16GB installed, only 12GB usable in Win7 x64 Ultimate
#1
Posted 06 October 2012 - 12:24 PM
So I rebooted and checked the BIOS and that shows 16GB. I changed it from 1600Mhz to 1333mhz and rebooted in Windows and now it shows the entire 16GB as usable. It's Corsair Vengeance LP DDR3, which is specified to work at 1600Mhz at 1.5v. CPU-Z shows JEDEC#4 as 666Mhz@1.5v and XMP-1600@1.5v. I've left the Voltage and timings in the BIOS on Auto.
I actually just had one 2*4GB kit replaced as one of the sticks went bad and the other I just bought so that I could use the PC whilst I was waiting for the original kit to be replaced, so I hope another stick hasn't gone bad!
#2
Posted 06 October 2012 - 02:16 PM
#3
Posted 06 October 2012 - 06:34 PM
#4
Posted 06 October 2012 - 07:06 PM
doveman, on 06 October 2012 - 06:34 PM, said:
I bet your machine got nothing wrong but, despite the manufacturer's claims, it's northbridge simply cannot control that much RAM that fast, so it skips part of it. When you set the RAM a little less fast, however, then it's able to cope with all of it. Set it again to 1600 MHz... do 4 GiB diappear? Return it to 1600 MHz... do all 16 GiB get detected once again?
#5
Posted 06 October 2012 - 09:11 PM
dencorso, on 06 October 2012 - 07:06 PM, said:
Wouldn't surprise me. Manufacturer's always obfuscate about the capabilities of their boards
I'll test as you suggest but just want to be clarify that from the tests I already did the BIOS detects all 16GB at 1600Mhz or 1333Mhz, as does Windows but 4GB of it is unusable in Windows at 1600Mhz, whereas it's all usable at 1333Mhz.
If I can't run it at 1600Mhz, should I still be able to overclock the NB to 2600Mhz as I did when running with only 8GB? That would help make up for it. In fact, when I tested before that made more difference, although obviously ideally I'd be able to run both the RAM at 1600Mhz and the NB at 2600Mhz.
#6
Posted 06 October 2012 - 10:47 PM
doveman, on 06 October 2012 - 09:11 PM, said:
That tells one that Windows Memory Control is tighter than the one done by the BIOS, but that's just what's expected, anyway.
doveman, on 06 October 2012 - 09:11 PM, said:
The only way to know for sure is by testing.
#7
Posted 07 October 2012 - 06:26 AM
dencorso, on 06 October 2012 - 10:47 PM, said:
doveman, on 06 October 2012 - 09:11 PM, said:
That tells one that Windows Memory Control is tighter than the one done by the BIOS, but that's just what's expected, anyway.
doveman, on 06 October 2012 - 09:11 PM, said:
The only way to know for sure is by testing.
OK, thanks.
#8
Posted 07 October 2012 - 09:31 AM
- WHAT is the need for 16 GB? <--"absolute" question
- WHICH is "better" (i.e. faster/more useful/advisable)? (12 Gb@1600Mhz or 16 Gb@1333Mhz) <--"relative question"
and a third one coming purely from curiosity
- WHAT the heck are you running on that machine that *somehow* prompted you to decide that the previous 8 GB were not enough?
jaclaz
#9
Posted 07 October 2012 - 01:36 PM
jaclaz, on 07 October 2012 - 09:31 AM, said:
- WHAT is the need for 16 GB? <--"absolute" question
- WHICH is "better" (i.e. faster/more useful/advisable)? (12 Gb@1600Mhz or 16 Gb@1333Mhz) <--"relative question"
and a third one coming purely from curiosity
- WHAT the heck are you running on that machine that *somehow* prompted you to decide that the previous 8 GB were not enough?
jaclaz
Well the third question is addressed in my first post
I can't answer your other questions but perhaps these benchmarks with 8GB I did a while ago will be of some use. Note I don't think I ever got it stable with the NB at 2600Mhz though, so settled for 2400Mhz.
DDR3-666 (1333) Mhz (9-9-9-24-1T), NB 2000Mhz:
Copy 10466 MB/s
Read 8107 MB/s
Write 6831 MB/s
Score 7.47GB/s
Latency 66.9 ns
---
DDR3-800 (1600) Mhz (9-9-9-24-1T), NB 2000Mhz
Copy 10560 MB/s
Read 8667 MB/s
Write 6734 MB/s
Score 7.70 GB/s
Latency 65.0 ns
---
DDR3-800 (1600) Mhz (9-9-9-24-1T), NB 2400Mhz
Copy 11602 MB/s
Read 9457 MB/s
Write 7883 MB/s
Score 8.67 GB/s
Latency 61.5 ns
---
DDR3-800 (1600) Mhz (9-9-9-24-1T), NB 2600Mhz, CPU-NB 1.32v, NB 1.3v
Copy 11867 MB/s
Read 9822 MB/s
Write 8457 MB/s
Score 9.14 GB/s
Latency 60.4 ns
---
DDR3-666 (1333) Mhz (9-9-9-24-1T), NB 2600Mhz, CPU-NB 1.32v, NB 1.3v
Copy 11350 MB/s
Read 9010 MB/s
Write 8583 MB/s
Score 8.80 GB/s
Latency 62.5 ns
#10
Posted 07 October 2012 - 04:50 PM
doveman, on 07 October 2012 - 01:36 PM, said:
DDR3-666 (1333) Mhz (9-9-9-24-1T), NB 2000Mhz:
Copy 10466 MB/s
Read 8107 MB/s
Write 6831 MB/s
Score 7.47GB/s
Latency 66.9 ns
---
DDR3-800 (1600) Mhz (9-9-9-24-1T), NB 2000Mhz
Copy 10560 MB/s
Read 8667 MB/s
Write 6734 MB/s
Score 7.70 GB/s
Latency 65.0 ns
---
So, 1600 vs 1333 means:
Copy 0.9 % faster
Read 6.9 % faster
Write 1.4 % slower
Latency 2.8 % less
jaclaz, on 07 October 2012 - 09:31 AM, said:
- WHAT is the need for 16 GB? <--"absolute" question
- WHICH is "better" (i.e. faster/more useful/advisable)? (12 Gb@1600Mhz or 16 Gb@1333Mhz) <--"relative question"
My replies to your questions:
1) None at all.
2) Since "faster" is almost imperceptible (see above: 7 % at most!), "more" is better (in the sense that there may be some rare occasion where all that memory may be actually filled, perhaps by running numerous virtual machines or the like). The benchmark above confirms it, but common sense alone should suffice to conclude that "more" is better than "faster", in this context, IMO.
#11
Posted 07 October 2012 - 09:34 PM
You would need to tweak it little by little to get the 16GB to work. BIOS update could help sometimes...
#12
Posted 08 October 2012 - 02:47 AM
dencorso, on 07 October 2012 - 04:50 PM, said:
Copy 0.9 % faster
Read 6.9 % faster
Write 1.4 % faster
Latency 2.8 % less
Be aware of the math:
(10560/10466)-1=+0.9%
(8667/8107)-1=+6.91%
(6734/6831)-1=-1.42%
@doveman
Well no.
If you have some "spare" 8 Gb testing them is OK, but nothing actually "prompts" or "forces" you to add it to an otherewise perfectly working machine with already 8 Gb (unless you use the machine with a few high end programs that need heaps of RAM).
jaclaz
#13
Posted 08 October 2012 - 04:01 AM
jaclaz, on 08 October 2012 - 02:47 AM, said:
Well no.
If you have some "spare" 8 Gb testing them is OK, but nothing actually "prompts" or "forces" you to add it to an otherewise perfectly working machine with already 8 Gb (unless you use the machine with a few high end programs that need heaps of RAM)
As I've already explained (twice
#14
Posted 08 October 2012 - 04:27 AM
puntoMX, on 07 October 2012 - 09:34 PM, said:
You would need to tweak it little by little to get the 16GB to work. BIOS update could help sometimes...
CPU-Z shows that my Phenom II X4 955 has revision RB-C3. I'm aware that the spec doesn't officially support 1600 DDR though but as it worked with 8GB I was surprised to find it doesn't with 16GB. I'll look at tweaking the voltages to see if I can get it to work though, thanks.
#15
Posted 08 October 2012 - 04:39 AM
doveman, on 08 October 2012 - 04:01 AM, said:
Yep, and I have understood.
The point I was trying to make is slightly different, more around the Occam's Razor approach:
http://en.wikipedia.org/wiki/Occam's_razor
With a "good" set of chips (8 Gb in total) does the machine work OK?
