Poor Triple Channel Performance? (with pictures!)

[chjade84]

I just finished the build of a new computer for work around the i7 and after running some benchmarks it appears that the OCZ 6GB (3 x 2GB) 240-Pin DDR3 SDRAM DDR3 1333 is about the same speed as my G.SKILL 4GB (2 x 2GB) 240-Pin DDR2 SDRAM DDR2 1066.

The Triple Channel runs at 7-7-7-20 and the Dual Channel runs at 5-5-5-15.

This is also my first real experience in overclocking RAM however I have done several simple CPU overclocks in the past. Making this even harder, I've never built a computer around the i7 until now.

The other specs:

i7
Asus P6T
i7 920 2.6 Ghz (OC to 3.2 Ghz)
6GB DDR3 1333

C2D
Asus P5Q-E
E8500 3.16 Ghz (OC to 4Ghz)
4GB DDR2 1066

Should I be expecting any more performance out of the Triple Channel in this setup? What am I (quite likely) doing wrong?

Here are the benchmarking results (Performance Test x64). The first 5 columns are the C2D setup; Work has XP, Home has Vista. The last column is the computer this new one will replace.



And my BIOS screens:

 

[merlinbadman]

I should be able to help you as I have exactly the same components, and at home I have my SANDRA memory bechmark scores written down (I can get them later if needed to compare).

Im not familiar with the bechmarking suite you are using, could you install SANDRA and have a go with that? Specifically to ascertain the memory bandwidth.

Anyway, for starters I suggest you use the XMP overclocking profile, rather than the DOCP, it uses info stored in the RAM's SPD to set the ram up automatically. Your/our OCZ ram supports the use of XMP.

While it didnt set my correct timings it did change alot of other things eg QPI Link speed. However I was easily able to manually change these afterwards to the correct 7-7-7-20 1T.

What is your QPI Link speet ATM? CPU-Z should tell you.

 

[chjade84]

Well... thanks for the reply.

Unfortunately I was trying to isolate the noise from the Raptor HDD and in the process of doing to shorted out part of the PCB on the all aluminum chassis of the Lian LI case. A small plume of smoke later and it looks like I'll be needing a new hard drive.

Lesson to myself: Turn off power before moving the hard drive around!

Now that I think about it I'm surprised this hasn't happened before considering how many times I've done the same thing in the past.

Check back in a few days and I'll have the Sandra data for you.

Thanks!

 

[merlinbadman]

Oh well, thats a shame.

What you going to get instead? Raptor again, SAS (P6T has built in SAS controller), or a normal 7200rpm drive?

 

[chjade84]

I'm sorry I've not heard the term SAS before. Is that the same as RAID?

As far as the next drive? I'm still one the fence. You can read my other post (still waiting for a reply) about that here. http://www.tomshardware.com/forum/248176-32-raptor-faster

What are your thoughts? RAID would be a possibility but I hear it's not (in practicality) faster.

 

[Inigo Montoya]

I believe SAS stands for serial attached scsi, and the asus P6T deluxe has a built in controller (version 1 of the p6t, it doesn't come on V2 of the P6T). Bit of info on SAS

I think I have the same setup as you guys, and I 'm having a little trouble figuring out the biossettings.
(I haven't been keeping up with all the latest info, up untill recently my old xp 2400+ with ti4600 was sufficient ) I didn't know the OCZ worked with XMP, I just tried it, and if I put the "AI overclock Tuner" into "XMP" mode, a grey line apears underneath that says " eXtreme Memory Profile [disabled]"
Do I have to change a setting somewhere else first? Or do I need to update the bios so the memory is recognized correctly? (I haven't flashed the bios yet, because I haven't encountered any other problems (the old: if it ain't broke...)

BTW, I have the 300 Gig velociraptor and I was pleasantly surprised by the (lack off) noise. It's considerably less noisy than the old WD Caviars in the other machine, but that's probably due to the assembly in the antec P182 vs the old chieftec case.

 

[geofelt]

In some real application benchmarks(vs. synthetic ram speed tests) different ram configurations and speeds seem to have little impact on i7 performance.
The i7 memory handler seems to be able to deliver data well from almost any ram configuration.

I suggest that you test your different ram setups with some real applications and see if your results are still good.

 

[Zorg]

Interesting, do you have any links that you could help me out with? If not I'll Google it.

 

[geofelt]

Interesting, do you have any links that you could help me out with? If not I'll Google it.

I don't recall if this was one I saw, but it is a start:
http://www.techspot.com/article/131-intel-corei7-memory-performance/page15.html

 

[cyberjock]

Just a theory, as I do not buy Asus products anymore after their stunt with their power efficiency game with their P5 series motherboard.

You installed three 2GB sticks. Are you using a 32bit OS? Although the triple channel design is supposed to be interleaved, maybe there is a design flaw that actually doesn't use interleaved with 32bit OSes and 3x2GB sticks? You'd only be using the first 2 sticks then totalling 4GB.

