CAS 2 vs. CAS 3 speed differences in RAM with Athlon 64?

[ANTant]

Hello.

I am going to upgrade my old gaming box to an Athlon 64 system. I
noticed the memory CAS speeds. Does CAS 2 (faster) and CAS 3 (slower)
make any big speed differences for 1 GB of PC3200 RAM on a Socket 754
ATX motherboard with an AMD Athlon 64 3000+/2.0 Ghz to 3400+ 2.4 Ghz
(both with 512 KB socket and 754 CPU)? CAS 2 is expensive so... I am
wondering if getting 3 is really worth the price. I am mainly gaming,
watching movies, using the Internet, etc. Gaming is the big one to
note.

You can see my current Athlon XP 2200+ system at
http://alpha.zimage.com/~ant/antfarm/about/computers.txt ... I will be
replacing the motherboard (ASUS K8V SE Deluxe or a MSI K8T Neo F...),
CPU, RAM, and sound card (getting an Audigy 2).

Thank you in advance.
--
"Oh, look what Kyle got me, it's a red Mega... Ants in the pants? Ants
in the pants?! Ants in the Pants?!! ..." --Eric Cartman in South Park's
Damien Episode (Season 1; Episode 8)
/\___/\
/ /\ /\ \ Phillip (Ant) @ http://antfarm.ma.cx & http://aqfl.net
| |o o| | E-mail: (e-mail address removed) or (e-mail address removed)
\ _ / Nuke ANT from e-mail address if your e-mail was returned.
( )

 

[DDStech]

The CAS ratings are latency cycles. What this means is how many clock ticks
before the memory can read/right again. For CAS 2, this is two mhz, for CAS
3 this is 3mhz. In theory, this calculates to CAS 3 being 30% slower than
CAS 2. Now you know why CAS 2 has a premium on it.

For gaming, if frames per second are your main concern, then you need CAS 2.

 

[David Maynard]

The CAS ratings are latency cycles. What this means is how many clock ticks
before the memory can read/right again. For CAS 2, this is two mhz, for CAS
3 this is 3mhz.

You were right when you said clock ticks. That is not 2, or 3, 'MHz' however.

In theory, this calculates to CAS 3 being 30% slower than
CAS 2. Now you know why CAS 2 has a premium on it.

If CAS delay were the only thing to memory access that would be right but,
in reality, that occurs once per data stream so the one 'difference' is
spread over multiple reads making it relatively minor in the total picture,
say a few percent.

 

[DDStech]

Can you explain the difference? The clock cycle is in mhz. Perhaps I am
jumping to oonclusions here.

However, there is a huge difference between cas2 and cas3 memory settings.
It is easily noticed by the human eye. Windows snap open alot quicker,
programs pop up faster. Games run noticibly faster.

 

[Lachoneus]

However, there is a huge difference between cas2 and cas3 memory settings.

It is easily noticed by the human eye. Windows snap open alot quicker,
programs pop up faster. Games run noticibly faster.

Are you sure you didn't just have window animation turned off when you
were running CAS2? Because I could barely tell the difference at all.
Benchmark improvements were almost in the noise. IMHO, CAS2 is not
worth the price premium.

 

[ANTant]

Wow, 30% slower? That's a big number! Thanks for the comments!

The CAS ratings are latency cycles. What this means is how many clock ticks
before the memory can read/right again. For CAS 2, this is two mhz, for CAS
3 this is 3mhz. In theory, this calculates to CAS 3 being 30% slower than
CAS 2. Now you know why CAS 2 has a premium on it.

For gaming, if frames per second are your main concern, then you need CAS 2.

--
"Oh, look what Kyle got me, it's a red Mega... Ants in the pants? Ants
in the pants?! Ants in the Pants?!! ..." --Eric Cartman in South Park's
Damien Episode (Season 1; Episode 8)
/\___/\
/ /\ /\ \ Phillip (Ant) @ http://antfarm.ma.cx & http://aqfl.net
| |o o| | E-mail: (e-mail address removed) or (e-mail address removed)
\ _ / Nuke ANT from e-mail address if your e-mail was returned.
( )

 

[ANTant]

Are you sure you didn't just have window animation turned off when you
were running CAS2? Because I could barely tell the difference at all.
Benchmark improvements were almost in the noise. IMHO, CAS2 is not
worth the price premium.

