G
Guest
Guest
It has been now some days since AMD released it's 760
chipset and so, the first DDR platform for pcs.
Benchmarks where quite nice and showed some potential
the DDR-tech. can offer with reasonable price, but
one think was left unmentioned, which i'd like to address.
DDR-tech isn't drop in upgrade, it needs support by the
programs using it to archive the best performance from
it. That is, there must be linear or atleast longer reads
and minimun amount of random access. This is because the
DDR-tech have about the same latency than equally clocked
SDR, but twice the bandwidth. If only small random reads
used, performance should be identical for both techs.
So randomreads should atleast be twice in width compared
to SDR. This doesn't affect the performance of SDR at all,
but will efficiently double the performance of DDR.
Another think where the benchmarking fails is highly
optimized games, like Quake3. Quake3 is optimized so
that it archives all performance it can get from SDR
and videocards. That is, it has been coded and balanced
to have certain factor of memory read and cpu usage.
As such, switching to DDR memory and board doesn't
help it at all, because the bottleneck will be cpu or
videocard. Situation would be different if it had been
optimized to use DDR. Then it could be a lot faster with
DDR-memory. Probably the DDR optimizations would not have
much affect as cpu can't do much better that it's doing
now.
Not to mention the we allso have quite big caches, which
efficiently hides the main memory performance.
(how could i otherwice get 150MB/s with my old p133, which
only had 133MB/s peak memory performance ).
Just my few.. ahem.. $.
chipset and so, the first DDR platform for pcs.
Benchmarks where quite nice and showed some potential
the DDR-tech. can offer with reasonable price, but
one think was left unmentioned, which i'd like to address.
DDR-tech isn't drop in upgrade, it needs support by the
programs using it to archive the best performance from
it. That is, there must be linear or atleast longer reads
and minimun amount of random access. This is because the
DDR-tech have about the same latency than equally clocked
SDR, but twice the bandwidth. If only small random reads
used, performance should be identical for both techs.
So randomreads should atleast be twice in width compared
to SDR. This doesn't affect the performance of SDR at all,
but will efficiently double the performance of DDR.
Another think where the benchmarking fails is highly
optimized games, like Quake3. Quake3 is optimized so
that it archives all performance it can get from SDR
and videocards. That is, it has been coded and balanced
to have certain factor of memory read and cpu usage.
As such, switching to DDR memory and board doesn't
help it at all, because the bottleneck will be cpu or
videocard. Situation would be different if it had been
optimized to use DDR. Then it could be a lot faster with
DDR-memory. Probably the DDR optimizations would not have
much affect as cpu can't do much better that it's doing
now.
Not to mention the we allso have quite big caches, which
efficiently hides the main memory performance.
(how could i otherwice get 150MB/s with my old p133, which
only had 133MB/s peak memory performance ).
Just my few.. ahem.. $.