DDR2 8500 RAM Bandwidth Seems Low

[champmanfan]

Hi. I've only had this PC for about 2 months now and whilst it fast for video editing & gaming, the RAM bandwidth seems low for my type of memory (4GB Corsair TwinX DDR2 8500 Dominator 5-5-5-15). I hope someone here is familiar with my type of system to help out please since I bought the best RAM I could afford - see signiture for brief details on this. My mobo is a BFG 680i SLi with latest BIOS.

I tested the memory bandwidth using SANDRA on XP 32bit three times and heres the screenshot of the highest score:


Does anyone here will have similar scores for 8500 rated RAM or is this low. The RAM is rated 1066mhz speed/2.2v. It would be better if I could compare to similar RAM here because the RAM in SANDRA for comparison isn't too good. How do you work out the bandwidth for my CPU + RAM to work out if the RAM is creating a bottleneck overall.

More detailed PC info is on my profile page. Thanks

 

[MetalX]

Well, I don't think you ever get all the advertised bandwidth, and since Core 2s don't have an integrated memory controller, they don't access the memory very efficiently.

However, I don't know all that much about memory, so you should wait until someone more experienced comes along.

 

[CMH]

Uhm... your memory seems to have automatically set itself to DDR2 533mhz speed, perhaps to match your CPU bus speed. This is quite strange since it should automatically be set to 800mhz.

The DDR2 8500 speeds are acheived only by overclocking, which you seem to have done. Did you set the CPU:RAM ratios right?

 

[MetalX]

No, it's automatically set to 533MHz (DDR2-1066MHz) which is PC2-8500.

Look in the middle section of the upper of those two silver windows. It says:

FSB:1068.00 (QDR) MHz
RAM: 1068.00 (DDR2) MHz

 

[champmanfan]

Uhm... your memory seems to have automatically set itself to DDR2 533mhz speed, perhaps to match your CPU bus speed. This is quite strange since it should automatically be set to 800mhz.

The DDR2 8500 speeds are acheived only by overclocking, which you seem to have done. Did you set the CPU:RAM ratios right?

My RAM is SLi certified memory which the mobo can utilise to set itself to the correct rated speeds (no need to set it manually) which is why I thought the benchmark would give better results. Its set to 'SLi Max' on the mobo, my CPU is set to 1066mhz FSB & My RAM is set to 1066mhz as well. My CPU & RAM isn't linked like 1:1 or anything but it looks like thats what my RAM has set itself to since they both match the same speeds. The speeds set are the rated speeds.

No, it's automatically set to 533MHz (DDR2-1066MHz) which is PC2-8500.

Look in the middle section of the upper of those two silver windows. It says:
FSB:1068.00 (QDR) MHz
RAM: 1068.00 (DDR2) MHz

I wasn't sure if thats how DD2 RAM worked but that makes sense how you've explained it, ta. So when DDR3 comes out would it say 355mhz which equals DDR3 if multiplied by 3?

I've only had 333Mhz RAM before on AMD Athlon 3400 so I'm not used to Intels ways but I'm getting there. I bought a QX6700 because its more powerful - at the time - for my work & games but I would have thought there was more bandwidth available. So does anyone know how to work out my bandwidth in Gbs.

Thanks for the replies

 

[MetalX]

No. DDR1, DDR2, and DDR3 are just different versions of DDR. DDR stands for Double Data Rate, meaning that the effective clock speed is double the actual clock speed because data is sent on the rising and falling edge of each clock cycle.

Bandwidth is worked out by multiplying the effective clock speed by 8, or by 16 for dual channel RAM. So DDR2-1066 is PC8528 (but the manufacturers round it down to PC8500 on the packaging), and DDR2-1066 Dual Chanel is PC17056(which would be rounded down to PC17000. These numbers stand for the number of MB transfered per second.

So with PC8528, that means that there is 8,528MB/s or 8.5GB/s memory bandwidth.

And just wanted to note, your FSB : DRAM ratio is not 1:1, it is 1:2 because the RAM is running at 533MHz (1066MHz DDR or Dual Data Rate) while the FSB is running at 266MHz (1066MHz QDR or Quad Data Rate).

If you wanted your RAM to be 1:1, it would have to run at the very slow 266MHz (533MHz DDR) which nets a memory bandwidth of 4.2GB/s (PC4200).

 

[champmanfan]

Thanks for clearing up the bandwidth question, very helpful answer thanks

I did think that DDR2 was like Dual Channel as in its referred to as double data rate because my RAM is 128bits from what I remember when I ordered it; like using two 64bit channels to use like Dual Channel. From the Scan page where I bought it it says 2x128Mx64 Module Configuration. I think the 64 afterwards is used to transfer data at a native 64bits for 64bit OSs.