Hmmm... not 30% speed difference? OK, I really need to know if CAS 2 is
really worth buying for major speed difference. 30% is a lot. If it is
5%, I would just get CAS 3 and save my money.
--
"Oh, look what Kyle got me, it's a red Mega... Ants in the pants? Ants
in the pants?! Ants in the Pants?!! ..." --Eric Cartman in South Park's
Damien Episode (Season 1; Episode 8)
/\___/\
/ /\ /\ \ Phillip (Ant) @ http://antfarm.ma.cx & http://aqfl.net
| |o o| | E-mail: (e-mail address removed) or (e-mail address removed)
\ _ / Nuke ANT from e-mail address if your e-mail was returned.
( )

 

[David Maynard]

Can you explain the difference? The clock cycle is in mhz. Perhaps I am
jumping to oonclusions here.

Well, even if it could be expressed in Mhz, 2 or 3 is way off no matter how
you look at it. Take a 100Mhz FSB. 1 'clock' represents '100 Mhz'. 2 would
then be half that speed, or '50 Mhz'. (I'm using SDR PC100 SDRAM here for
simplicity's sake). The number is obviously larger the faster the FSB is,
so there's no way we're going to come around to '2 or 3 MHz'.

However, 'MHz' refers to a repetitive, cyclic, thing and the CAS '2 clock
ticks' doesn't repeat every '2 clock ticks' (which would be needed to claim
it's '50 MHz'). It occurs at the beginning of a burst read and then again
before the next burst. The repetition rate, which one might say is a 'MHz'
number, is, then, more related to the number of bytes in the data stream
than it is to whether it's CAS 2 or 3. To make matters even more
complicated, there are other delays besides simply CAS (why you see SDRAM
specs with those three numbers. e.g. 3-2-2 [for CAS 3])

Let's take a simplified example for a read without precharge. The delay
before read is CAS plus Trcd, 3+2, 5 clocks. Then a 4 clock data read.
That's 9 clocks. At 100MHz FSB that's 90ns which translates to 11.1 Mhz, if
it repeated over and over.

However, the repetition depends on there being constant reads, that the
selected bank doesn't change, in which case the extra 2 (or 3) clock
precharge time must be added, and we're ignoring things like bank
interleaving and command overlap.

In other words, it's gets rather complicated to do simple calculations like
"2 is 33% faster than 3" and have much meaning. Note that in our simplified
example, going from CAS 3 to CAS 2 memory lowers the clocks needed from 9
to 8; roughly 11% better, not 33%. But, as noted, that doesn't take into
account other memory cycles, such as the 2 clock precharge.

And then, the main memory access is being L2 cached, so that it's from the
L2 that the processor really reads, and that tends to mask 'improvements'
in the main memory access times since that is, after all, the purpose of L2
cache to begin with: to minimize the effect of main RAM delays.

However, there is a huge difference between cas2 and cas3 memory settings.
It is easily noticed by the human eye. Windows snap open alot quicker,
programs pop up faster. Games run noticibly faster.

I seriously doubt that whatever difference you think you're seeing is
because of CAS 2 vs 3. Odds are something else changed at the same time.

 

[Yousuf Khan]

The CAS ratings are latency cycles. What this means is how many clock
ticks before the memory can read/right again. For CAS 2, this is two
mhz, for CAS 3 this is 3mhz. In theory, this calculates to CAS 3
being 30% slower than CAS 2. Now you know why CAS 2 has a premium on
it.
For gaming, if frames per second are your main concern, then you need
CAS 2.

No chance. CAS stands for Column Address Strobe, and its latency is the
number of cycles it takes to activate it. If you picture RAM as a grid, then
you will have the memory arranged in rows and columns. Similarly the RAS is
the Row Address Strobe, with its own latency timing.

Due to the mainly linear nature of memory accesses, you will only need to
activate the CAS only once in awhile and keep it open, while you cycle
through the memory using the RAS. This is known as a burst transfer. Really,
RAS is the more important measure of the speed of RAM, because it's done so
much more often. But RAS timings don't really change from module to module,
it's always stuck at about 3 cycles in DDR-RAM. So the only timing that
might be slightly better or worse is the CAS timing, so that's what they
advertise. Really, CAS timing is sort of a measurement of what kind of a
delay you can expect for a burst of data to begin coming to you after you've
told it to start, since CAS is one of the first activities done.

Now a DDR-200 (aka PC1600) runs at 100Mhz, DDR-266 (PC2100) runs at 133Mhz,
DDR-333 (PC2700) runs at 166Mhz, and DDR-400 (PC3200) runs at 200Mhz. CL3
would mean 3 cycles for CAS, and CL2 would mean 2 cycles. So a DDR-200 CL2
module would have a CAS latency time of 2/(100) = 50ns; while a DDR-400 CL3
would have a CAS latency of 3/(200) = 15ns. So a DDR-400 even with a CL3 has
considerably lower latency than a CL2 DDR-200.

So yes, CAS latency does make a small difference, but we're only talking
about nanoseconds at the beginning of a burst.

