[doveman]

It seems to me that Softperfect is probably trying to use part of the RAM it shouldn't as although I can copy the files to it OK as FAT32, they then don't work properly (game crashes on loading) and if I create an image and try to mount it, it BSOD my PC. As you can see from the RAMdisk Enterprise screenshot in my first post, the default settings make it avoid using some of the RAM, so perhaps that's what Softperfect needs to do.

Gavotte_RAMDisk_1.0.4096.5_200811130 looked promising but I can't seem to get it working under W7 x64 as it just showed error code 19 in Device Manager (missing registry entries apparently). It comes with a 64bit driver and ram4g.reg which contains

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\RRamdisk\Parameters]
"UsePAE"=dword:00000001

which I merged and there are other settings in that key. Has anyone else got this working under W7 x64?

[jaclaz]

Why don't you try IMDISK, which is:

• supported
  
• continuously updated

?

BTW, the results of the test with the Softperfect one seem to me like (possibly ) indicating that your RAM isn't really-really "stable" .

The gavotte .reg is ONLY for 32 bit (to add PAE access).
Still, at least on 32 bit XP, you simply install it running ramdisk.exe, most probably on 7 you will need every trick of the trade +1 (Administrator/UAC/driversigning).


jaclaz

This post has been edited by jaclaz: 09 October 2012 - 08:20 AM

[doveman]

jaclaz, on 09 October 2012 - 08:03 AM, said:

Why don't you try IMDISK, which is:

• supported
  
• continuously updated

?

BTW, the results of the test with the Softperfect one seem to me like (possibly ) indicating that your RAM isn't really-really "stable" .


jaclaz

Thanks, I overlooked that so I'll give it a go.

I don't think my RAM's unstable or else I'd have expected problems with RAMdisk Enterprise. Seems more likely that SoftPerfect just needs some work. If you know of some software that can test my RAM in Windows (as opposed to outside, which I've already done with memtest) I'll do that though.

[jaclaz]

"pure" RAM testing?
http://hcidesign.com/memtest/

jaclaz

[doveman]

jaclaz, on 09 October 2012 - 08:40 AM, said:

"pure" RAM testing?
http://hcidesign.com/memtest/

jaclaz

Thanks, that should do the trick

I've just tested playing DCS World from an IMDisk and that worked fine, although to be honest I didn't notice it made much difference to loading times and I still get the occasional stutter, so that's obviously not being caused by the HDD being unable to stream the data fast enough and it would be a waste of money buying an SSD to run it from. I double-checked straight after by loading it from my 3.5" HDD and the experience was much the same.

I did a CDM on the IMDisk, my 2TB Samsung drive, which contains W7, the swapfile and all my data (plus my games) and my 2.5" 500GB Samsung drive, which I thought I might use just for games to keep them separate from my WIndows/swapfile drive. Clearly the 2.5" drive is a lot slower than the 3.5" one and they're both massively slower than the IMDisk but as I say, the IMDisk didn't seem to benefit DCSW over the 3.5", so it seems the latter provides as much bandwidth as DCSW can use, at the moment at least. Last time I tried it from the 2.5" it wasn't great but there's daily updates coming out for DCSW at the moment and I haven't tried the latest one on the 2.5" drive, so I need to do that to see if it's fixed the problem.

[jaclaz]

Still,it remains a mistery (at least to me) why all your Crystal Diskmark benchmark reports show roughly half the speed Raymond got .

jaclaz

[doveman]

jaclaz, on 09 October 2012 - 09:56 AM, said:

Still,it remains a mistery (at least to me) why all your Crystal Diskmark benchmark reports show roughly half the speed Raymond got .

jaclaz

Well I've been told it's because he's probably running Intel and I'm running AMD, which has a crappy memory controller That sounds like it could be right actually, as I've seen RAM benchmarks on Intel systems that are much higher than I could ever hope to achieve with my system.

Hey, 2965MB/s is still 25x faster than my HDD and 5.93x faster than a SSD and that's just the Seq speeds Sucks that it doesn't make much difference to loading/playing DCSW though. I'll have to test some other games, including ArmAII, when I get a chance.

[doveman]

Just wanted to mention that the author of SoftPerfect tells me he's identified the problem and is looking into it, so hopefully that will be an option in the near future.

