Originally Posted by Blameless
The AIDA64 memory bandwidth tests have been multi-threaded for quite some time.
I'm not sure about the latency test.
Ah yes, you're right. It's only the latency test that always uses Core 0. Bandwidth should be multi-threaded.
Originally Posted by larrydavid
Elmor: Have you been able to get the PBO settings to actually do anything on the CH6/CH7?
Not tested yet.
Originally Posted by rv8000
It's a bit unclear as to how ASUS is addressing the settings of the FCLK (synchronous/asynchronous) in the bios settings.
Is the setting tied to the memory speed selection, i.e. anything above 3600 will force asynchronous mode, or can you select the fclk ratio separately from the memory speed?
Thanks for sharing these results!
P.S. Did you try any timings scaling testing at 3600?
If you leave FCLK Frequency at Auto, it will keep 1:1 until some point, I believe it's 1800 MHz. When you keep increasing DRAM Frequency, FCLK will still stay at that point if left Auto.
If you manually set both, it will be 1:1 if you set them to that (actually 1:2 as DRAM Frequency is shown as DDR MT/s). There's no separate switch for it, the values just needs to match.
Originally Posted by Nizzen
Why does Ryzen 3k support 4400mhz+ memory, when it doesn't help for performance over 3733mhz?
It does help in some cases for performance. Theoretically if you can push memory high enough you can also get better results in most than at 1:1, but I don't see that happening for most users on this generation. Then again many just look at the frequency not the performance. It's the same on Intel platforms, most high frequency memory does not make much sense in terms of actual application performance. Additionally the decision to go with this implementation was most likely made before they knew how high memory frequency they would be able to get out of the design.
Originally Posted by Anty
Why SOC 1.25?
Why not? I wanted to keep things consistent and it's working well.
Originally Posted by Bart
Awesome stuff, thanks for the great info Elmor! Turns out I might not have to buy new memory after all, unless I want crystal pimp RAM.
I wonder if it's 'safe' to assume that most current good B-die kits are still useful on X570 (ie most 3200CL14 kits)?
From what I'm seeing, the requirements are the same as on previous Ryzen platforms. The main uplift is that you should now be able to get higher memory capacity at much better performance than before (2x16GB, 4x8GB) which goes well with the increased core count and lower DRAM prices.
Originally Posted by Lexi is Dumb
My 3200C14 rated kit isn't working on my X370-F now at all with the 3700X, it just winds up posting in safe mode.
Probably the same issue I had, that the memory voltage is always 1.200V when starting the system and it has had time to execute BIOS code.
Originally Posted by lordzed83
nice to see You post I'm still waiting for 3900x to come and box of toys to build my mate 3600system.
If You got time Would be great if You could drop a quck test of 3600cl16 vs 3600cl14 on latency and lets say quick cb15 score. Very interested how latencies affect performance. All I'w seen is 3200cl14 vs 3600cl16 tests
I'll add it to the list
Originally Posted by gamervivek
Single chiplet CPUs suck at memory write speeds since AMD have reduced it to half. Happens to 3700X too,
It doesn't seem to have a large real world impact, I'm sure AMD did their research before going with this decision. Keep in mind these are very specific benchmark numbers testing only raw bandwidth and not actual application performance.
Originally Posted by Lexi is Dumb
Okay, I just noticed that the board is putting the voltage at about 1.46v during idle. Should I be worried?
Idle and low load it's fine, for heavy loads it's a bit much.