Yousuf Khan

 

[Yousuf Khan]

Can you explain the difference? The clock cycle is in mhz. Perhaps I
am jumping to oonclusions here.

However, there is a huge difference between cas2 and cas3 memory
settings. It is easily noticed by the human eye. Windows snap open
alot quicker, programs pop up faster. Games run noticibly faster.

Did you make the observations of the RAM on the same machine, or is this an
observation based on comparing two separate machines (i.e. one running CAS2
and the CAS3, obviously)?

Yousuf Khan

 

[Yousuf Khan]

Hmmm... not 30% speed difference? OK, I really need to know if CAS 2
is really worth buying for major speed difference. 30% is a lot. If
it is 5%, I would just get CAS 3 and save my money.

You'd be lucky if it made even 5% of a difference. The best way to look at
CAS latency is as a measurement of how quickly a RAM module reacts to a
command. It's actual sustained speed is represented by its PC rating
(PC2100, PC2700, etc.), or alternatively its DDR rating (DDR-266, DDR-333,
etc.), same thing. The sustained speed is the same no matter whether you are
getting an expensive CAS2 or a cheaper CAS3. The CAS rating is just a
measurement of its initial reaction time.

The above is assuming that that we're comparing two modules with the exact
same sustained speed ratings, but different CAS ratings. If we're talking
about PC2100 CL2 vs. PC2700 CL3, then by all means get the PC2700, as it
would be much faster all around.

Yousuf Khan

 

[ANTant]

You'd be lucky if it made even 5% of a difference. The best way to look at
CAS latency is as a measurement of how quickly a RAM module reacts to a
command. It's actual sustained speed is represented by its PC rating
(PC2100, PC2700, etc.), or alternatively its DDR rating (DDR-266, DDR-333,
etc.), same thing. The sustained speed is the same no matter whether you are
getting an expensive CAS2 or a cheaper CAS3. The CAS rating is just a
measurement of its initial reaction time.

The above is assuming that that we're comparing two modules with the exact
same sustained speed ratings, but different CAS ratings. If we're talking
about PC2100 CL2 vs. PC2700 CL3, then by all means get the PC2700, as it
would be much faster all around.

This was assuming PC3200 like Kingston brand. I guess I will go with CAS
3 then to save money and since the speed difference isn't that big. I
have until Wednesday night to decide.
--
"Oh, look what Kyle got me, it's a red Mega... Ants in the pants? Ants
in the pants?! Ants in the Pants?!! ..." --Eric Cartman in South Park's
Damien Episode (Season 1; Episode 8)
/\___/\
/ /\ /\ \ Phillip (Ant) @ http://antfarm.ma.cx & http://aqfl.net
| |o o| | E-mail: (e-mail address removed) or (e-mail address removed)
\ _ / Nuke ANT from e-mail address if your e-mail was returned.
( )

 

[Tony Hill]

I am going to upgrade my old gaming box to an Athlon 64 system. I
noticed the memory CAS speeds. Does CAS 2 (faster) and CAS 3 (slower)
make any big speed differences for 1 GB of PC3200 RAM on a Socket 754
ATX motherboard with an AMD Athlon 64 3000+/2.0 Ghz to 3400+ 2.4 Ghz
(both with 512 KB socket and 754 CPU)? CAS 2 is expensive so... I am
wondering if getting 3 is really worth the price. I am mainly gaming,
watching movies, using the Internet, etc. Gaming is the big one to
note.

Generally speaking you're looking at about a 0-3% difference in
performance, probably about 1.5% on average. It makes sense if you're
going for a high-end system, where the extra $50 for CAS2 memory makes
about as much difference as an extra $200 worth of CPU speed.
However, given that you're looking at a 3000+ to 3400+, the saved on
CAS3 memory would be better spent elsewhere.

If you were to compare a system with an Athlon64 3000+ and CAS2 memory
to one with an Athlon64 3200+ with CAS3 memory, the two systems would
cost about the same amount but the latter system would usually be
faster.


Note that the difference is definitely *NOT* 33% as another poster
mentioned. From the test bellow, an Athlon64 3200+ (2.0GHz) has a
total latency for an uncached memory access of 83ns, or 166 clock
cycles for the processor (this is a slight over-simplification, but it
will suffice for the example).

http://www.aceshardware.com/read.jsp?id=60000258

This test used CAS2 memory. Going to CAS3 memory would add a single
200MHz clock cycle on the memory side of things, or 10 clock cycles as
far as the processor is concerned. The result would be an uncached
memory access time of 88ns. So we get a performance improvement of
about 6% when going from CAS3 memory to CAS2.

