Memory Bandwidth Test Result

Message

whizkid · #1 Post by **whizkid** » Tue May 08, 2007 12:48 pm

Hey all, rather than bury it in some other thread, I thought I'd start a new one.

I have a T60 (2623-D7U), 2.0GHz Core Duo with two 512MB PC5300 memory sticks. Both were installed by Lenovo. One is branded Samsung and has Samsung chips, the other is branded Hynix and has Hynix chips.

With either stick in either socket on my T60, memtest86+ indicated 2106MB/s memory bandwidth.

With both sockets filled, in either orientation, the measurement was 2293MB/s, for an improvement of 8.879%. I was hoping for a lot more speed.

Is there another test I should try? A bootable CD image is preferred.

#2 Post by **RonS** » Tue May 08, 2007 12:55 pm

With the two modules installed, you should be running in symmetric dual-channel mode. I've heard you get a real-world memory speed boost of about 10%, so your results make sense.

Splaktar · #3 Post by **Splaktar** » Tue May 08, 2007 1:17 pm

I'm running dual channel (Interleaved) mode as well with 2 Micron 1 GB PC5300 SODIMMs.

I've tried out Rightmark Memory Analyzer with success. It's not a bootable CD image though.

It gave me the following numbers:

Real Read Bandwidth, MB/s: Average 3682.41, Max 4265.95
Real Write Bandwidth, MB/s: Average 1489.75, Max 3503.24

Read Latency, ns: Average 106.45, Minimal 105.54, Max 114.21

Real RAM Bandwidth, MB/s: 2517.18, Max 2764.04

I don't really have much to compare these against though, other than the fact that I have twice as much RAM but not nearly twice as much bandwidth. The RAM are running at 332.50 Mhz.

whizkid · #4 Post by **whizkid** » Tue May 08, 2007 2:03 pm

Thanks, but that's not really useful by itself. Can you repeat the test after removing one stick so we can compare the difference, please?

Splaktar · #5 Post by **Splaktar** » Wed May 09, 2007 6:19 pm

OK, 1 GB of RAM removed:

Channel Mode: Single

Real Read Bandwidth, MB/s: Average 3775.05, Max 4312.76
Real Write Bandwidth, MB/s: Average 1298.24, Max 3509.05

Read Latency, ns: Average 105.21, Minimal 105.04, Max 111.96

Real RAM Bandwidth, MB/s: Average 2213.24, Max 2391.03

So that gives me a Bandwidth gain on average of 303.94 MB/s using Dual Channel SODIMMs. This is a gain of ~12-13% if I'm not mistaken.

whizkid · #6 Post by **whizkid** » Thu May 10, 2007 10:51 am

I get these numbers:

Average: 13.7% faster
Read: 97.5% as fast (2.46% slower)
Write: 14.8% faster

How interesting that average read speed went down a little. I guess it's the bus bandwidth that's the bottleneck, and the logic to deal with dual-channel causes a small delay.

But that's a nice write speed improvement. For some very few folks it could make a nice difference.

Anyone up for doing some gaming benchmarks comparing single-channel to dual-channel?

Dodge DeBoulet · #7 Post by **Dodge DeBoulet** » Thu May 10, 2007 12:34 pm

Pleading ignorance of the memory technology used here, but if I, at some point, swap one of my 1GB SODIMMs for a 2GB, will I still see the same performance? Or do dissimilar-in-size memory modules preclude the use of dual-channel?

I'm hoping that the first two gigs will be operating at the dual-channel speed, while the remaining gig operates at the somewhat lower speed.

whizkid · #8 Post by **whizkid** » Thu May 10, 2007 12:38 pm

To get dual-channel to work (in the T60 anyway), you must use identical memory sticks.

Splaktar · #9 Post by **Splaktar** » Thu May 10, 2007 1:01 pm

whizkid wrote: Anyone up for doing some gaming benchmarks comparing single-channel to dual-channel?

Just ran 3DMark06 with 1 GB of RAM and got the same exact score as my previous tests with 2 GB of RAM (and same drivers).

1612

Splaktar · #10 Post by **Splaktar** » Tue May 22, 2007 10:04 am

Got my replacement RAM from IBM. My old being bad and causing lock ups.

I'm running dual channel (Interleaved) mode.

Real Read Bandwidth, MB/s: Average 4032.36, Max 4618.78
Real Write Bandwidth, MB/s: Average 1514.67, Max 3507.72

Read Latency, ns: Average 94.08, Minimal 92.65, Max 100.83

Real RAM Bandwidth, MB/s: 2559.48, Max 2810.87

Looks like a solid improvement over the old stuff (~1.6%).

ahe · #11 Post by **ahe** » Fri May 25, 2007 6:33 am

Are you guys certain that you are actually running dual-channel?

I've tried to find factual information about this, and it seems like you will get an actual doubling of the memory bandwidth if it's running in dual-channel. The specifications on Intel's pages for my processor (the T7600) does not mention dual-channel at all. The motherboard apparently has support for it, though.

The wikipedia page for the Intel Core 2 processor clearly shows that it is not supported for the processors with 667MT/s, but I don't know the source. See the "Matched processor and RAM ratings" section here: http://en.wikipedia.org/wiki/Intel_Core_2

It seems like the only dual-channel mode is supported for T7x00 processors in a "Socket P" (ie. front side bus with speed 800MT/s). See http://en.wikipedia.org/wiki/List_of_In ... re_2_Duo_2

Splaktar · #12 Post by **Splaktar** » Fri May 25, 2007 7:40 am

Well I'm using a Core Duo processor, not a Duo 2.

RightMark Memory Analyzer is telling me that I'm in 'Dual Channel Interleaved' mode. It's a free download if you'd like to check it out yourself.

ahe · #13 Post by **ahe** » Fri May 25, 2007 8:16 am

I have only one memory chip, a 1GB PC-5300 that came with the laptop, so it reports Single-Channel for me. Which test in particular did you run? I get values that are way higher than yours, e.g. average read bandwidth of 13432MB/s.

Edit: memtest86+ surprisingly reports that the memory speed is 15212MB/s.

From the documentation I've read it seems that dual-channel allows you to use half-speed memory to achieve the full bus transfer rate. I might be wrong.

Splaktar · #14 Post by **Splaktar** » Fri May 25, 2007 8:40 am

The 'Quick Tests' will give you details on memory bandwidth and such. This is what I'm using.

ahe · #15 Post by **ahe** » Fri May 25, 2007 9:15 am

Okay, there it gave me these results:

Code: Select all

Real Read Bandwidth, MB/s: Average 1787.02, Max 2030.88
Real Write Bandwidth, MB/s: Average 601.77, Max 1511.97

Read Latency, ns: Average 117.57, Minimal 107.52, Max 128.54

Real RAM Bandwidth, MB/s: 994.35, Max 1121.32

But when I changed some settings in the General tab (Min. Data Set Size to 1GB, thread lock to enabled, and memory allocation to standard), it reported these results:

Code: Select all

Real Read Bandwidth, MB/s: Average 4418.96, Max 4737.89
Real Write Bandwidth, MB/s: Average 1476.60, Max 3535.37

Read Latency, ns: Average 53.26, Minimal 51.32, Max 54.90

Real RAM Bandwidth, MB/s: 2428.88, Max 2709.33

Memory Bandwidth Test Result

Memory Bandwidth Test Result

Who is online