Great results, but I can't explain why. Insight appreciated.

Hi all,

Here are my specs.

Core2Duo 8500
G.Skull F3-10666CL9D-4GBNQ, 4x2 GB, stock freq 1333 Mhz, 9-9-9-24 lat, 1.5v
9800 GX2
Custom watercooling for the CPU, GPU and MCP
750W Corsair

With everything set to defaults in the BIOS (FSB 333) I was getting a read speed for the memory of 7700 MB with Everest.

When I was trying to OC the CPU with upping FSB I quickly ran into stability issues after 350 FSB which was equal to 3.4 Ghz CPU and 1400 Mhz for the memory. At that speed the memory was reading with about 8200 MB/s.

I though the memory was holding me back and strapped it down @ 1066 Mhz. Now I increased the FSB to 390 which put my CPU @ 3.7 Ghz with an absolutely stable system ( played games for a week with it now).

The thing is that even with FSB 390 my memory is still underclocked because it is running on 1248 Mhz now. So I expected results slightly lower that stock reading speed of 7700MB/s. But to my absolute amazement Everrest showed 9005 MB/s reading speed!! This is way over the 8200 I was getting even with the memory on 1400 Mhz! Write and copy speeds showed just as dramatic improvement. I changed timing from 9-9-9-24 to 8-8-8-22 but this actually made insignificant difference.

There is something I don't know here and I am really curious to find out.

Thanks to all try answer!

That microarchitecture has the memory controller on the Northbridge. When you up the fsb, you're increasing the rate that you're pushing data from RAM -> Northbridge -> Processor and back the other way too.

Shuffling all that data back and forth was apparently more of a bottleneck than grabbing it from RAM alone.

I'm not sure if the data has to go through the Northbridge itself but its something like that.

Kampfzerstorer said:

That microarchitecture has the memory controller on the Northbridge. When you up the fsb, you're increasing the rate that you're pushing data from RAM -> Northbridge -> Processor and back the other way too.

Shuffling all that data back and forth was apparently more of a bottleneck than grabbing it from RAM alone.

I'm not sure if the data has to go through the Northbridge itself but its something like that.

Click to expand...

yeah, pretty much. your just increasing the throughput of the cpu and nb which has more of an influencing factor that the RAM speed.

That explains it. I had the same happen to me last night. I had the FSB at 500mhz with a 3:2 divider that put my RAM back to stock 1333mhz with tight 7-7-7-20 1T timings with P1 and P2 Enabled and low tRC, etc, and all of a sudden I was getting 9.7Gb/s out of the memory. Heck of a bottle neck at the northbridge indeed!

This just gave me an idea for tonights testing session (heh I've had my system for all of 4 days and it's just now realizing it came to life to suffer and get beat down!!). I was hitting 500mhz FSB to try to get to 4.5Ghz, and I did get there stably but my air cooling is not enough for the VCore required, maybe I could mess with other things to bring VCore down a bit and try to control the temps but I think I might just try and keep the 500mhz FSB, lower my multiplier to go back to 4Ghz with stock VCore and have all this healthy Memory bandwidth at my disposition.

yeah I always go for higher FSB.

The chipset is limiting the memory, as the best DDR3 in dual channel can reach 20Gb/s on an i7, with over 30Gb/s with triple channel.

Badbonji said:

The chipset is limiting the memory, as the best DDR3 in dual channel can reach 20Gb/s on an i7, with over 30Gb/s with triple channel.

Click to expand...

Damn that makes me feel like my DDR3 @ 10Gb/s is dog slow.

ravaneli said:

Hi all,

Here are my specs.

Core2Duo 8500
G.Skull F3-10666CL9D-4GBNQ, 4x2 GB, stock freq 1333 Mhz, 9-9-9-24 lat, 1.5v
9800 GX2
Custom watercooling for the CPU, GPU and MCP
750W Corsair

With everything set to defaults in the BIOS (FSB 333) I was getting a read speed for the memory of 7700 MB with Everest.

When I was trying to OC the CPU with upping FSB I quickly ran into stability issues after 350 FSB which was equal to 3.4 Ghz CPU and 1400 Mhz for the memory. At that speed the memory was reading with about 8200 MB/s.

I though the memory was holding me back and strapped it down @ 1066 Mhz. Now I increased the FSB to 390 which put my CPU @ 3.7 Ghz with an absolutely stable system ( played games for a week with it now).

The thing is that even with FSB 390 my memory is still underclocked because it is running on 1248 Mhz now. So I expected results slightly lower that stock reading speed of 7700MB/s. But to my absolute amazement Everrest showed 9005 MB/s reading speed!! This is way over the 8200 I was getting even with the memory on 1400 Mhz! Write and copy speeds showed just as dramatic improvement. I changed timing from 9-9-9-24 to 8-8-8-22 but this actually made insignificant difference.

There is something I don't know here and I am really curious to find out.

Thanks to all try answer!

Click to expand...

Is that read or write . ?

I was talking about read speed shown in Everest Ultimate.

What do you guys measure the bandwith with? Sandra? Can Everest measure that?

I think it can yeah...

ravaneli said:

I was talking about read speed shown in Everest Ultimate.

What do you guys measure the bandwith with? Sandra? Can Everest measure that?

Click to expand...

Yes in everest there is a read bench and a write bench

Ha! Cool. Never came upon that, but wasn't relly looking for it. So much info in Everest..

Can't wait to go home and check it out..

OK, everest doesn't have bandwidth information. It has read and write and copy, but not bandwidth.

Anyway, I downloaded a cracked Sandra, and it showed same reading speed (9200 MB/s right now @ 400 FSB) like in Everest. And then Sandra showed me bandwidth of ~ 7.8 GB/s.

How da hell is that possible? The bandwidth is the maximum possible speed, but my reading speed is faster?

Not exactly sure, but could it be do to with buffering or how the RAM works with latencies? : \
Everest doesn't like my Windows 7, BSoD trying to launch it each time so I just keep to Sandra

talk of piracy here is a no no, might want to edit your post

Honestly, my trust is Sandra is kinda shaky. Like it reports 12 000 degree temperatures, my friend's DDR2 800 ram is faster and has more bandwidth than mine, when Everest says mine is twice as fast..
Something about the display of the data looks unprofessional too..

Everest looks better and is more consistant in the numbers..

Sandra's benchmarks are fine, seem to be in line with my pc compared with other setups. As for sensors, yeah both tend to have problems showing correct temps etc., happened on my 780i when it said my northbridge or something was -49C, if only