Hello,

  • Thread starter Thread starter Richard K Rabbat
  • Start date Start date
Rob said:
I'd tend to agree with the RAM argument. Super-fast disks come into their
own on server apps and throwing huge video files around, but for most
other uses anything above 7200 SATAs represents a large diminishing
return. Speaking as a user of a Mac Mini (4200, 1Gb RAM) and a PC (7200
SATA, 1Gb RAM).

Rob
Click to expand...

Rob;

What you fail to realize is that the data just does not magically appear
in RAM from out of the blue. The data has to be read from the drive.
Caching, prefetching, long branch predictions...but the data originally
reside *ON THE HARD DRIVE*. RAM is faster. I don't dispute that. What I
wish you and the others to understand is that NCQ allows faster *SUSTAINED*
data transfer, thus giving data to the RAM faster. If you have a file on
the HDD, your RAM cannot just "make it up". It has to access the data on
the HDD. Because NCQ allows out of order execution, it is significantly
faster than any standard IDE device (UDMA), and is also faster than the
Raptors in sustained read, which is what most reads from the HDD are.
Before data can be allocated to RAM, it has to come from *SOMEWHERE*; it
just does not magically appear from the ether. Because NCQ is
non-sequential, allows faster *SUSTAINED* access to data on the HDD, and
with today's faster data buses yielding much faster throughput, NCQ is a
clear winner.

I never disputed RAM is faster. But the data in the RAM comes from
somewhere...

Do you understand? Can you comprehend? RAM it not psychic. It is not
clairvoyant. It gets its data from elsewhere. It releases the data it has
when it is asked. Simple. Elegant. But the date comes from somewhere
*OUTSIDE* the RAM.

I cannot make it any simpler. If you cannot make the connection, I am
sorry. It is your loss, not mine.

Bobby
 
NoNoBadDog! said:
What you fail to realize is that the data just does not magically appear
in RAM from out of the blue. The data has to be read from the drive.
Caching, prefetching, long branch predictions...but the data originally
reside *ON THE HARD DRIVE*. RAM is faster. I don't dispute that. What I
wish you and the others to understand is that NCQ allows faster
*SUSTAINED*
data transfer, thus giving data to the RAM faster. If you have a file on
the HDD, your RAM cannot just "make it up". It has to access the data on
the HDD. Because NCQ allows out of order execution, it is significantly
faster than any standard IDE device (UDMA), and is also faster than the
Raptors in sustained read, which is what most reads from the HDD are.
Before data can be allocated to RAM, it has to come from *SOMEWHERE*; it
just does not magically appear from the ether. Because NCQ is
non-sequential, allows faster *SUSTAINED* access to data on the HDD, and
with today's faster data buses yielding much faster throughput, NCQ is a
clear winner.

I never disputed RAM is faster. But the data in the RAM comes from
somewhere...

Do you understand? Can you comprehend? RAM it not psychic. It is not
clairvoyant. It gets its data from elsewhere. It releases the data it
has
when it is asked. Simple. Elegant. But the date comes from somewhere
*OUTSIDE* the RAM.

I cannot make it any simpler. If you cannot make the connection, I am
sorry. It is your loss, not mine.

Bobby
Click to expand...
I understand. But the point is that as time goes on, there is less and less
need for "sustained" disk access, especially on desktop systems. Even on
database servers (Oracle, DB2, MS SQL Server) it is more and more common
that a large portion of the data remains in memory (except for very large
data warehouse databases). Database products, like many applications and
operating systems, are exploiting memory to a larger degree than ever
before. This is one reason that 64 bit computing is coming, because it is
needed to address more than 4GB of memory.

So big deal, the first time the data is accessed, it comes from disk, but on
subsequent accesses it is more and more likely that it will be in memory
cache.
 
NoNoBadDog! said:
Rob;

What you fail to realize is that the data just does not magically appear
in RAM from out of the blue. The data has to be read from the drive.
Caching, prefetching, long branch predictions...but the data originally
reside *ON THE HARD DRIVE*. RAM is faster. I don't dispute that. What I
wish you and the others to understand is that NCQ allows faster *SUSTAINED*
data transfer, thus giving data to the RAM faster. If you have a file on
the HDD, your RAM cannot just "make it up". It has to access the data on
the HDD. Because NCQ allows out of order execution, it is significantly
faster than any standard IDE device (UDMA), and is also faster than the
Raptors in sustained read, which is what most reads from the HDD are.
Before data can be allocated to RAM, it has to come from *SOMEWHERE*; it
just does not magically appear from the ether. Because NCQ is
non-sequential, allows faster *SUSTAINED* access to data on the HDD, and
with today's faster data buses yielding much faster throughput, NCQ is a
clear winner.