IMDisk is OK and I tested ArmAIIOA running from it yesterday and it did load a lot quicker (unlike DCS World) although I haven't had time to test whether it makes gameplay/streaming smoother. The only issues I have with IMDisk is that it seems to take a lot longer to load images than the other programs I tested (it doesn't show a progress bar either, making it hard to know how long it's going to take) and that the 4k and 4k QD32 scores were quite a bit slower than some of the other RAMdisks, although I don't know if those or the Seq and 512k speeds are the most important for my games. It's great being able to easily load/save images with IMdisk though, unlike RAMdisk Enterprise which also didn't appear to provide any way to unmount the RAMdisk to release the memory, whcih IMDisk does

[jaclaz]

doveman, on 10 October 2012 - 07:02 AM, said:

Just wanted to mention that the author of SoftPerfect tells me he's identified the problem and is looking into it, so hopefully that will be an option in the near future.

IMDisk is OK and I tested ArmAIIOA running from it yesterday and it did load a lot quicker (unlike DCS World) although I haven't had time to test whether it makes gameplay/streaming smoother. The only issues I have with IMDisk is that it seems to take a lot longer to load images than the other programs I tested (it doesn't show a progress bar either, making it hard to know how long it's going to take) and that the 4k and 4k QD32 scores were quite a bit slower than some of the other RAMdisks, although I don't know if those or the Seq and 512k speeds are the most important for my games. It's great being able to easily load/save images with IMdisk though, unlike RAMdisk Enterprise which also didn't appear to provide any way to unmount the RAMdisk to release the memory, whcih IMDisk does

You can try with IMDISK to load a small image, extend it and then copy to it the files, it may be quicker
A 2x factor between Intel based and AMD "sounds" still a lot, possibly some of the guys with more specific hardware experience might want to comment.

jaclaz

P.S.: Also, since your disk is biggish, you should try awealloc, compare with the reply that Olof in the meantime posted on the previously referenced thread:
http://reboot.pro/17610/

This post has been edited by jaclaz: 10 October 2012 - 07:50 AM

[doveman]

jaclaz, on 10 October 2012 - 07:44 AM, said:

You can try with IMDISK to load a small image, extend it and then copy to it the files, it may be quicker :unsure.

Yeah, I could but that's too much hassle. I just want to load an image and play

Quote

A 2x factor between Intel based and AMD "sounds" still a lot, possibly some of the guys with more specific hardware experience might want to comment

It does but when you look at the RAM benchmark Raymond's just given me, which is pretty typical for Intel RAM benchmarks I've seen, it's more understandable

Copy 21168 MB/s
Read 19418 MB/s
Write 18784 MB/s
Score 18.10 GB/s
Latency 52.4 ns

compared to my

Copy 10466 MB/s
Read 8107 MB/s
Write 6831 MB/s
Score 7.47GB/s
Latency 66.9 ns

Quote

P.S.: Also, since your disk is biggish, you should try awealloc, compare with the reply that Olof in the meantime posted on the previously referenced thread:
http://reboot.pro/17610/

I'm not sure that would help. It seems that's mostly for use on 32-bit systems to be able to use the RAM above 4GB and both the thread you reference and this one suggest that awealloc will just make things slower Then again, that thread suggests that the method used to allocate RAM for IMDisk could allow it to then be paged out to the swapfile back on the HDD whereas the method used for awealloc prevents this, so maybe it's better to use that even if it is slower.

[doveman]

I tried making a 400MB RAMdisk using awealloc - imdisk -a -f \\.\awealloc -s 400M -m S: -p "/fs:ntfs /q /y"

As you can see from the CDM results, it's much slower:

[jaclaz]

doveman, on 10 October 2012 - 10:26 AM, said:

I tried making a 400MB RAMdisk using awealloc - imdisk -a -f \\.\awealloc -s 400M -m S: -p "/fs:ntfs /q /y"

As you can see from the CDM results, it's much slower:

This makes little sense.
400 Mb is NOT a "biggish"disk size, you were talking till now of a several Gb one.