So 8.5GBps of RAM bandwidth seems better than I thought, but why does SANDRA say 4.9Gbps when it works out higher than this? As you say, my FSB is 267mhz multiplied by 4 cores = 1068mhz. I'll post some BIOS screenshots later to you can see exactly how its set up on my BFG 680i SLi.

As for 1:1 ratio, it seems my FSB is the bottleneck and not my RAM as first thought - unlike the AMD CPUs - when you said "If you wanted your RAM to be 1:1, it would have to run at the very slow 266MHz (533MHz DDR) which nets a memory bandwidth of 4.2GB/s (PC4200)." So my FSB would have to be double to get true 1:1 but thats not possible since it would have to be a crazy 2132Mhz if I understand correctly - my mobo supports 800/1066/1333mhz FSB buses.

 

[CMH]

Well, thats what overclocking is... You shouldn't end up using your mem to full speed if you're going 1:1, unless you lower your CPU multiplier to something really low, which may adversely affect performance.

But do try reducing your CPU multiplier and set your RAM to 1:1, you should get least get to 6400Mb/s (If I'm right, at 3.2ghz, CPU multiplier 8x, FSB 400mhz).

After that, you can try other RAM ratios. Its quite divided on whether running your ratios at 1:1 will give a better performance, but it doesn't hurt to experiment and benchmark to see.

 

[champmanfan]

I've tried different FSB settings in BIOS but it was keep crashing just before the loading screen to XP Pro so I can't set it to 1600QDR FSB or 400mhz as you suggested. I've got it currently on FSB 1280mhz & RAM on 1066mhz.

I'm curious; would there be much difference between my 8500 RAM & whatever is the fastest RAM available for the nVidia 680I Sli?

Here's the updated benchmark which gives the RAM quite a bit more bandwidth:


It seems that 1280mhz FSB is the sweet spot & I don't know if it can go higher or if anyone has tried higher FSB on the Core2 Quad Extreme 6700. Its still way off from your suggested 6.4Gbps because I can't get it to boot up when I set the FSB to 1600 and x8 FSB.

Heres some more shots of my BIOS beforehand using the CPUMAX RAM SLi so you know how it setup on my mobo.

CPU Multiplier Settings


CPU RAM OCMax Settings


RAM Timings Settings On Optimal


Voltages (never touched them)


If you have anymore suggestions since looking at the shots I'd appreciate the help. I will post more shots when I get CPUZ installed later & another utility. Is there other utilies that anyone uses here to benchmark your system for RAM & CPU bandwidth?

 

[Asher]

I'm having the exact same problems! It's very frustrating.

I've an ASUS Striker Extreme (680i chipset).

I've tried two brands of RAM: Mushkin XP2-9200, and OCZ PC2-8500 (Reaper edition), each one are 2 1GB DIMMs. I tried them separate and together, and both of them have low memory bandwidth (measured in both SiSoft and Memtest).

I've triple-checked (and experimented with) the BIOS settings. I also have the latest BIOS.

In the BIOS, my system bus and memory bus speeds are both set to 1066MHz (effective). My memory timings are all 100% correct.

When I benchmark in SiSoft, I get 5348MB/s int, and 5413MB/s float.

According to SiSoft, a 680i with PC2-6400 (much slower than PC2-8500) gets 5799MB/s int an 5777MB/s float.

Something is clearly very wrong here. Why are PC2-8500 RAM DIMMs (two different brands with two different manufacturers) both consistently slower than PC2-6400? Something's gotta be wrong with the motherboard (hardware, BIOS, or drivers...)

I'm running Windows Vista 32-bit by the way.

 

[Asher]

This may be a BIOS/chipset bug...I noticed both of us have QX6700 processors.

 

[champmanfan]

This may be a BIOS/chipset bug...I noticed both of us have QX6700 processors.

I've got my mobo updated with the latest BIOS update to support CPUs with FSB of 1333+.

It would be good if anyone with the core 2 duos could see if theres a bandwidth issue with the Core 2 series or just our quad cores. Our PC8,500 is still pretty fast and the fastest RAM we could get is PC10,000 so unless you manage to experiment with better tweaks than me as posted above, let me know

 

[Adhmuz]

Sorry but that makes no sense at all... I have 2 1GB DDR 3200 and i do get the suggested 6.4 GB/s bandwidth exactly as i should, heck i even get 6.5 GB/s, I've noticed in your sandra screen that it shows a "Bandwidth Efficientcy" which is 57% Estimate, not sure if thats a problem or not, but it would make a little more sense, either that or ddr2 is complete and total garbage, you decide. Try using Everest to test your memory.

 

[Asher]

I've used Memtest86, SiSoft, Everest, and RightMark...all come with similar values.

There's something that's screwing up the efficiency on our chipsets somewhere.

