Printable View
This is my first thread concerning memory so please go easy on me! :D
This is a little study on the performance of the G Skill F2-8000CL5D-4GBPQ in a 8GB configuration on an eVGA 780i board.
First the stability tests: I had to go down a bit in FSB over the 680i board to reach stability. Please note the vcore is wrong in cpuz. It is 1.42 under load.
Next, I tried to increase the performance in many ways but I was not really suceessful. However I decided to plot the FSB vs read, write and latency in Everest. One can see that the read and write speed increase rather linearly from FSB=1333 to 1440 with the latency staying at around 54 ns.
At 1460 -1480 the computer cannot boot into Vista 64
At 1500-1600, the read speed drops alot. The write speed on the other hand keeps increasing linearly all the way to FSB 1800 (although I had to change the multi down one notch) but the latency changes to 74 ns.
What does all this mean? Well, I'm new at this...someone got comments?
A little speculation.
There is (are) some chipset primary latency timing(s), similar to Intel's performance level (tRD). On Intel platform, you have direct control over this timing in BIOS as well as with Memset utility, on NVIdia, there seems to be no way you can control it.
Sadly, NVidia does not publish the datasheets for their chipsets as Intel does.
Over 1480 this chipset timing jumps to higher value. This also may explain why you cannot boot in 1460-1480 region - the timing is too tight already for this speed.
I've got a 4GB (2x 2GB) set ordered and with luck I'll see them Friday. I saw a review done that compared 4GB to 8GB and the 8GB didn't always come out on top. I don't remember where I saw this but if you've got the time you could run some tests and see which one is the winner on the 780i. Of course I want to know if I should buy another set!
I ordered mine for $114.99 + Free 3-day shipping from Newegg and I just notice they are now $109.99 + Free 3-day shipping. Will someone please let me know when they are free!
Well in my case, some of the photoshop work I do takes 10 seconds with 8G and 45 minutes with 4G! How's that for an overclock!
Quote:
Originally Posted by Cronos
Any advantages to running at FSB 1800 vs 1400? The CPU speed is the same (3.6 GHz) and most of the benches I have run show little difference. Any disadvantages?
I noticed that you're using a version of Everest Ultimate Edition that doesn't support the 780i motherboard. I'd suggest that you use one of the following beta versions ...
4.20.1257 Beta
4.20.1285 Beta
You can use Google to search for a download site. The last time I checked they weren't listed on the Everest beta forum. I use the 1285 version but I haven't seen any difference with the 1257 version.
When I first saw that you tested 1800 QDR I couldn't believe it but now I see that you dropped the multiplier to 8. The only way you're going to find out if 1800 is better than any others is to test them against a set benchmarks.
For me that includes the ones you listed but also SuperPI 32M, 3DMark06 and the Crysis Benchmark. I'll use the Crysis Benchmark scores as the deciding factor.
The fun begins Friday.
Ya, the multiplier is down to 8. I see little difference in UperPi 1M (0.4 seconds) and in 3DMark06 a couple of hundred. Hardly worth bothering about.
I looked on everest yesterday and didn't find the new version. Thanks for the tip.
Very interesting, could you elaborate on this some? Also, what is the rest of the system set up?Quote:
Originally Posted by tomb18
My guess is opening VERY files that are more than your combined physical memory so the system needs to use pagefile just to open, much less work on. I'm surprised though, since photoshop will only use 3.5gb of memory and I've never found a file to be RAM dependent on a 4gb setup.Quote:
Originally Posted by tomb18
here is a tom's article on 8gb : http://www.tomshardware.com/2008/02/15/vista_workshop/
Its' not something I do very often, but I recently produced a photo of 48"x72" @300 dpi.
If you work in layers, your ram requirements increase substantially. The photo files that I work with are about 65mb (12 mega pixel SLR).
Once you duplicate this layer (you never work on the original layer) you are up to 130mB.
I then have a couple of adjustment layers and then I do mask based sharpening using some automated procedures. I also create some other layers for various processes. Each one of these creates additional layers.
By the time the photo is ready for uprezzing (it's still at about 12x18"at 240 dpi) the file size is 1.2G. I like to keep the layers so that I can go back a quickly change something that is not right. But even if it is flattened (bringing you back to 65M) the uprezzing brings you to about 800 MB. I then do a masked based output sharpening giving a file size of 1.2GB, and two scratch files, one of 2.4 GB and the other around 6GB.
So in 4G of ram the last step uses the disks for the scratch space and takes around 45 minutes. With 8G it all fits and finishes in around 10 seconds.
Of course if you have a lot of files opened in Adobe Bridge (real memory hog) and I'm doing some HDR panorama's 4G is also not enough. This I do more often.
Yellowbeard:
The system is now 8G ram, 3.4 GHz QX6700
150G raptor system disk,
2X400G raid 0 for data,
1X500(for now) scratch space (and backups).
I talked to Fugger recently about exactly this type of work. He's found it beneficial to use a Gigabyte I-RAM for the Photoshop scratch disk. It apprears that you have your Photoshop scratch file on that spare 500gb HD and off the OS/system drive?
That's pretty useless dribble.Quote:
Originally Posted by zfactor
Here's an article from adobe: http://kb.adobe.com/selfservice/view...0005&sliceId=1
On a 32 bit os, photoshop can use a maximum of around 1.6 GB ram for it's images. Anything over this and it uses the scratch disks themselves.
In Photoshop CS2 and CS3, on a 64 bit operating system, with 4G or more ram, PS will use 3 GB for it's image data and images that are larger that used to require scratch disks now can be cached in ram by the OS instead of being written and read to disks. The result is vastly increased performane of course. So far comparing Vista 64 with XP64 it seems that Vista does a much better job of managing the memory.
The tom's hardware must of used an older version of PS which did not have this cabability.
Yes the scratch disk is dedicated to photoshop scratch space. Although I do not need it much anymore :clap: .Quote:
Originally Posted by Yellowbeard
have a look at the adobe article above, it gives a good overview of capabilities of 64bit OS's and more than 4G for photoshop cs2 and cs3.
Nice job tomb18. I appreciate the insight and thank you for your work. :up:Quote:
Originally Posted by tomb18
How much further down did you have to go in FSB with the 680i vs. the 780i?
ThanksQuote:
Originally Posted by ShoNuff
It wasn't much. On the 680i I was running at 1400 qdr with a 10x multi giving 3.5 GHz on the QX6700 (4 sticks 1G ram). I found that with the same settings on the 780i and 8G I would get errors in prime95 at that FSB. All it took though was lowering the FSB by 20. I now run 3.45GHz. Hardly worth mentioning. I suspect that this is a vcore issue since the droops under load goes down alot more than the 680i board I had.
The 680i board that I had (one of the later revisions as well) never let me get my FSB higher than 1450, no matter what I did.
The 780i board, now lets me overclock the FSB at ease, easily getting to 1800. I would run the ram here, but overall performance drops considerably due to the above ram speed changes.
I guess the 780i bios is still immature. What I can't understand is that the write performance continues to increase linearly with FSB while the read drops so much. I would have expected both.
Also, why did nVidia see fit to change the latencies in the chipset at 1480 with 8G? Why not leave it be as they do with 4G of ram? (The letency changes does not seem to occur in this case) Let us overclockers play with all of the options (voltages etc) to try to get our own stability. nVidia seems to be forcing stability on this case according to default system parameters basically leaving us with a system that will only do 1333-1460. Hardly a top of the line overclocking board.