If yes, then do you have any benefit of any kind that you can actually measure/feel with adding to it the other "spare" 8?
If yes, then do it.
If not, keep the spare ones as "spare" (or give away them, or keep them for another machine you plan to buy).
It seems to me like there is NO actual difference (or at least one that you can actually perceive without using a benchmark) between:
- 8 Gb @1333 <- stable and working OK
- 8 Gb @1600 <- stable and working OK
but we know that:
- 16 Gb @1333 <- possibly stable and working OK but largely unused
- 16 Gb @1600 <- found UNstable and NOT working OK besides being largely unused
If you prefer, you have proven that 16 Gb@1600 is NOT good on that motherboard and - at the moment - found that by downclocking the RAM the machine works OK with 16 Gb @1333, but what actual practial advnatages do you have with 16GB@1333 over running 8Gb@1600?
If, as I presume, the answer is "None", the only use of that additional 8 Gb is to be able to tell your friends how powerful is your machine with 16 Gb and have a slightly increased power consumption and have the "spare" be worn out a bit.
jaclaz
#16
Posted 08 October 2012 - 06:18 AM
jaclaz, on 08 October 2012 - 04:39 AM, said:
With a "good" set of chips (8 Gb in total) does the machine work OK?
If yes, then do you have any benefit of any kind that you can actually measure/feel with adding to it the other "spare" 8?
If yes, then do it.
If not, keep the spare ones as "spare" (or give away them, or keep them for another machine you plan to buy).
It seems to me like there is NO actual difference (or at least one that you can actually perceive without using a benchmark) between:
- 8 Gb @1333 <- stable and working OK
- 8 Gb @1600 <- stable and working OK
but we know that:
- 16 Gb @1333 <- possibly stable and working OK but largely unused
- 16 Gb @1600 <- found UNstable and NOT working OK besides being largely unused
If you prefer, you have proven that 16 Gb@1600 is NOT good on that motherboard and - at the moment - found that by downclocking the RAM the machine works OK with 16 Gb @1333, but what actual practial advnatages do you have with 16GB@1333 over running 8Gb@1600?
If, as I presume, the answer is "None", the only use of that additional 8 Gb is to be able to tell your friends how powerful is your machine with 16 Gb and have a slightly increased power consumption and have the "spare" be worn out a bit.
jaclaz
Well as we've seen, there's so little difference between running @1600 or @1300 that we can ignore that as a consideration. For the benefit of anyone else who reads this in future, it seems worth clarifying the relatively small contribution memory performance makes to overall performance, so even a 6.91% increase in read speeds won't make much practical difference. It also seems unlikely that 16GB @1333, which is the proper supported "non-overclocked" spec of the chipset will be unstable and I'm not worried about wearing them out a bit as they have a limited lifetime warranty, so let's just concentrate on any possible benefits I might find with 16GB.
To expand on what I was saying before about certain games, DCS World has a problem on my system (and others apparently) of stuttering, which is probably caused by not being able to stream the data from HDD quickly enough. The game folder is approx 9.45GB, so by creating a RAMdisk to hold these files and running it from there, I can see if that is indeed the problem and whether it's worth spending £115 on a 256GB SSD (if I get one, I want one big enough to run Windows and several games from). I may in fact decide to just stick with the RAMdisk and the extra 8GB cost me about £35, so I'll avoid spending more money on a SSD if I decide to do that.
Another game that tends to struggle with streaming data fast enough is ArmAII. That's a bit tricker as the Operation Arrowhead data folders are about 7.7GB whilst the ArmAII data folders are about 15GB, so 22-23GB for Combined Ops, which is obviously not practical to put on a RAMdisk (and that's not even including the mods, not all of which necessarily will benefit from running from RAMdisk, but some, like custom islands/terrain (or at least some of their files) of which I have 7.8GB, probably would. However, other people have done lots of test to determine which files get used the most and put only those on a RAMdisk with good results, so I plan to try that sometime and if that works, again it will make it less likely that I buy a SSD.
I also use a 512MB RAMdisk currently for Mediaportal, so that when watching TV it creates the timeshifting files on that rather than constantly writing to the HDD, helping to keep noise/power down and reduce wear and tear on the HDD. I also put some temp files on there but it's impractical to use for all temp files as when installing games it might create several GB of temp files and when downloading with IE it creates a temp file before moving it to the final destination, so if downloading a 4GB file it would take 4GB on the RAMdisk. I actually use Opera or Iron rather than IE most of the time, which don't have this problem and just create a partial file in the actual download folder but for someone who wants to use IE and use a RAMdisk to store both the IE and system temp files, being able to use around 4GB for the RAMdisk would be useful and that wouldn't be practical with only 8GB. For my 512MB RAMdisk I don't really need more than 8GB though. That gives me approx 10mins timeshifting for SD TV and 4mins for HD and I don't currently receive HDTV so I find that plenty but other users of Mediaportal find it far too small and would use at least 2GB for the RAMdisk.
Apart from the RAMdisk, as I mentioned DCS World, being 64-bit, may benefit from more than 8GB and there's a 64-bit version of X-Plane 10 due soon, which it's been suggested will also use as much RAM as is available to increase performance and I believe ArmA3 will also be 64-bit and give better performance with more RAM, although all these need testing to confirm if there's any benefit and if so, how significant.
Now, I'm building a PC for my brother and I decided I'd put 8GB in that to somewhat future-proof it and was going to buy that 8GB when prices come down, as they're predicted to do soon. However, if I determine that there's no point having 16GB in my PC and I'd be better off buying a 256GB SSD for £115 than trying to use £35 worth of 8GB RAM for a RAMDisk, then I'll just put the extra 8GB in my brother's PC, which is one of the reasons why I wasn't too concerned about having to buy it to use my PC whilst I was waiting for the original kit to be replaced. If I didn't have the option of putting it in his PC if it turned out the running games from RAMdisk idea didn't work out for me, I probably would have been more hesitant to buy it. Then again, if the RAMdisk doesn't make my games work better, I can't be sure that a SSD will either (although many people have told me that both DCS World and ArmaII perform a lot better since they got a SSD to run them from).
#17
Posted 08 October 2012 - 07:30 AM
doveman, on 08 October 2012 - 06:18 AM, said:
And that's just a memory-intensive benchmark, which is very unlike most software people will use. For example, since you mention games, here's the real-life performance gains you can expect:

The biggest improvement, from DDR3 2133 CL9 vs DDR3 1333 CL9 in Metro 2033 is 1% gained. In your case it would be far less. There's essentially zero gains to be made here.
But 16GB vs 8GB is a decent boost for many. If you use any memory hungry apps (vmware workstation, most CAD apps, photoshop on large images, etc) and especially if you heavily multitask, and keep in mind things like superfetch still benefit from it quite a lot too. And sure, why not ramdisks as well...
#18
Posted 08 October 2012 - 12:57 PM
and created a 10GB RAMdisk, copied the DCS World folder to it and updated the shortcuts and registry to point to the RAMdisk (S:\Games\DCS World).
I'm not sure if it loads any quicker, maybe a bit but the stuttering/freezing is even worse, which doesn't seem to make much sense. In Cold Start in Mozdok, if I rotate the view at the start in the cockpit by holding down Num4, it sticks at one point (the point varies) every one or two rotations and when flying along it keeps freezing constantly.
I thought when running from HDD this was being caused by not being able to stream the data from the HDD quickly enough but this obviously shouldn't be an issue with the data on the RAMdrive. I've still got about 3GB free even with the 10GB RAMdrive and Black Shark loaded, so it's hasn't run out of RAM.
This is what my RAmdisk Enterprise settings are (most of them are default as I don't understand them!)

