• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Response to ALL-RAM machine article

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Captain Slug

Helpful Senior Member
Joined
May 23, 2001
Location
Asteroid B-612
In response to this article - http://www.overclockers.com/tips00621/

I have a problem with one of the concepts

- If you're transferring data from a hard drive to a RAMdrive card, it's still going to be bottlenecked by the speed of the hard drive when it needs to cached between the two.
- RAM is dynamic, but ONLY if it's power is cutoff. If the RAM could simply take advantage of being powered from the 5v standby it could keep the data active as long as the machine is plugged in and the card is booted.

Perhaps the RAMdrive could have it's own secondary on-card storage. For example a 2.5" notebook-sized hard drive (or even a microdrive) would be in the right power consumption range for such an application. This drive would only need to be utilized to pre-cache the RAM. If the card were independantly powered and self-booting through the 5V standby, it could pre-cache the RAM from the drive as soon as the machine receives power as it is pluggged in.

Cut out the middle man and you have all the problems solved. What is really needed is not complication, but independence of the RAMdrive itself so that it can take better advantage of the timing that's inherent in normal operation.

Thoughts?
 
I was reading about ramdrives a year or three ago. The ones I saw had a battery backup, and I think they were connected via IDE.
 
The problem they are having is the lack of bandwith... they are bottlenecked at the current PCI standards... hence why they use PC133 memory and don't put ddr memory in these "drives". I think we would need to see a motherboard that could support 3 or 4 PCI-E x16 slots before the RAM Drives can be fully utilized...

Slug - They want RAM Drives to do exactly what you're thinking. My concept of the RAM Drive was that the hard drive would probably be there so the RAM Drive would load the OS from the hard drive and that's about it. Think of it as a "Ghost" procedure... At start up, the hard drive clones itself to the RAM Drive, and then the RAM Drive clones itself to the hard drive on shutdown.
 
dark_15 said:
The problem they are having is the lack of bandwith... they are bottlenecked at the current PCI standards... hence why they use PC133 memory and don't put ddr memory in these "drives". I think we would need to see a motherboard that could support 3 or 4 PCI-E x16 slots before the RAM Drives can be fully utilized...
That's the part of the article I agree with. Any other means (IDE or SATA) simply wouldn't be able to take advantage of the RAM's speed.

dark_15 said:
Slug - They want RAM Drives to do exactly what you're thinking. My concept of the RAM Drive was that the hard drive would probably be there so the RAM Drive would load the OS from the hard drive and that's about it. Think of it as a "Ghost" procedure... At start up, the hard drive clones itself to the RAM Drive, and then the RAM Drive clones itself to the hard drive on shutdown.

The ghosting procedure between a RAMdrive and hard-drive done through an IDE or SATA interface would have to take place after the motherboard has booted and after the regular hard drive is powered on with 12v and this would add considerable time to the boot sequence. My solution is to make the card inependently powered on just 5V-standby so that the Ghosting procedure could be done at the same time as the system post. This would prevent any delays during the booting of the machine after it has been unplugged.
 
Any way you slice it... it's going to take a while to boot up... let's face it... permanent storage is just slow...

As for me... I could handle a 5-10 minute boot if it meant my apps and OS would work 100x faster... Just think of how fast Video would encode with a Pure Ram setup...

Stedeman - That sounds like a good possibilty... why isn't it being implemented? Could cost (price/performance ratio) be a factor in PCI-X 2.0???
 
I'm not to sure why. It looks like an easy upgrade to the regular PCI-X standard so it will probably trickle down from the server market in a few years.
 
In the good old bad days when I used to have a Mac, my favorite gamming trick was to make a RAM drive (easy to do with MAC OS) and copy whatever game to it and blast away. It makes a BIG difference!
so even if you dont boot from it, just having the application in memory frees up the regular drive to read and write only for the OS. Right there is performance enhancement.

All you need to do is press Apple key S errr, Control key S often!
 
With folding and seti membership growing, the number of people that leave there computers on 24/7 also grows. I'd say you turn it on, wait for it too boot and then leave it on. That's what I do with mine. It's on for months at a time, usually without a reboot.
Rob
 
If the card were secondarily powered by 5v standby you wouldn't see a long 5-10 minute boot unless you had JUST plugged the machine in. So in normal desktop situations you wouldn't be waiting any longer period of time to get the machine up and running on a daily basis.
And yes, I'm in the 24/7 group.
 
