AMD Eyzen cpus and ram speed

An item in a recent issue of a UK PC magazine said, in essence, that
using ram with the right speed was essential to get the best out of
AMD Ryzen cpus, which wasn't something that I'd come across before.
But how does ram speed correlate to cpu speed? I didn't get an answer
when I asked the magazine.
I've put the article here for a few days:
https://1drv.ms/u/s!AkhvgPvXD0_1jdpw...EuZ0Q?e=Nyxdhg
(PC Pro No 311 September 2020)

Peter Johnson wrote:
An item in a recent issue of a UK PC magazine said, in essence, that
using ram with the right speed was essential to get the best out of
AMD Ryzen cpus, which wasn't something that I'd come across before.
But how does ram speed correlate to cpu speed? I didn't get an answer
when I asked the magazine.
I've put the article here for a few days:
https://1drv.ms/u/s!AkhvgPvXD0_1jdpw...EuZ0Q?e=Nyxdhg
(PC Pro No 311 September 2020)


I'm not an expert. But think about two devices each working at a
different clock speed that are "talking" together. Lets look at 2
examples (over-simplified):

A: CPU: 12 cycles per minute (one every 5 seconds).
RAM: 6 cycles per second (one every 10 seconds)

The clocks of the CPU and RAM can be perfectly synchronized (timed).
Look at your watch.

B: CPU: 12 cycles per second (one every 5 seconds)
RAM: 5 cycles per second (one every 12 seconds)
IN B it is much trickier for the CPU and RAM to be synchronized . Their
may be more "waiting" (by Each device).

The CPU and RAM can only communicate when both "devices" are ready for
it to happen. Also relevant is how long it takes to execute each part
(micro-instruction) of an instruction among other things, like latency.

I hope that is helpful. I did not read the article you mentioned, but
perhaps I will go look at it now.

Cheers,
Bill

On Tue, 8 Dec 2020 22:54:11 -0500, Bill wrote:

Used to be memory was the smaller of overall improvements, e.g. the
biggest improvement of late being SSDs.

Peter Johnson wrote:
An item in a recent issue of a UK PC magazine said, in essence, that
using ram with the right speed was essential to get the best out of
AMD Ryzen cpus, which wasn't something that I'd come across before.
But how does ram speed correlate to cpu speed? I didn't get an answer
when I asked the magazine.
I've put the article here for a few days:
https://1drv.ms/u/s!AkhvgPvXD0_1jdpw...EuZ0Q?e=Nyxdhg
(PC Pro No 311 September 2020)

The UnCore doesn't clock the same way as the memory.
The performance curves can have a stairstep in them,
caused by the ratio of clocks.

Chiplet IOchip External

CPU ----- UnCore ---- memory
^ ^ ^
| | |
clock clock clock

You can see discussions of the latest blather, here.

https://www.reddit.com/r/overclockin...ng_suggestion/

Any time the clocks are not nice multiples, there can be
additional cycles of latency caused by the need to
enter a "different" clock domain. You want the UnCore to
perhaps "go as fast as it can go", then select a memory speed
that is exactly 2X of that (Being careful when discussing
clocks, because the "rate" on a DIMM like DDR4-3800, relies
on double data rate, both edge clocking, and so the clock
is actually half of the stated rate.)

What you can do, is buy two 4GB (single sided) sticks of
some extremely fast (but not necessarily low CAS) stuff,
and do some clock ratio tests and tease out the details
yourself. The reason you're selecting 4GB sticks, is to
reduce the amount of money wasted on the test case. Maybe
your plan is to buy two 16GB sticks of something tailor
made for your optimized clock values, in which case the
2x16GB might cost more than 4X your test sticks.

For an enthusiast site to do these tests, you have to wait a
bit past "release day" before efforts of this complexity
are ready to be published.

And while the manufacturers and AMD/Intel are getting better
at providing tools, they might still not communicate that
effectively with the public and help people make the right choices.
The enthusiast web sites receive "personal service" from
company representatives, and while the people on the enthusiast
site might appear clever, they're just as clever as the
help they're getting from their contact network. An engineer
at Asus who has worked on three versions of a motherboard,
likely knows all these clock rules, inside-out. They have to
do stuff like figure out what the time-of-flight should be
set to, so they have some harder things to do than the users.
(There has to be sufficient time for bus-turnaround, before
a DIMM output pads are turned on.) There are a few hobbyists
who have even tried to optimize that :-)

Paul