Intel G3 SSD Roadmap

[flamenko]

80/160/300/600

And another area to consider is that Intel has also moved to full disk encryption which may effect performance results...

 

[Makaveli]

80/160/300/600

And another area to consider is that Intel has also moved to full disk encryption which may effect performance results...

I'm waiting for more details on this will encryption be on by default on the drives?

Will a 32nm i7 that supports AES help with this drive level encryption?

If the encryption is hardware based on the SSD will there really be a big performance hit ?

So many questions!

 

[eric.kjellen]

Will a 32nm i7 that supports AES help with this drive level encryption?

If the encryption is hardware-based in the SSD then the CPU wouldn't be able to help at all. Maybe depending on how performance turns out it will be better if you have an AES-accelerated CPU to leave hardware FDE off and let software take care of drive encryption, from just a performance perspective that is.

 

[Mark R]

Also IMFT's flash is still 2 bit per cell right? I thought the 3 bit per cell flash was meant more for USB flash drives and stuff like that, not high performance applications like SSD.

That's right. No manufacturer has released or announced 3 bpc flash which is suitable for SSDs. In all cases, the flash is too slow, has too high an error rate, and has too short an endurance.

The current generation IMFT 25 nm 2bpc Flash is specified to have the same write cycle endurance as the older generation 32 nm 2bpc. Whether this means that they are genuinely the same, is open to debate. There is a fair bit of research out now, which suggests that the bench tests used to measure flash write-endurance can seriously underestimate the actual lifespan under normal conditions. As a result, the manufacturers may now have changed to a method which doesn't under-estimate so much.

However, it would be interesting to see benchmarks for both the mainstream and enterprise series drives. One technique for improving flash endurance is to use slower programming. This is the major difference in high-endurance 'enterprise MLC'. A similar approach may also have been used to improve the endurance of the standard 25 nm MLC flash. So, I am curious to see what the benches are, and what mitigation techniques have been used for the 'high-end' enterprise drives, which use a significantly slower memory.

 

[Emulex]

the capacitor backed write cache (no intel has had write-back cache!) will allow the chipset to increase speed and optimize write strategy.

remember all intel flash up till now was READ ONLY and Write strategy use only. the competition has been rocking mad numbers using unsafe non-backed write cache.

so if they throw another 256meg ram chip for write cache and write optimization - that should give them the 3x longer write cycles and offset the slower write speeds at the same time. It may not be linear write but in business iops count for most of us (SQL acceleration, vm acceleration, heavy queue depth).

only time i need linear speed is for backups and traditional dasd is far cheaper per GB to handle that.

 

[tweakboy]

Im just going to wait until the 320GB Sandforce is 100 dolllars. Maybe 2012 or 2013 hmm!!,,

 

[DesktopMan]

And another area to consider is that Intel has also moved to full disk encryption which may effect performance results...

I'd be utterly shocked if the hardware encryption doesn't have the same throughput as the drive. This isn't particularly difficult to do. Hardware implementations can easily do a lot more than the limits of the sata 6gbps bus. See e.g. VIA PadLock which is something like 5 years old. SandForce already does this in their first gen controllers.

 

[Dark Shroud]

The problem is that Intel is supposedly not going with SATA 6gbps speed for G3 drives. If this is true then Intel will have to compete with price vs both Indilinx Sandforce drives.

Intel really needs step up in adding USB3 & SATA 6gbps to their chipsets.

 

[rgallant]

-at least no one is fighting over the name [G3 should be 3x faster than G2 to have the G3 name lol] -like the upcoming release of the amd video cards.
-but really the ssd's on the market to day can't even flat line[all benches\real world] the sata I spec's let alone needing sata III.except for large files, where sata II would be great. so flat lining sata II spec. should be the goal post -not sata III.
all marketing\dreaming.
-would be like -fighting over the max. speed signs that state -60 miles per hour for right turns on city corners , but a F1 car can do it at 62 mph so the speed limit should be changed to 100 mph max. and yet no car can do the turn faster than 20 mph without rolling ,
-a good ssd review
http://www.behardware.com/articles/794-1/ssd-2010-report-15-models-compared.html

 

[Emulex]

really depends on your intent - what i said a year or so ago - the OS then the APP needs to be aware of its storage - so it can participate in tiering - (ram,swap,ssd,hard drives) in unity.

Encryption if you think about it like compression can result in write patterns that are to your advantage or not. think really hard about that.