The autor of the article is ill-informed about what's on the market. Solid state drives are coming and PCI or PCI-e are not needed. Have a look here.
 
What really needs to appear is magnetic ram that is non-volatile. It will happen someday. the question is when? It would be awesome to see a system with 4-8gigs of magnetic ram, then another 80+ gigs for mass data storage. Then combine that with a 32 megabyte SRAM cache on the processor with a built in memory controller, maybe dual-cores, and and a powerful GPU. That would be a next generation system. It might even be the architecture that holds us over till quantum computing hits mainstream in 30 years or whatever.
 
woutertal said:
The autor of the article is ill-informed about what's on the market. Solid state drives are coming and PCI or PCI-e are not needed. Have a look here.

The fastest hard disks today with ATA133 Buses provide a sustained data rate between 20 and 55 MB per second depending where the data is on the mechanical rotating platter. This figure is 93 MB per second for the HyperDrive III with an ATA100 Bus. The seek time for the HyperDrive III is at the silicon level (less than 100 microseconds) rather than the mechanical magnetic level (8 milliseconds). So the read and write times for the HyperDrive III are over 80x faster than a Hard Disk.

XP installs on a HyperDrive III in around 4 minutes, rather than the 40 minutes that it takes to install on a Hard Disk.

That's useful... but it still is a bottleneck. Data stored in the RAM transfers data from itself to the processor in gigabytes per second... not megabytes per second. If you were to give plenty of bandwith required for a pure RAM Drive, Windows XP will install in SECONDS (and that's on a very slow day)... not 4 minutes as stated on the site. And with that setup... you would be able to encode video and do amazing calculations in a few seconds... then the CPU, not the Hard Drive, would be the bottleneck...
 
Thats why you buy an Opteron box and build your ramdisk out of system memory.

The TYAN-K8QS will take 32gb of memory. That'll be more than enough to fit your games into, and it'll be alot faster than some PCI or PCI-e bottlenecked device.
 
XWRed1 said:
Thats why you buy an Opteron box and build your ramdisk out of system memory.

The TYAN-K8QS will take 32gb of memory. That'll be more than enough to fit your games into, and it'll be alot faster than some PCI or PCI-e bottlenecked device.
*takes out a $10,000 loan to build a new machine
 
Captain Slug said:
*takes out a $10,000 loan to build a new machine
$10K will be a few short for just the RAM! The RAM is $10,320 (16X 2GB REGISTERED ECC PC2700 $645.00 each) + the Mobo is $1,722. Don't forget this sucker is going to need a VERY powerful PSU and oh yeah, 4 CPU's!!! I would say it's more like refinancing your home...I wonder if the bank would take a kidney for one.:eh?:
 
And with that setup... you would be able to encode video and do amazing calculations in a few seconds...

No. The write-to-disk speed isn't the primary factor, the CPU speed is. Video encoding requires a *lot* of math, which is something only the CPU can do. The amount of work required means that, unless you're using four Opterons, odds are the write-to-disk speed is not the bottleneck. On my rig, it takes roughly 10 hours to encode an hour-long MPEG-2 video (Constant Bit Rate @ 8000 Kbit/sec). The resulting file is about 1900 MB. 1900MB/10 hours = 190MB/hr = 3.16MB/min = .0527MB/sec of data being output by the CPU. Last time I checked, the sustained transfer rate of ATA-100 was greater than 1/20 of a MB per second.

If you're going to spend exorbitant amounts of money to improve video encoding speeds, spend it on the CPU first.
 
dark_15 said:
As for me... I could handle a 5-10 minute boot if it meant my apps and OS would work 100x faster...
If I never installed anything or changed configurations, I would probably reboot about once a month.

However, I tend to go spurts of a few weeks at a time, then I will mess with some driver or other, and I will reboot 7 to 10 times in an hour, Or, I'll get a new CPU or new RAM, and I'll reboot 50 times in a day checking stability and changing BIOS settings. That would not be fun with 5-10 minute boots.

As Captain Slug says, you wouldn't need to wait 5-10 minutes if the RAMdrive is powered by 5V standby. But even if it is, a power failure or a power cord disconnect wipe out the drive anyway.


No one has yet mentioned the risk RAMdrives pose to your data. If your computer powers down normally, everything copies back to your hard drive normally. If you lose power say goodbye to whatever was on your RAMdrive, and say goodbye ot it for good of it wasn't also saved on non-volatile media.
 
Back