NAND-based Flash drives hit the market

[Vatroslav]

As the awaiting for SSD flash drives is coming to an end, Fusion has just announced their even more innovative ioMemory storage solution. What may turn as a revolutionary architecture eliminates the wait time between memory and storage, achieving access rates of 700 MB/s (read) and 600 MB/s (write) and thereby improving the overall speeds by almost a thousand times.

PCIe 4x card-based, being able to serve as either local storage or storage cache, and with the longevity of 8 years and error-correction technology, it seems like the end days of the spinning disc disks are nearing.

Starting at $2400 and 80 GB of storage space (640 GB is the maximum announced capacity), they are pricey, but from what it seems, well worth the investment for realtime streaming-based computing.

I'd like to know what some of the tech guys here think about this.

http://www.fusionio.com/faq.html

[Pietro]

Incredible stuff! Seems like an ultimate solution indeed. That would need a better CPU for streaming this amount of data, I guess, but then 32bit system/sequencer problem is puff - gone.

Microtechnology is cheaper and cheaper every day. I bought 2x1GB of 800MHz RAM for my PC in August 2006, today for the price, I could get even 4 times that (4x2GB 800MHz).

If the price for 320Gb drives goes down to $1000 - I'm in. I don't know how, but I'm in

- Piotr

[paulwr]

Originally Posted by Pietro

Incredible stuff! Seems like an ultimate solution indeed. That would need a better CPU for streaming this amount of data, I guess, but then 32bit system/sequencer problem is puff - gone.

Microtechnology is cheaper and cheaper every day. I bought 2x1GB of 800MHz RAM for my PC in August 2006, today for the price, I could get even 4 times that (4x2GB 800MHz).

If the price for 320Gb drives goes down to $1000 - I'm in. I don't know how, but I'm in

- Piotr

The guys over at VisionDAW last spring or summer did some experimenting with 5 gig or so of a transistor drive thing, the point being to find out how much a DAW could be pushed if streaming samples wasn't limited by hard drive performance. I wish I could remember where the choke exactly was, but it choked way sooner than anyone had hoped. It seems like we are on the brink of so much, but the single super system so many are dreaming of is still a ways off I think.

So much software and hardware must climb up to a similar high level to make it feaseable. Who will be the brave one to get 80GB for $2400 and report back? Perhaps a dual 4-core Xeon system on a beefy server MOBO and loads of ram will help, if the software can take advantage of the arrangement. And maybe the fact that this new ioMemory sits on a pci-e slot will make a difference, too. I'm all ears!

-Paul

[luima]

I am not a real expert in hardware but I know that there is bandwidth limit in the PCI bus even in the PCI-Express, so this slots never will reach neither approach the speed of RAM memory directly attached to the Processor bus.

[Vatroslav]

Originally Posted by Pietro

Incredible stuff! Seems like an ultimate solution indeed. That would need a better CPU for streaming this amount of data, I guess, but then 32bit system/sequencer problem is puff - gone.

Well, actually, an SSD such as this would only eliminate the need for multiple drives to stream from. The 4 GB limit remains in a x32 enviroment. The key here is a single drive system (preferbaly a x64 one) doing better than any multiple drive one today.

The biggest downside compared to the mechanical disks which understandibly fails to be mentioned on the site are the somewhat slower random write speeds, but I don't think it should make too much difference in the sequencial streming work.

Originally Posted by paulwr

The guys over at VisionDAW last spring or summer did some experimenting with 5 gig or so of a transistor drive thing, the point being to find out how much a DAW could be pushed if streaming samples wasn't limited by hard drive performance. I wish I could remember where the choke exactly was, but it choked way sooner than anyone had hoped. It seems like we are on the brink of so much, but the single super system so many are dreaming of is still a ways off I think.

So much software and hardware must climb up to a similar high level to make it feaseable. Who will be the brave one to get 80GB for $2400 and report back? Perhaps a dual 4-core Xeon system on a beefy server MOBO and loads of ram will help, if the software can take advantage of the arrangement. And maybe the fact that this new ioMemory sits on a pci-e slot will make a difference, too. I'm all ears!

-Paul

Since it was 5 GBs, I'm guessing it was a 64-bit system? Which DAW was it?

Do you remember any specs of the experiment? Because it might have been an overload which is hardly to ever happen in any production.

[Vatroslav]

Originally Posted by luima

I am not a real expert in hardware but I know that there is bandwidth limit in the PCI bus even in the PCI-Express, so this slots never will reach neither approach the speed of RAM memory directly attached to the Processor bus.

The PCI-E x4 max bandwith is 1000 MB/s, so since the total bandwidth is not shared between the devices, I don't think it will create any problems to the first generation ioMemories.

[Pietro]

Originally Posted by Vatroslav

Well, actually, an SSD such as this would only eliminate the need for multiple drives to stream from. The 4 GB limit remains in a x32 enviroment. The key here is a single drive system (preferbaly a x64 one) doing better than any multiple drive one today.

Yes, I ment, if you had such drive at your hands, you could probably have VERY low pre-load buffer settings for your software instruments, thus being able to load much more not even getting close to the 32bit memory treshold.

Hopefully, someday.

- Piotr

[Vatroslav]

Originally Posted by Pietro

Yes, I ment, if you had such drive at your hands, you could probably have VERY low pre-load buffer settings for your software instruments, thus being able to load much more not even getting close to the 32bit memory treshold.

Hopefully, someday.

- Piotr

Yeah, that's true ... Personally, haven't spent much time experimenting with the pre-loads as my resources are very limited, but the idea does sound interesting. I wonder how much an 87500 / 700 would be able to handle before choking. It might depend on as little as which instruments are being used and how (percussion beds with tons of sixteens would probably make things harder).

[paulwr]

Originally Posted by Vatroslav

Since it was 5 GBs, I'm guessing it was a 64-bit system? Which DAW was it?

Do you remember any specs of the experiment? Because it might have been an overload which is hardly to ever happen in any production.

I'll try to get more info next week, but it was a 64bit system, not sure if 2 or 4 cores.

I think we are inching toward a super system, but that it will be 12-18 months before we have enough components and software up to speed to think about dedicating a DAW for broad composing to such a system. Even then, I think we'll be wanting slaves but just not as many.

Slaves done up with 64bit , now that is another story because you can isolate a group of compatible stuff if it exists for what you need software-wise. I'll dedicate a slave or more to 64 bit and tons of ram when Play comes for SO or when the new Gigastudio 4 shows up AND proves itself. If Gigastudio actually lives up to the promise of running vsti's within its standalone on a slave, that could be very interesting. Supposedly you can even apply vst effects to a vsti that doesn't in itself allow vst plugins. So say on Kontakt, you could apply your favorite compression to a buss or something like that, on the slave. I can make a lot of use of that kind of flexibility.

-Paul