RAMBUS memory size benchmarks?

[FOOL]

I don't know whether or not any sites have done this sort of benchmarking, but I think it would be interesting. Many people have said that due to the serial nature of RDRAM, the more memory you have, the slower it will run. Has anyone actually tested this to see the degree to which this is true?
Maybe going in 64 MB increments such as 64 MB, 128 MB, 192 MB, 256 MB, 320 MB, 384 MB, 448 MB, 512 MB, etc...
Personally I like to have at least 256 MB of RAM in my system, so I would like to know how RDRAM performs in this situation.
Ideas, thoughts, places this has already been done?
Thanks.

 

[Ulysses]

RDRAM is too expensive and currently offers no performance improvement over cheaper SDRAM systems. This could change in the future. Generally, sophisticated people avoid it:
http://www.pcworld.com/current_issue/article/0,1212,17284+1+0,00.html.

See also:
http://www.tomshardware.com/mainboard/00q2/000529/index.html

http://www.anandtech.com/showdoc.html?i=1239

http://www.anandtech.com/printarticle.html?i=1245

http://www.inqst.com/ddrv840.htm

 

[FOOL]

Thanks for the post Ulysses, but those aren't quite what I'm looking for. I've already seen a ton of benchmarks and such comparing DDR to RDRAM, but what I'm interested in is the performance of different memory sizes when using RDRAM. Every article mentions how the serial nature of RDRAM causes memory performance to slow with larger memory amounts, and I was wondering if anyone had done any sort of benchmarks to not only validate this, but to quantify it as well.

Thanks,

FOOL

 

[Adul]

Maybe Anand would want to step up to the plate?

 

[xtreme2k]

maybe no one wana piss intel off

 

[KarsinTheHutt]

RDRAM just isn't worth it right now. Intel basically admitted as much over the weekend when it posted some benchmarks comparing the RDRAM 820 platform to the PC133 815e platform. the PC133 won in 9 of 11 Intel benchmarks. I don't know whether this is due to RDRAM being slow as hell or the 820 chipset being a piece of junk. Regardless, I'd go with pc133 anyday.

www.theregister.co.uk has the dirty details i think.

 

[Prodigy^]

just say f*ck Rambus and forget it

and it's "Rambus" dammit, not RAMBUS! it's a company name, but an abbreviation! and there's no RAM type called Rambus, Rambus is the company who makes RDRAM. geh!

 

[lomiller]

RDRAM can get slower as more memory chips are added and as trace lengths become longer. The problem is basically the propagation time of the signal. Because of its high frequency and its long path lengths the time it takes for data to get from memory chips to the Chipset becomes important.

Since all data has to travel through all the memory chips in a channel and each chip adds a shot delay latency is affected by the number of memory chips in a given channel. The overall amount of memory is not a problem; just the number chips in the channel. The chipset compensates for this propagation time by adding clock cycles to the latency. The effect would be similar to going from CAS 2 to CAS 3. This is automatically detected and set on boot-up.
There can also be benefits to having more chips in a given channel, so performance results will vary depending on, Software used, motherboard, number of RDRAM chips, make of RDRAM chips.

I don't know of any benchmarks that show the results of adding more RDRAM chips to a channel, but the overall effect would be to slow the system down as chips are added.

 

[FOOL]

Okay, thanks for all the posts. I just went over the results that intel released on the two types of memory, and had a good laugh after reading toms most recent rant(?) on the subject.
Also, thanks for setting me straight on the number of chips being important rather than the amount of total memory.

- FOOL