These timings are measured in clock cycles,
so they vary in accordance with the clock
speed of the RAM. As such, a faster clock
speed with a slower latency can be faster than
a slower clock speed with a faster latency,
and vice versa. This is also why latencies have
increased from older generations of DRAM,
yet our RAM doesn’t feel any less responsive.
For the likes of the original DDR standard, tCL
was around 3 whereas now it’s upwards of 15.
However, DDR back then ran at 200-400MHz,
whereas DDR4 runs at ten times that figure. As
such, the real-world latency is little different.PERFORMANCE MATTERS
We’ve generally skipped over any talk of
memory clock speed so far here, as the
concept is so broadly understood. In short,
the same general rules apply to DRAMas any other component: all other factors
being equal, a higher clock speed is better.
However, the exact role of clock speed
depends on two factors.
The first is simply whether your system will
benefit from higher clock speed. Once you get
past a certain point (around 3600MHz for both
AMD and Intel’s latest desktop platforms – as
we saw in our recent RGB memory Labs test
in Issue 199 ), faster-clocked RAM unlocks only
a tiny fraction of extra performance from your
CPU, GPU and other components. It simply no
longer becomes a significant bottleneck for
your system beyond a certain point.
Also, the dependence on memory
speed varies significantly with the type of
processing that’s being performed. Some
calculations are complicated, but are only
being performed on relatively small orquite fixed chunks of data, so they don’t
require a lot of data to be written to and read
from memory. On the other hand, some
calculations are relatively simple, but are
being performed on massive data sets. For
instance, a GPU is dealing with the latter
situation, with it having to constantly churn
through millions of simple calculations on
a huge, ever-changing data set. That’s why
memory bandwidth is so crucial for GPUs.
Along similar lines, memory latency
affects different applications in different
ways. In the same way that vastly faster
access times make SSDs feel faster than
hard drives (even back when overall
SSD read and write speed wasn’t much
better than hard drives), lower-latency
memory also improves the experience for
applications that require a quick response.Four memory chips are packed
together to form a rank of memory,
with two ranks on each DIMMThese timings are measured in clock cycles,
so they vary in accordance with the clock
speed of the RAM. As such, a faster clock
speed with a slower latency can be faster than
a slower clock speed with a faster latency,
and vice versa. This is also why latencies have
increased from older generations of DRAM,
yet our RAM doesn’t feel any less responsive.
For the likes of the original DDR standard, tCL
was around 3 whereas now it’s upwards of 15.
However, DDR back then ran at 200-400MHz,
whereas DDR4 runs at ten times that figure. As
such, the real-world latency is little different.
PERFORMANCE MATTERS
We’ve generally skipped over any talk of
memory clock speed so far here, as the
concept is so broadly understood. In short,
the same general rules apply to DRAM
as any other component: all other factors
being equal, a higher clock speed is better.
However, the exact role of clock speed
depends on two factors.
The first is simply whether your system will
benefit from higher clock speed. Once you get
past a certain point (around 3600MHz for both
AMD and Intel’s latest desktop platforms – as
we saw in our recent RGB memory Labs test
in Issue 199 ), faster-clocked RAM unlocks only
a tiny fraction of extra performance from your
CPU, GPU and other components. It simply no
longer becomes a significant bottleneck for
your system beyond a certain point.
Also, the dependence on memory
speed varies significantly with the type of
processing that’s being performed. Some
calculations are complicated, but are only
being performed on relatively small orquite fixed chunks of data, so they don’t
require a lot of data to be written to and read
from memory. On the other hand, some
calculations are relatively simple, but are
being performed on massive data sets. For
instance, a GPU is dealing with the latter
situation, with it having to constantly churn
through millions of simple calculations on
a huge, ever-changing data set. That’s why
memory bandwidth is so crucial for GPUs.
Along similar lines, memory latency
affects different applications in different
ways. In the same way that vastly faster
access times make SSDs feel faster than
hard drives (even back when overall
SSD read and write speed wasn’t much
better than hard drives), lower-latency
memory also improves the experience for
applications that require a quick response.Four memory chips are packed
together to form a rank of memory,
with two ranks on each DIMM