Intel's Braidwood looks to take another stab at Turbo Memory
By Ross Miller 
posted June 12th 2009 7:14AM
posted June 12th 2009 7:14AM
Intel's jumped into familiar waters this month at Computex when it showed off Braidwood, a flash memory-based accelerator that works by caching I/O from the processor, reportedly enabling applications to launch much faster. Though not directly stated, for all intents and purposes it looks like this is a spiritual successor to Intel's Robson / Turbo Memory, a similar initiative from the company's past that didn't quite meet the commercial or critical success it had hoped. Whether we see better results this time is gonna be a story we revisit in the more distant future: Braidwood's coming as an option with the Nehalem-based Clarkdale processor, which isn't ramping up production until late this year.
Next topic: return of the math co processor :D.
We are already there, they are called GPUs.
IO by its very nature needs to be 'live', how can you successfully cache it?
No it doesn't. For example, let us say that you are editing a picture. The picture can appear to be edited on screen, but the write-back cache can hold off actually editing the file on the HDD or RAM until a whole bunch of write-backs are necessary, thereby reducing interrupts and CPU cycles that are used by the HDD, or CPU cycles used to write to RAM.
First off that would cause lots of confusion, and lost effort if anything goes wrong, and when it can it will eh, and secondly that would require extensive support from both the OS and even the graphicscard drivers.
And it's a quaint example seemingly desperately trying to prove an untenable position I feel.
But the problem is that the original author used the term I/O but didn't explain enough what he means with it (probably because he doesn't know himself), leaving us a bit groping too, because he says:
The architecture accelerates I/O (input/output) accesses by saving that data to flash memory, according to Crooke. In a demonstration at Computex, Crooke showed Braidwood "caching the I/O...And then, when it launches that application again, it happens very quickly,"
But that's just disk cache described there and not CPU-level I/O what I, and it seems you too, are thinking of when you use the term I/O.
In that description of that "I/O" it's just old readyboost, but surely intel would not introduce that as new and link it to their new architecture, so it can't be readyboost.
That's why my other post is lambasting the author.
so processors that already have 3 levels of cache, still need ANOTHER slower cache, the same speed other flash based technology, like flashdrives and SSDs to store temporary data.
if only there was some sort of intermediary place for processors to store data and access it randomly, and even more quickly in some sort of dual channel configuration
I don't know.
If it works like readyboost, but built into the mobo, rather than taking up my SD slot, I'd be OK with that.
That's exactly how turbo memory works right now. I installed it in my Asus G1-s laptop, cost me about $20 for a 512 stick (2 years ago) and I noticed a bit of a performance bump. Not much, but I'd say it was worth the cost of admission.
I for one will only be happy if it comes with a button on the case labled "Turbo"
The whole cnet article is the worse piece of writing ever I think, and the author should be forced to follow some courses how to present information or if he won't be fired so he can work for some EU government and write their publications.
I can tell you from experience: Turbo Memory, in any form, is absolutely useless compared to an adequate amount of actual memory, which you can generally buy for cheaper. On a related note... I have a 1 GB module of Turbo Memory for laptops, if there are any takers! (fits any mini-pci slot).
Looks like Arab Micro Devices just can't keep up with Intel releasing something
every quarter.