On the Image File Processing sub-page, I've just pointed it to F:\DCSW.img, clicked Save RamDisk content and ticked Load RAMdisk content from Image at Driver Start, although I've unticked this now as it made Windows take about 10mins to start! I prefer DataRAM RAMdisk as it lets you easily load and save different images. stop/start the RAMdisk but that's limited to 4GB RAMdisks so I can't use that for DCSW.
With my browser open, which does have 34 tabs open at the moment, there seems to be a lot more HDD activity than before I installed the RAMdisk, even though I've still got 3GB free so it shouldn't be needing to use the swapfile any more than before.
I've changed the RAMdisk from FAT32 to NTFS now so I'll see if that's helped but I'd like to find a program to benchmark it first. HDTunePro recognises my 512MB DataRAM RAMdisk and let's me bench that but the 10GB RAMdisk Enterprise drive isn't listed at all. Also, is there a program to test the RAM in Windows? I've done a memtest from boot which was fine but as we've already seen, Windows can be a bit particular about RAM settings, so I'd like to check that it's all working fine in Windows as well.
#19
Posted 08 October 2012 - 01:38 PM
doveman, on 08 October 2012 - 12:57 PM, said:
Casually (today's topic):
http://reboot.pro/17610/
jaclaz
This post has been edited by jaclaz: 08 October 2012 - 01:38 PM
#20
Posted 08 October 2012 - 03:04 PM
jaclaz, on 08 October 2012 - 01:38 PM, said:
http://reboot.pro/17610/
Thanks, there's a co-incidence
So I tested with ATTO then CDM and had rather disappointing results compared to Raymond's http://www.raymond.c...nd-write-speed/
This is my 10GB NTFS RAMDisk Enterprise disk in ATTO:

and this is my 512MB FAT32 DataRAM RAMdisk:

This is the 10GB NTFS RAMdisk Enterprise disk in CDM:

and this is the 512MB FAT32 DataRAM RAMdisk:

and lastly I made a 512MB FAT disk with Bond Disc (which did very well in Raymond's tests):

As you can see, the Seq and 512k results are about half what Raymond got and only on the 4K and 4K QD32 figures do I match, or in some cases exceed his results. In fact, on the 4K QD32 Write I exceeded his best result (324MB/s) quite considerably but that's not much consolation when I want to use a RAMdisk for loading games (i.e. only reading) and I'm getting such poor Seq and 512k results.
I also tried the SoftPerfect RAMdisk and was able to create a 10GB NTFS RAMdisk but whilst copying the DCSW files to it, it said it was unable to continue as the files were corrupt. Checking the RAMdisk showed that that the DCSW folder was now inaccessible or and non delete-able and so I had to unmount the RAMdisk and re-mount it (now empty) to clear that problem. Trying again gave the same result so I gave up on that software.



Help


Back to top