 

[champmanfan]

I've used Memtest86, SiSoft, Everest, and RightMark...all come with similar values.

There's something that's screwing up the efficiency on our chipsets somewhere.

I'm downloading EVEREST Home Edition v2.20 now so I'll post the results later. I did search & find this page for RAM comparison to ours of 66% efficiency.

On the Intel weirdness of things, there might be an issues with MS Vista because one of the latest updates was to fix a compatibility prob with the CPUs so you should be using XP for all your testing otherwise your results aren't reliable or trustworthy. Although SANDRA should really be tested on a 64bit OS since our QX6700, 680i & RAM is designed for better things than 32bit. I wonder if there is much difference?

I always use XP & have done for these tests but if Adhmuz thinks his RAM is good for bandwidth then could you run the same quick benchmark in SANDRA for comparison

I'm sure theres more to this than just software issues so the more people that can work together on this then we can get to bottom of this. I don't think even AMDs have ever had problems with bandwidth so I would prefer we target any of the Intel chips from the Pentium 4 onwards for benching memory bandwidth. Theres more on getting more Ghz/Mhz from Answers.com on Core2's here for some useful info if this helps with more bandwidth

No Celeron's please

 

[Asher]

I benchmarked this on Vista before the Core 2 microcode update, which had no impact on my results.

 

[champmanfan]

I benchmarked this on Vista before the Core 2 microcode update, which had no impact on my results.

Maybe it fluctuates with Intel's CPU so maybe the update stabilizes it or it just doesn't happen to our Quads.

I've updated my previous post with 2 URLs if those help.

 

[sellmesanity]

The ASUS P5K-E is the only board that can run that memory at it's native speed.

 

[MetalX]

My P5B Deluxe has a BIOS option for DDR2-1066 when the FSB is at stock. It increases based on FSB changes.

 

[champmanfan]

My P5B Deluxe has a BIOS option for DDR2-1066 when the FSB is at stock. It increases based on FSB changes.

My mobo is BFG nForce 680i SLi (Socket 775) PCI-Express DDR2 Motherboard and supports Dual Channel DDR2 at 400/533/667/800mhz from what I read in my manual but also support DDR2-1200 & beyond in SLi so the problem isn't with the mobo as I have the latest BIOS update. My CPU also supports FSB of 1333mhz & beyond. I have a 1Kw Enermax Galaxy PSU so I can rule out a lack of power for the components.

As it says on the BFG website: "DualDDR2 Memory Architecture
A state-of-the-art DualDDR2 memory controller allows high bandwidth and low latency data access to the CPU and GPU. Ensures data and information are relayed through the system as quickly as possible for incredible performance."

Its one of the best boards for quad core and SLi DDR2 RAM but its annoying that I can't as yet pinpoint where I can best achieve more bandwidth beyond what I already have clocked the FSB at because it crashes/restarts just before Windows loads up. Don't know why it does that if it gets past the BIOS ok.

 

[Adhmuz]

I always use XP & have done for these tests but if Adhmuz thinks his RAM is good for bandwidth then could you run the same quick benchmark in SANDRA for comparison

Well after running SANDRA like you asked my results are about half of what I'm getting with Everest, about 3GB/s which is actually half exactly so I dunno maybe it's SANDRA throwing off the results.

 

[Mirob]

I have a set of OCZ PC8500 SLi memory in a Biostar Tforce 965PT with a E6300. It will run the memory at 1066 5-5-5-15 but not when I'm overclocking much. At 1.86GHz and DDR-800 3-4-4-12 it has a Sandra bandwidth score of 5333, at 3.045GHz and DDR 870 4-4-4-12 it's 7388. If I only could get 4:5 ratio to work at that speed it might be closer to 8500.

 

[sellmesanity]

I am telling you it is the board

 

[champmanfan]

At 1.86GHz and DDR-800 3-4-4-12 it has a Sandra bandwidth score of 5333, at 3.045GHz and DDR 870 4-4-4-12 it's 7388.

I can't even get my RAM to lower lantencies of 4-4-4-12 from 5-5-5-15 because it won't boot. My voltage is 2.2v which are the defaults so do you have to increase anything to get it stable or would lowering to 800 like you help if thats what you had to do?

I did try a FSB of 1332mhz but my PC got the BSOD after 20mins even though in that time I did manage to get SANDRA to benchmark the memory bandwidth: 6942mbs/6964mbs.

Its easy enough to get to 1280 FSB but anything beyond that ends up in a restart so stability is a problem for me. It may be the CPU that caused the crash cause the max FSB on the QX6700 is 1066 whilst the QX6800 is 1333FSB. Upon reading this article it may seem that the CPU isn't as good as the QX6800 for upping the FSB to 1600+ to achieve 8.7gb/s in SANDRA but even this crashed their setup.