PS Scratch Disk size optimization / allocation

I'm considering the use of an SSD (Solid State Disk) as a Photoshop 1st scratch disk and need some specific info about how scratch disks are used. Here are the specifics:

PC running XP has 4G of RAM.
Photoshop can only use 2G.
Memory slider must be set at around 75% - 80%
So that means only 1.5G or so of RAM available for image operations. But I regularly work on images files which start out at 500M or higher before any layers are added. So I very quickly end up with heavy scratch disk utilization.

I can purchase a solid state disk (SSD) which is an adapter card filled with memory chips but which appears to the OS as a hard drive. Its advantage over a traditional hard disk is extremely fast access times (~0.6us) and data transfer rates (80-100MB/s) and no need to defragment.

I would use the SSD as the first scratch disk and then use a traditional and very large hard drive as the second scratch disk. So I'd like to know how the scratch disk space is allocated.

For example, if I purchased a 4GB SSD, and made that the 1st scratch disk, would Photoshop create a 4GB scratch file there first and then use the hard disk if it needed still more space? A better understanding of how this 2 scratch disk scenario works will help me pick the right size SSD.

Thanks in advance.

The few SSDs we've tested haven't been any faster than 2 IDE drives setup as a RAID array. But we haven't tested too many.

Yes, it works as you described.

Thanks Chris. Can you say anything about the Cenatek Rocket Drive?

Nope.

O.K. Thanks much.

:

I'm considering the use of an SSD (Solid State Disk) as a Photoshop 1st scratch disk and need some specific info about how scratch disks are used. Here are the specifics:

PC running XP has 4G of RAM.
Photoshop can only use 2G.
Memory slider must be set at around 75% - 80%
So that means only 1.5G or so of RAM available for image operations. But I regularly work on images files which start out at 500M or higher before any layers are added. So I very quickly end up with heavy scratch disk utilization.

I can purchase a solid state disk (SSD) which is an adapter card filled with memory chips but which appears to the OS as a hard drive. Its advantage over a traditional hard disk is extremely fast access times (~0.6us) and data transfer rates (80-100MB/s) and no need to defragment.

As Chris notes, you'll see very little advantage over a good RAID array, which are pretty cheap now that SATA has 10K drives.

The reason is that we have reached saturation on the PCI bus. The theoretical bandwidth of 120MB is shared between ALL your PCI devices, and "chatter" and inneficiencies on the bus means that theoretical limit is seldom reached.

A RAM drive over PCI, like the RocketDrive, is limited by this. And it's bandwidth would be shared by your traditional hard drives.

There are a few solutions that alleviate slightly this already, such as devoted pathways to gigabit network adaptors on some Intel and the new NVIDIA NF3-250 motherboards, but real relief won't come until PCI-EX is available (and a SSD made for it).

A RAMDisc using some of your system memory (say, the top 1GB) as that 1st scratch disc might help. Cenatek has a decent one you could experiment with.

<snip>

Thanks Jerry,

Good point about the bus. But in my situation, there are two PCI-X busses on the workstation. I don't know much about PC architecture but it seems to me that if the SSD has performance parameters like 6 MICRO-sec access times and higher transfer speed than disks, and, if it's placed on a separate bus, this has got to be a large improvement (say double performance) over even RAID, yes? Not to mention no issues with fragmentation, etc.