PCI-E bus overclockers check in here!

[Rubycon]

Anyone running > 100MHz please tell your story here.
Why do you do it? (i.e. higher bclk stability, increased 3D performance, higher i/o throughput, etc.)
What was the cracking point?

 

[TurtleCrusher]

I thought there was zero performance gain until the lanes were completely saturated. Overclocking the PCI-E bus is useless, isn't it?

 

[Rubycon]

My Areca 1680ix-24 says otherwise. It's in an 16X electrical (running at 8X) and performance increased considerably.

 

[Sylvanas]

There was a thread on XS a while back about this and some have said by increasing the PCI-E frequency allowed them a higher GPU core clock. I tried this out back in the day on my X850XT and PCI-E 1.0 motherboard and I did see about 10mhz stable improvement. It can help alleviate bandwidth issues too for motherboards with multiple cards example being the P35 (16x and 4x when in Crossfire mode). I had mine at 125mhz which was what the Nvidia based mobos where increasing the frequency automatically to when SLI or an Nvidia card was detected- they called it 'Linkboost' a feature which I think has faded in their more recent chipsets. I had no problems at this frequency but I didn't go above this as per XS some people experienced HDD corruption on the SATA bus at 130+.

 

[Idontcare]

Originally posted by: Rubycon
My Areca 1680ix-24 says otherwise. It's in an 16X electrical (running at 8X) and performance increased considerably.

Pushing those IOP processor's clockspeed, active cooling on the heatsink?

 

[VirtualLarry]

I had my friend's system at 115Mhz. It was an IP35-E, with a 250GB IDE HD and IDE DVD burner, and a PCI-E 9600GSO.

Didn't notice any real performance gains, but did it as a test, because I was planning on sticking a 9600GSO in a different friend's system, with a 945GC chipset, and due to overclocking, the PCI-E bus would have ended up at around 115Mhz.

Well, one day, my friend went to burn some DVDs, and got "IDE DMA CRC Error" several times, along with the machine locking up (apparently it wasn't able to reset the IDE bus). The reason is that the IDE bus is on the Jmicron controller, which is attached to the PCI-E bus. It does not like being overclocked!

Interestingly, it never caused a problem with just the HD, but when using both the IDE HD and the IDE DVD burner, it got really screwed up.

 

[rogue1979]

I am currently running a 120MHz fsb speed. Not out of choice, my motherboard only allows overclocking with the "pci-e sync" mode, "async" limits it to very little overclocking on the fsb.

The pci bus is on a different controller and is still running 100MHz. This on a 945C chipset.
I haven't noticed any extra pci-e performance, though when I was running a 9600GT @ 775/1900/2050 it would overclock the core and shaders 26% and I would have to adjust the speeds down accordingly. It didn't affect the overclock the 9600GT memory.

My motherboard tops out a 352MHz, because after that the sata bus disappears. I tested briefly on IDE and got it to post all the way past 400MHz. I have no actual benchmarks, but I'm pretty sure it overclocks the sata bus and it makes the whole system "feel" snappier by a very noticeable amount

Been using the board for about a year and half and no hdd corruption or dvd-rw problems ever.

 

[Rubycon]

Originally posted by: Idontcare

Originally posted by: Rubycon
My Areca 1680ix-24 says otherwise. It's in an 16X electrical (running at 8X) and performance increased considerably.

Pushing those IOP processor's clockspeed, active cooling on the heatsink?

Not sure if it stresses the IOP more as its speed is locked at 1.2GHz. The extra speed on the bus gives it more overhead to push data, however. 110 is the highest I feel comfortable with. I may try higher though. At least with a smart host if there is a problem you know before it starts eating your data. The onboard NIC is unreliable at this speed so I had to put a Intel PCI-E server NIC in there.

Cooling is stock - a small heatsink fan like used on older video cards like the Geforce2 series. The card will work without a fan but needs good channeled airflow (common in server environments).