Of course, uncached memory access is only part of the story. In
reality, well over 90% of all memory access is handled by the L1 and
L2 cache, though even at that the MUCH slower main memory access takes
a disproportionate amount of time (for comparison to that 166 clock
cycles for memory, L1 cache normally takes 2 clock cycles and L2 cache
takes usually 7-20 clock cycles).

In the end, main memory access makes up at most about 50% of
performance. So, 6% faster memory gets you, at most, a 3% speed
boost. Typically the benefit is actually smaller than that, or about
the 1.5% improvement I mentioned above.

 

[Jon Danniken]

No chance. CAS stands for Column Address Strobe, and its latency is the
number of cycles it takes to activate it. If you picture RAM as a grid, then
you will have the memory arranged in rows and columns. Similarly the RAS is
the Row Address Strobe, with its own latency timing.

Due to the mainly linear nature of memory accesses, you will only need to
activate the CAS only once in awhile and keep it open, while you cycle
through the memory using the RAS. This is known as a burst transfer. Really,
RAS is the more important measure of the speed of RAM, because it's done so
much more often. But RAS timings don't really change from module to module,
it's always stuck at about 3 cycles in DDR-RAM. So the only timing that
might be slightly better or worse is the CAS timing, so that's what they
advertise. Really, CAS timing is sort of a measurement of what kind of a
delay you can expect for a burst of data to begin coming to you after you've
told it to start, since CAS is one of the first activities done.

Now a DDR-200 (aka PC1600) runs at 100Mhz, DDR-266 (PC2100) runs at 133Mhz,
DDR-333 (PC2700) runs at 166Mhz, and DDR-400 (PC3200) runs at 200Mhz. CL3
would mean 3 cycles for CAS, and CL2 would mean 2 cycles. So a DDR-200 CL2
module would have a CAS latency time of 2/(100) = 50ns; while a DDR-400 CL3
would have a CAS latency of 3/(200) = 15ns. So a DDR-400 even with a CL3 has
considerably lower latency than a CL2 DDR-200.

So yes, CAS latency does make a small difference, but we're only talking
about nanoseconds at the beginning of a burst.

Of all the explanations I have read concerning memory timings, that is far and away
the most concise and practical one I have ever had the pleasure to read. Thanks,
Yousuf!

Jon

 

[AD.]

Can you explain the difference? The clock cycle is in mhz. Perhaps I am
jumping to oonclusions here.

Ignoring the mega/milli prefixes for a second, aren't you confusing
'somethings' with 'somethings per second'?

Or did I misread your question?

Cheers
Anton

 

[AD.]

DDR-200 CL2 module would have a CAS latency time of 2/(100) = 50ns;

Great post, but shouldn't that be 20ns?

Cheers
Anton

 

[Lachoneus]

You'd be lucky if it made even 5% of a difference. The best way to look at

... snip ..

This was assuming PC3200 like Kingston brand. I guess I will go with CAS
3 then to save money and since the speed difference isn't that big. I
have until Wednesday night to decide.

....

Okay, I don't want to throw a monkey wrench in things here, but there's
also CAS2.5, which is what I got for my new Athlon 64. It was the same
price as the CAS3 when I bought it, so I thought it wouldn't hurt.

I don't know exactly what it means, but I'd imagine it's somewhere
between CL2 and CL3 in performance, meaning you'd be lucky if the
performance difference in any real-world situation wasn't lost in the
noise.

 

[DDStech]

After reading Toms hardware article on memory timings I stand corrected. It
appears to gain about 6% over CAS 3. However, 6 % performance gain is worth
a few extra bucks in my opinion. Also, most of the time, your only quoted
the CAS latency. What you find when you get the chip is that its not
3-2-2-5, but, 3-4-4-9. The ram I'm using is 2-2-2-5.

I ran into this problem with my last system. I bought one stick of ram, that
I was able to run at 2-2-3. When I added ram, I bought a similar type, but
it would only run at 3-3-4. This made a noticible difference in performance.
Windows didn't snap open as fast, and games ran slower.

 

[chrisv]

The CAS ratings are latency cycles. What this means is how many clock ticks
before the memory can read/right again. For CAS 2, this is two mhz, for CAS
3 this is 3mhz. In theory, this calculates to CAS 3 being 30% slower than
CAS 2. Now you know why CAS 2 has a premium on it.

Clueless.

 

[Flow]

I ran cas 2,5 memory but when it went bust i upgraded to a cas 2 stick.
I find that it overclocks better and runs just that little smoother opposed
to the cas 2,5 memory.
Dunno about percentages though.Im a gamer and will never go back to slower
ram.
I have all mem settings in bios at most agressive possible and it runs
better then with slower ram.
This is ofcourse my opinion,can't back it up with figures.
If you like to save some money then go for the cheaper ram at a small cost
of performance,if you want the best then get the more expensive memory.