Why is my memory slower on B550 despite being clocked higher?

Associate
Joined
16 May 2008
Posts
2,486
Location
Bristol
I recently moved from Z170 i5 6600 non K at 4.28GHZ to B550 3700X @ stock but kept using the same memory, ssd, gpu and psu as before. In userbenchmark I had substantially faster memory performance on the old setup even though it's running at a higher speed on the new one and on a 4 year newer chipset!

I haven't touched timings on either system. On the B550 I've set FLCK to 1800 and increased the memory multiplier to 36. The memory is Crucial Ballistix DDR4-3200 16-18-18-36.

Z170:
pBUTPuJ.png
B550:
77zL2IU.png

Is this a common issue?
 
Soldato
Joined
18 May 2010
Posts
22,372
Location
London
You say it's running at 3600MHz but what timings are the ram modules using?

Sometimes to compensate for the higher frequency the mobo will set looser timings so you don't get instability.

I'd guess at 3200MHz you where using tighter timings than you are now at 3600MHz.
 
Man of Honour
Joined
22 Jun 2006
Posts
11,624
It's difficult to understand what these graphs are actually showing, would you be ranked higher if you have memory that's faster than everyone else on the same platform? I'd imagine such fast memory was uncommon on your old system, but common on your new system.

From what I understand of the architecture, Skylake-S and derivatives generally outperformed AMD until (or including?) the 3 series with memory access times, because of the ring bus.
 
Associate
OP
Joined
16 May 2008
Posts
2,486
Location
Bristol
You say it's running at 3600MHz but what timings are the ram modules using?

Sometimes to compensate for the higher frequency the mobo will set looser timings so you don't get instability.

I'd guess at 3200MHz you where using tighter timings than you are now at 3600MHz.

They are set to whatever XMP is choosing as I've never manually configured them and I didn't record them from the old setup unfortunately. I've set it to 3400MHz to match the previous setting but it's no better.. I'm trying to play with the timings manually but it's getting quite tedious resetting the BIOS constantly. Z170 automatically recovered itself on these failed overclocks.

It's difficult to understand what these graphs are actually showing, would you be ranked higher if you have memory that's faster than everyone else on the same platform? I'd imagine such fast memory was uncommon on your old system, but common on your new system.

From what I understand of the architecture, Skylake-S and derivatives generally outperformed AMD until (or including?) the 3 series with memory access times, because of the ring bus.

Ignore the graph on the left, its the right 3 columns that are showing comparable numbers. I thought it would be on par at least but certainly not slower :confused:
 
Soldato
Joined
18 May 2010
Posts
22,372
Location
London
They are set to whatever XMP is choosing as I've never manually configured them and I didn't record them from the old setup unfortunately. I've set it to 3400MHz to match the previous setting but it's no better.. I'm trying to play with the timings manually but it's getting quite tedious resetting the BIOS constantly. Z170 automatically recovered itself on these failed overclocks.

What I am saying is if you're ram only has an 3200MHz XMP profile then when you set 3600MHz in the BIOS the mobo will compensate for the speed by loosing the timings.

I apologise if I have misunderstood.

Additionally if you want to play with our ram on Ryzen download Typhoon Burner export a profile in to the dram calculator and see what settings it suggests you.

If you fire up the dram calc and go to the memtest section it will show you your timings also, my guess s at 3600MHz they are very loose which is why your speed is slower than 3200MHz.
 
Last edited:
Associate
Joined
10 Dec 2007
Posts
1,689
I wouldn't look at userbenchmark to tell you how your RAM is doing. You also can't directly compare RAM between Intel and AMD there is more going on. For example let's take the latency measurement. A finely tuned 5600x chip with bleeding edge stability 4000+ Mhz RAM is around 50ns. On Intel a simple XMP 3200 Mhz 16-18-18-38 kit with 0 tuning is about 48ns. Basically the world record AMD RAM latency is barely able to keep up with a simple XMP latency of a prebuilt Intel PC. Is this bad? No because the platforms are completely different.

If you want to make sure your RAM is performing how it should for the AMD platform then do the following. Download ZenTimings: https://zentimings.protonrom.com/

Make sure MCLK, FCLK, UCLK are the same value and nothing looks out of the ordinary. You can even screenshot and post it here for extra clarification.

Do a proper RAM benchmark with Aida64.
 
Last edited:
Associate
OP
Joined
16 May 2008
Posts
2,486
Location
Bristol
Thanks for the advice above.

I've spent way too long playing with these settings now but in the end I found 3600 was generating errors so I had to dial it back to 3466Mhz then tweaked the timings using DRAM calculator.
eRXFnyu.png

I think this is the best I'm going to get, not as good as Intel but in reality probably wont make any difference.
Inzb1Lz.png
 
Associate
OP
Joined
16 May 2008
Posts
2,486
Location
Bristol
Using some man maths I managed to justify getting some B-die 3200 CL14.

Clocked to CL14 3600 / 1800 FCLK with help of DRAM Calc
9oOnKnh.png

It's not as fast as the old system but close enough for me
XQnNgvR.png
 
Back
Top Bottom