Again, just an idea. Further testing would have to be performed to confirm it. I personally run 3x1GB sticks on a Gigabyte motherboard, but they will be supplemented later today with 3x2GB sticks.

 

[Zorg]

I don't recall if this was one I saw, but it is a start:
That]http://www.techspot.com/article/131-intel-corei7-memory-performance/page15.html

That works, thanks.

 

[chjade84]

I believe SAS stands for serial attached scsi

That's right. I have yet to use SCSI drives and I don't plan to as they are way too expensive for what they are (IMHO).

Are you using a 32bit OS?

No, it's XP x64. I see all 6GB in windows, or at least real close to that.

I'll be getting a new hard drive on Monday so I can post some benchmarks then.

Thanks!

 

[merlinbadman]

That's right. I have yet to use SCSI drives and I don't plan to as they are way too expensive for what they are (IMHO).

Actually the 147GB SAS drives are very similar in price to the 150GB raptor. The SAS drives out perform the 150 raptor in every test I have seen (THG charts).

Plus since the P6T supports SAS (no need for an expansion card like traditional SCSI), I think SAS is a very attractive option. I would have gone for one in my new build, but I have a an old 150 raptor so didnt see the point.

 

[geofelt]

Actually the 147GB SAS drives are very similar in price to the 150GB raptor. The SAS drives out perform the 150 raptor in every test I have seen (THG charts).

Plus since the P6T supports SAS (no need for an expansion card like traditional SCSI), I think SAS is a very attractive option. I would have gone for one in my new build, but I have a an old 150 raptor so didnt see the point.

On a previous build, I bought some 15k scsi drives on e-bay, along with a pci scsi controller cheap. They worked fine, but realistically, they were no faster than the raptor.

You can now buy 148gb 15k SAS drives on e-bay for $50 or so. Server farms are selling them off in favor of larger and cooler devices. They work well, but they really aren't well suited to the single user desktop environment. The access times are better because of the 15k rotation speed. That does not translate to faster data transfer speeds because their aerial density is not as great. The velociraptor is actually much faster in most tests. Look at www.storagereview.com for some benchmarks, particularly for the office drivemark 200t which most closely represents what you and I might do every day.
http://www.storagereview.com/php/benchmark/bench_sort.php

 

[geofelt]

On a previous build, I bought some 15k scsi drives on e-bay, along with a pci scsi controller cheap. They worked fine, but realistically, they were no faster than the raptor. That is, in part, because their microcode is optimized to the fast random access pattern needed by servers. The raptor and velociraptor are more optimized to a mix of sequential and random processing.

You can now buy 148gb 15k SAS drives on e-bay for $50 or so. Server farms are selling them off in favor of larger and cooler devices. They work well, but they really aren't well suited to the single user desktop environment. The access times are better because of the 15k rotation speed. That does not translate to faster data transfer speeds because their aerial density is not as great. The velociraptor is actually much faster in most tests. Look at www.storagereview.com for some benchmarks, particularly for the office drivemark 200t which most closely represents what you and I might do every day.
http://www.storagereview.com/php/benchmark/bench_sort.php

 

[blackened144]

On a previous build, I bought some 15k scsi drives on e-bay, along with a pci scsi controller cheap. They worked fine, but realistically, they were no faster than the raptor.

You can now buy 148gb 15k SAS drives on e-bay for $50 or so. Server farms are selling them off in favor of larger and cooler devices. They work well, but they really aren't well suited to the single user desktop environment. The access times are better because of the 15k rotation speed. That does not translate to faster data transfer speeds because their aerial density is not as great. The velociraptor is actually much faster in most tests. Look at www.storagereview.com for some benchmarks, particularly for the office drivemark 200t which most closely represents what you and I might do every day.
http://www.storagereview.com/php/benchmark/bench_sort.php

Your experiment with 15k drives was stunted by your use of a PCI controller card. The theoretical bandwidth for the entire PCI system 133Mbps. Thats about as fast as 1 drive and Im assuming (based on what I made bold above) you had multiple drives connected to the card. Did you have any other PCI devices? They would have cut into the max bandwidth available to the controller. Also if it was an older system like my old nForce3 board where the gigabit controller was incorporated into the PCI bus, you dont even need a card installed to eat away at the max bandwidth for the bus. I ran into the same problem with my raided 10k SCSi drives on my PCI controller, it was no better than my 7200 raid array. Had you used a PCIx or PCIex controller card, you would have seen serious improvement. Anyway, as far as the vraptor being faster than the 15k drives, that is correct. But at $50 for 146gb SAS, you can get 2, put them in RAID0 and blow away the performance of the vraptor and because of the 15k spindle speed the access time overhead of RAID0 would be insignificant. All cheaper than the one vraptor.

 

[wimcle]

But at $50 for 146gb SAS, you can get 2, put them in RAID0 and blow away the performance of the vraptor and because of the 15k spindle speed the access time overhead of RAID0 would be insignificant. All cheaper than the one vraptor.

of course carrying that argument one step farther, at the same price you could stripe 7200rpm sata drives and blow away the vraptor

 

[chjade84]

Now see, I had a bad experience with RAID 0 so I haven't had the desire to use it again. Everything works just fine but the performance is not nearly what I would have expected.



