<br>Elite Member
Oct 31, 1999
30,699
1
0
Originally posted by: pinki
Some time ago ( a year maybe?) i read this really great article on unknown website with benchmarks concerning what happens when you use more than 1 gig of ram with intel and amd processors. the contents of the article are vague but i remeber that the p4 was adversly affected in about half the benchmarks (i know sandra mem benches were among them) and the AMD machine was effected the other half of the time.

so im looking for this article because i cant remeber if the benchmark was done on an amd 64/opteron or if it was done on a barton core or older athlon xp.

the findings were quite clear that after you add more than 1 gig of ram to a system it would cut the bus width in half on a p4 to 32 bits and though many things werent effected some benchmarks were drastically effected. Though i dont remember about the amd machine in spesific i do remember that it was effected similiarly but in different benchmarks and that the method of adressing more than a gig of memory was different.

what i want is a link to the article, not a discussion on y 32 bit processors can physically adress 4 gigs of ram cause i know that they can just not at full speed depending on the processor.

i found out this info while searching for many hours for this lost article that discusses similiar issues with the pentium 2 processors and how when using more than 512 megs of ram the L1 cache and L2 cache would have to be disabled to adress that much and there fore performance dropped so much that it was never worth it.

thank you for your time and expertice
Pinki
The closest thing I can think of is the issue that the i430TX chipset had when equipped with more than 64MB of RAM. I experienced that firsthand. Some motherboards might gag at really high-capacity DIMMs or lots and lots of DIMMs, but I keep up on the CPU/motherboard/memory news fairly well (as you might guess from my five-digit post count :D) and I don't recall anything that really fits your description. I've personally run 1.5GB of RAM in an nForce2 system and there are plenty of other folks around here who've run 1.5GB+ as well. I'm pretty sure you'd hear them hollering bloody murder if it resulted in halved memory-bus width. :D

Maybe you were thinking of the tendency for some memory controllers to slow down if they are loaded with too many modules, which does happen in some circumstances (Athlon64 usually will only run a max of two double-sided modules at DDR400 speeds, for example).
 

zephyrprime

Diamond Member
Feb 18, 2001
7,512
2
81
I remember no such information. Some older processors are indeed not capable of caching the entire memory space but even so, you don't have to disable the cache. The memory space above the limit is uncached and the memory below is cached.

I bet what's most likely is that you are remembering an article that talks about what happens when you go from 2 to 1 dimm in a dual channel memory system.
 

obeseotron

Golden Member
Oct 9, 1999
1,910
0
0
If you were using lots of low quality dimms at the same time i could see them having a bad effect on bandwith, but even then it's unlikely. Older operating systems didn't efficiently use lots of ram, but anything nt based -2000, xp, should be fine. I would always say try to have as few dimms as possible and have the same exact ram for dual channel, but don't worry about a 1GB barrier, some motherboards might limit you to 2gigs, but most new ones should take 4 1gb dimms fine.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,572
10,208
126
Originally posted by: Accord99
I think you are refering to this Firingsquad article, where it was conjectured that the 4 DIMMs caused performance degradation for the P4 platform:
http://firingsquad.com/hardware/building_gaming_opteron_2003_Part2/page15.asp

However it's findings were not confirmed by Lost Cirucuits:
http://www.lostcircuits.com/memory/2gb/

I think I remember some article (on TH, maybe? I don't really recall), discussing how PAT works on i865 systems (which aren't technically supposed to have it), and the way that the mobo companies worked around it to enable it, resulted in a limitation on one particular memory configuration, I think running dual-channel with all slots filled, at 200Mhz, or something, that caused PAT to be disabled in that configuration on (some? all?) i865-based boards.

That's all I can remember, in terms of modern systems.

Way back in the day of Socket5/7, and Slot-1, there indeed were some limitations on both L1 and L2 cachability of RAM, but those issues are mostly just of historical curiousity at this point. (Higher physical addresses would end up uncached, which, combined with the fact that Win9x allocated memory from the top down, caused the OS and apps to primarily run out of uncached memory.) Also, Slot-1 PII chips with half-speed cache, had a seperate cache-controller on the PCB, and really early ones couldn't handle more than 512MB of RAM either.
 

Erehwon

Member
Jun 12, 2004
55
0
0
Originally posted by: pinki

so im looking for this article because i cant remeber if the benchmark was done on an amd 64/opteron or if it was done on a barton core or older athlon xp.
Pinki

Well if it was over a year ago I don't think the A64's were out were they?