IMHO what you should try is making a disk of the size you need it using IMDISK "native" allocation and test it in both ATTO and CDM, then make another disk of the same size using awealloc and test it with both ATTO and CDM, then disregard BOTH results of BOTH benchmarks and simply try running that game with one and the other method and choose the one that "feels" better (if a difference can be felt) pr that results more stable.

I presume that if the Author of the tool says:

Quote

The various memory access/allocation methods are good for different things. It is quite expected that measured performance will differ a lot. But my general opinion is that for memory backed virtual disks, there is no good reason to use internal ImDisk allocation for disks larger than, say 100-200 MB or so. Awealloc should really be used instead.

it is very likely that there are reasons...

jaclaz

This post has been edited by jaclaz: 10 October 2012 - 11:45 AM

[doveman]

jaclaz, on 10 October 2012 - 11:44 AM, said:

doveman, on 10 October 2012 - 10:26 AM, said:

I tried making a 400MB RAMdisk using awealloc - imdisk -a -f \\.\awealloc -s 400M -m S: -p "/fs:ntfs /q /y"

As you can see from the CDM results, it's much slower:

This makes little sense.
400 Mb is NOT a "biggish"disk size, you were talking till now of a several Gb one.

IMHO what you should try is making a disk of the size you need it using IMDISK "native" allocation and test it in both ATTO and CDM, then make another disk of the same size using awealloc and test it with both ATTO and CDM, then disregard BOTH results of BOTH benchmarks and simply try running that game with one and the other method and choose the one that "feels" better (if a difference can be felt) pr that results more stable.

Well Raymond's tests were done with a 512MB RAMdisk and he got good speeds, so it seemed reasonable. Just for you , I've done the CDM and ATTO tests with a 10GB awealloc RAMdisk and as I expected, it's just as slow (ATTO and CDM results for 10GB non-awealloc RAMdisk are in posts #20 (RAMdisk Enterprise) and #25 (IMDisk, only CDM but sufficient to show it's slower than RAMDisk Enterprise and faster than awealloc))

It's still many times faster than HDD and three times faster than SSD though, so plenty fast enough to rule out slow data streaming as being the cause of any stuttering. I've now tested DCS World from HDD, IMDisk and awealloc and I get the same stuttering with all three, so obviously that's not the problem.

Quote

I presume that if the Author of the tool says:

Quote

The various memory access/allocation methods are good for different things. It is quite expected that measured performance will differ a lot. But my general opinion is that for memory backed virtual disks, there is no good reason to use internal ImDisk allocation for disks larger than, say 100-200 MB or so. Awealloc should really be used instead.

it is very likely that there are reasons...

Well yes, as I said IMDisks can end up paged to the swapfile on the HDD, which rather defeats the purpose of a RAMdisk, so that's a good reason

[doveman]

There's a new version of SoftPerfect RAMDisk available now which fixes the problems I was having before.

Speeds look good. FAT32 is a bit faster than NTFS but I think that's normal because of NTFS' overheads. I just need to confirm that it prevents any of the RAM used being paged out, like awealloc.

[doveman]

Well running ArmAII from my 11000MB RAMDisk was working fine but now it's started randomly locking up the entire PC, requiring a hard reset.

It happened with SoftPerfect, which I've been using for ages but I've moved away from that now as a) it annoyed me that it auto-mounted the last mounted RAMDisk when starting the app after a reboot and b) it started throwing up errors and failing to mount RAMdisks anyway.

So I'm using Primo RAMdisk at the moment but I ran through one of the benchmark scenarios in ArmAII a few times, until the last time when it locked up the PC again. After that I ran from HDD and it didn't lock up at all.

I could try IMdisk awealloc again, I'm not sure if I've tried that since experiencing these lockups. It's a lot slower than the other apps but still twice as fast as a SSD.

Note I'm still running at the BIOS defaults I think, certainly CPU-Z shows NB 2000MHZ, DRAM Freq 666.7Mhz, 9-9-9-24-34-2T. The voltage is probably at default 1.5v as well. Might I need to boost that slightly for 16GB and if so, what should I boost it to or should I loosen the timings instead?

This post has been edited by doveman: 25 November 2012 - 06:10 AM

[doveman]

I ran Memtest overnight and that completed two passes no errors.

Any ideas why using the RAMdisk is causing my PC to lockup?