The work and home computers are identical except for the video card and the fact I tried RAID 0 using the Asus P5Q-E onboard RAID controller. Both computers have Seagate Barracuda 7200.10's. The i7 columns have the Raptor. The final column is the computer this (Raptor) will replace. Not sure on the drives though.

As you can see there is not that much of an improvement in using RAID, at least in this setup. What gives?

Wimcle what does that RAID consist of? That looks real nice!

SAS is out of the question as I can't build work computers with used drives and the new ones are out of budget. (I already went over by about $250/each)

I just need the best performance for $100 and 75+ GB.

(BTW how do you post images full size?)

 

[wimcle]

thats 4 wd6401aals drives $70 apiece

as far as the picture, I just clicked on the picture button above the message edit box and paste in the flickr url

 

[blackened144]

Now see, I had a bad experience with RAID 0 so I haven't had the desire to use it again. Everything works just fine but the performance is not nearly what I would have expected.

[url]http://img171.imageshack.us/img171/3706/diskty5.th.jpg[/URL]

The work and home computers are identical except for the video card and the fact I tried RAID 0 using the Asus P5Q-E onboard RAID controller. Both computers have Seagate Barracuda 7200.10's. The i7 columns have the Raptor. The final column is the computer this (Raptor) will replace. Not sure on the drives though.

As you can see there is not that much of an improvement in using RAID, at least in this setup. What gives?

Wimcle what does that RAID consist of? That looks real nice!

SAS is out of the question as I can't build work computers with used drives and the new ones are out of budget. (I already went over by about $250/each)

I just need the best performance for $100 and 75+ GB.

(BTW how do you post images full size?)

There are not many advantages to RAID0. I have 2 RAID0 arrays in my machine and I use them for working with large files, but my OS is on a single 400gb drive. Most often I will extract from one array to the other and doing that is almost twice as fast. Ive installed my OS on the 7200rpm RAID0 array and even though the throughput is higher than a standard disc, the OS seemed slower because of the higher access times associated with RAID0. This is where the cheap 15k SAS drives in RAID0 would kill the 7200 RAID0. But you mentioned only buying new, so that would make the SAS solution much more expensive and not worth it in my opinion. What I have personally been looking at to run the OS are some of the cheaper SSD drives. From what I have read the newer ones dont suffer from the lag as much as the older ones, and putting them in a RAID0 array virtually eliminates the lag altogether. You can get 2 30gb SSD drives for ~$120. I have the RAID0 arrays for storage, so 60gb would be enough for me. Anyway, if you were happy with your setup before, I would just get another vraptor and be done with it.

 

[blackened144]

of course carrying that argument one step farther, at the same price you could stripe 7200rpm sata drives and blow away the vraptor

http://img187.imageshack.us/img187/2159/hdtunelo7.png

It would blow it away in raw throughput, but because of the access time overhead inherent in RAID0 the OS will "feel" more sluggish with the 7200rpm drives in RAID0 than the vraptor by itself. This is why i ahve 2x400gb RAID0, 2x500GB RAID0, and my OS installed on a lone 400gb drive. The SAS RAID0 would overcome that inherent overhead regarding access times due to the 15k spindle speed and blow them all away, especially at $50 a drive..

 

[geofelt]

Your experiment with 15k drives was stunted by your use of a PCI controller card. The theoretical bandwidth for the entire PCI system 133Mbps. Thats about as fast as 1 drive and Im assuming (based on what I made bold above) you had multiple drives connected to the card. Did you have any other PCI devices? They would have cut into the max bandwidth available to the controller. Also if it was an older system like my old nForce3 board where the gigabit controller was incorporated into the PCI bus, you dont even need a card installed to eat away at the max bandwidth for the bus. I ran into the same problem with my raided 10k SCSi drives on my PCI controller, it was no better than my 7200 raid array. Had you used a PCIx or PCIex controller card, you would have seen serious improvement. Anyway, as far as the vraptor being faster than the 15k drives, that is correct. But at $50 for 146gb SAS, you can get 2, put them in RAID0 and blow away the performance of the vraptor and because of the 15k spindle speed the access time overhead of RAID0 would be insignificant. All cheaper than the one vraptor.

It was actually only one drive. The performance was not measured, but I sensed no improvement in snappiness of the OS or any improvement of boot times. The controller was only $25, so it did not cost much to experiment. As I recall HDTACH transfer numbers were similar to the raptor, and access times were a bit better. I figured that two drives would be ok, since data transfer from both would not be simultaneous that often and a 133 bus would not be much of a problem. I had no other pci devices. I think a good raid card with a large cache might have done better, but that would have been really pricey.

I gave some consideration to using the SAS controller on my P6T and trying a pair of 15K drives in raid-0, but the native data transfer rates of the drives is much less than the vraptor. In the end, I think that SSD development will produce a much better solution by the end of the year. For now, I think I will stick with the simplicity and consistently good performance of the vraptor.