Getting a little OT:

The performance of SSDs on these hosts is impressive. I replaced a 10 disk SAS 15K array with just four GS Kill Falcons (128) and the performance increase is noticeable. I may sound crazy but I sort of miss the sound of the 15K disks clicking when working hard. I do NOT miss the off zero beat throbbing spindle motor hum, however. Perhaps I could connect a simple device up using the drive activity lights header that clicks when the LED's fire to simulate drive noise! Kids put devices on their car that squeak when they let off the gas that makes a blow off valve noise. At least the former the performance is still real! :laugh:

 

[arkcom]

My computer won't boot with anything above 100. I don't know if it's because of my video card (4850) or the motherboard (ex58-ud3r).

 

[imported_Shaq]

It's the motherboard. Mine does it too.

 

[Rubycon]

Tried 115. Areca BIOS won't load.

Funny thing about the beep on the Areca. Its pitch increases with PCI-E speed! :laugh:

 

[Grnkjr0]

With Gigabyte X-58 mobos...

I have seen over at XtremeSystems forum, that some people are doing mods that allows them to get much higher overclocks than normal, and that is a PCI-E mod using a F8x BIOS version.

Normally you run into a wall with BCLK around 222 on most, but with PCI-E at 130 they are doing 235 BCLK and above. My X-58 UD4P will not boot above 103 PCI-e and BCLK 205.

I have not tried this myself and the mod is not out in the public it seems.

 

[Idontcare]

Originally posted by: Rubycon
Not sure if it stresses the IOP more as its speed is locked at 1.2GHz. The extra speed on the bus gives it more overhead to push data, however.

I was under the distinct impression that the base clock for the IOP processors, cache ram, etc were all multiplier'ed off the PCIe bus (i.e. the bus clock IS the reference clock)...but I will confess I do not have that in writing from anyone at Areca so if you know this to be untrue for fact then I will accept it as such.

But consider how much bandwidth your x8 PCIe slot is making available to the areca card, versus the bandwidth saturation point of the card as you load it with drives, I would be REALLY surprised if the performance bottleneck of a 1680ML is the x8 PCIe interface and not the operating specifications of the controller IC.

(also consider that the entire evolution of the areca card line performance-wise has solely focused on improving the controller IC and not the PCIe interface, further suggesting the bottleneck in performance is not the interface bandwidth, or at least has not been up until the most recent models)

Is the peak performance of a 1680ML really interface bandwidth limited?

 

[vj8usa]

Originally posted by: Scholzpdx
I thought there was zero performance gain until the lanes were completely saturated. Overclocking the PCI-E bus is useless, isn't it?

Even with video cards, it can help a lot when you're filling your VRAM and constantly swapping back and forth.
http://www.pcgameshardware.com...cards/Practice/?page=2

 

[Rubycon]

Originally posted by: Idontcare

I was under the distinct impression that the base clock for the IOP processors, cache ram, etc were all multiplier'ed off the PCIe bus (i.e. the bus clock IS the reference clock)...but I will confess I do not have that in writing from anyone at Areca so if you know this to be untrue for fact then I will accept it as such.

But consider how much bandwidth your x8 PCIe slot is making available to the areca card, versus the bandwidth saturation point of the card as you load it with drives, I would be REALLY surprised if the performance bottleneck of a 1680ML is the x8 PCIe interface and not the operating specifications of the controller IC.

(also consider that the entire evolution of the areca card line performance-wise has solely focused on improving the controller IC and not the PCIe interface, further suggesting the bottleneck in performance is not the interface bandwidth, or at least has not been up until the most recent models)

Is the peak performance of a 1680ML really interface bandwidth limited?

Good luck getting any information from Areca! :laugh:

I don't have the software or hardware interface to measure IOP core frequency and cache frequency (and internal bandwidth!) to verify for sure. If I put the card in a 4X slot it is capped around 1GB/S and 4K random suffers. Overclocking PCI-E did not improve Gig-E performance that I can see although I did not compare side by side. From NAS to my box it will hit 99% (995+ Mbps) so another 10% is kind of hard to see without special software.

The 1680ix is definitely faster with SATA drives however. With SAS drives STR is capped at about 900MB/S. I hit this with just four Falcon 128's.