I never disputed RAM is faster. But the data in the RAM comes from
somewhere...

Do you understand? Can you comprehend? RAM it not psychic. It is not
clairvoyant. It gets its data from elsewhere. It releases the data it has
when it is asked. Simple. Elegant. But the date comes from somewhere
*OUTSIDE* the RAM.

I cannot make it any simpler. If you cannot make the connection, I am
sorry. It is your loss, not mine.

Bobby
Click to expand...

Yes, thanks Bobby, I do understand that data originates from the HD, and
that moving it to RAM is an issue if the HD is slow. My point was that,
for most users, there's little between HDs that would make RAID or NCQ
worthwhile (hence 'diminishing returns'), and a move from (say) 512 to
1Gb of RAM would be wiser.

I accept that a faster HD will result in faster opening of programmes
for example - but it's such a small amount of time difference I, for
one, wouldn't consider it a limitation.

Now, if you're talking about throwing around 1Gb video files, or access
from 300 users on a network, then yes, I'm with you. Just out of
interest - where do you notice the best improvements in use (not
benchmarks!)?

Rob
 
Rob said:
Yes, thanks Bobby, I do understand that data originates from the HD, and
that moving it to RAM is an issue if the HD is slow. My point was that,
for most users, there's little between HDs that would make RAID or NCQ
worthwhile (hence 'diminishing returns'), and a move from (say) 512 to 1Gb
of RAM would be wiser.

I accept that a faster HD will result in faster opening of programmes for
example - but it's such a small amount of time difference I, for one,
wouldn't consider it a limitation.

Now, if you're talking about throwing around 1Gb video files, or access
from 300 users on a network, then yes, I'm with you. Just out of
interest - where do you notice the best improvements in use (not
benchmarks!)?

Rob
Click to expand...

I notice it in several areas...as you mentioned, when I am working with
video files, there is a dramatic difference.
Also, when I am working with multiple files under Microsoft Office (with
nested tables, relational DB, etc), since the data is being moved and
updated in several locations simultaneously, I noticed a marked improvements
Such as when file in Excel must update a file in PowerPoint). Also, when
streaming audio, my CPU usage is lower because the NCQ controller handles
the disk reading...nice benefit.
My whole intention in this thread was to impress upon readers that in
everyday use, NCQ was better than Raptors...it gives an increase across the
board in routine usage. I do not dispute that in certain scenarios that
Raptors might be a better choice, but since NCQ drives are cheaper, more
reliable, and in most real world applications faster, then they are clearly
the better choice. Why spend the money on Raptors when most of us would not
really place enough demand on them to warrant the cost. NCQ gives its
benefits all the time (sort of like Hypertransport, if you can understand
the analogy).

Bobby
 
NoNoBadDog! said:
I notice it in several areas...as you mentioned, when I am working with
video files, there is a dramatic difference.
Also, when I am working with multiple files under Microsoft Office (with
nested tables, relational DB, etc), since the data is being moved and
updated in several locations simultaneously, I noticed a marked improvements
Such as when file in Excel must update a file in PowerPoint). Also, when
streaming audio, my CPU usage is lower because the NCQ controller handles
the disk reading...nice benefit.
My whole intention in this thread was to impress upon readers that in
everyday use, NCQ was better than Raptors...it gives an increase across the
board in routine usage. I do not dispute that in certain scenarios that
Raptors might be a better choice, but since NCQ drives are cheaper, more
reliable, and in most real world applications faster, then they are clearly
the better choice. Why spend the money on Raptors when most of us would not
really place enough demand on them to warrant the cost. NCQ gives its
benefits all the time (sort of like Hypertransport, if you can understand
the analogy).

Bobby
Click to expand...

Fair enough - I'll certainly look into NCQ on the back of what you say
come next upgrade.

Rob
 
Back
Top