Question Ryzen 1600AF Overclock. RAM or IMC Slowing me down?

beastykato

Member
Jul 20, 2017
36
7
51
I recently got a 1600AF and I've been playing with it a little on a Asrock B450M Pro4.

I had a junk set of DDR4 laying around that I used for the build so I didn't have to waste money on memory. It's a patriot set 4gbx2 with Spectek (Micron) chips and is dual rank, 2133, CL15 rated @ 1.2v.

Now I have two issues here as I'm new to overclocking Ryzen. 1. I've read that the 1600AF, despite being 12nm, still uses the 1st gen memory controller that is often speed limited. 2. My RAM is $30 bargain basement ram lol

So, I've gotten the chip to 4.0GHz @ 1.25v and the memory at what I believe is stable at 2733. That's with 1.1v SOC and 1.2v DRAM. I also loosened the RAM timings to CL 17 (17-18-18-18-36). I've got no idea what these timings mean I just raised them up because CL15 wasn't stable at 2733.

Anytime I go higher I get crashes of applications in Windows or sometimes a failure to boot altogether. I've gotten into Windows at as high as 3200mhz on the RAM, but it was very sketchy with programs basically doing as they pleased.

My question now is how do I differentiate between whether it's my ram holding me back or the IMC? I've raised the DRAM voltage all the way to 1.45v and I've raised the SOC to 1.2v trying to get anything above 2733 stable and it has all failed, so I just reset everything back because I didn't want to keep pushing it.

I can continue to try to loosen the CL as well maybe all the way to CL 19 or 20? At what point does that become a losing proposition though? Is 2733 CL17 > 3200 CL20, for instance?

Thanks for any clarification.
 

Gideon

Golden Member
Nov 27, 2007
1,608
3,573
136
If your RAM is really rated that low you're almost surely limited by it, not the memory controller.

Higher clocked memory at the same CAS latency actually has lower actual latency. Here is a calculator for that. 2133, CL15 truly is garbage (~14.06 ns) that's about the same as 2666 @ CL19 (14.25 ns). I'm actually surprised you even got 2733 @ CL 17 (12.44 ns) out of it.


If you want to try further, then also tuning secondary tunings is a must (and knowing which memory chips the modules are using) You can do that using thaiphoon burner and Ryzen timing calculator. This video explains it quite thoroughly:


Overall, you should be able to improve things a bit just with better seconday timings. With some luck you might also improve upon primary timings or Mhz. I would first try the conservative presets though but no further than the calculator says is possible. I've only gotten FAST to work stabley with Samsung B-Die (and higher level stuff like Micron E-Die is also probably good). Hynix has usually ended up stable just around the SAFE preset.

You could try better memory modules as well, but you'll be hitting diminishing returns really soon. I'd say that for this ultra-budget CPU 2733 @ CL17 is just about good enough, that I wouldn't upgrade the memory unless you're also limited by the 8GB capacity (which is a lot more likely).
 
  • Like
Reactions: Thunder 57

VirtualLarry

No Lifer
Aug 25, 2001
56,229
9,990
126
I'd say that for this ultra-budget CPU 2733 @ CL17 is just about good enough, that I wouldn't upgrade the memory unless you're also limited by the 8GB capacity (which is a lot more likely).
This. I got some OLOy DDR4-2400 (Apparently, native JEDEC 2400) stable at 2800 in a Ryzen APU system, but I didn't feel like "pushing it". No, I didn't use "Ryzen DRAM calculator".
 

thor23

Member
Jul 13, 2019
80
22
81
I had a junk set of 2 x 8gb spectek, I couldn't get it to run stable at cl16 -2666mhz or higher. It overheated I think whenever I went over 1.3v. It ran ok at cl14-2400mhz-1.3v.
I would run yours at a lower frequency like 2400-2666mhz and try to get cl14-16 working.
btw my soc only needs 0.775v to run 2733mhz, 0.725v for 2400mhz.
 

beastykato

Member
Jul 20, 2017
36
7
51
Thanks for the replys.

Memory overclocking has long been something I've neglected to fully learn and understand I'll definitely go over all the information posted.

I'm totally happy with the ram at 2666 or 2733 considering it's base specs it's hard to complain. I just wasn't sure how to figure out what the limitation really was.

I'll definitely look into that lower SOC voltage too. I know I'm stable here so I intend on just lowering both SOC and core in small increments until they can no longer hold these speeds. I'm on the stock stealth cooler so I'm not expecting any more than this; 4.0Ghz seems more than adequate.
 

beastykato

Member
Jul 20, 2017
36
7
51
Just wanted to clarify here since thor suggested I actually tighten timings at a lower ram speed.

From what I've read regarding Ryzen would it not be more beneficial for me to run a loser timing at a higher speed if the effective latency is the same because it also increases the infinity fabric speed?

For example from the ram latency chart I'm viewing on reddit my ram stock at stock 2133 @ Cas 15 has a latency of 14.06, but 3200 at Cas 20 has a latency of 12.5.

So, 3200 even at such crap latency would be superior?
 

LightningZ71

Golden Member
Mar 10, 2017
1,627
1,898
136
From what I've seen on most benchmarks that bother to do actual comparisons of a variety of RAM speeds, for MOST cases, it appears that focusing on memory clock up to 1600Mhz (3200 Mxfr) seems to be the way to go. Once you get to DDR4-3200, you start trying to optimize your timings vs. speed to get the best combination you can manage. Zen/Zen+ seems to have, at least in the chips I've fooled with, a decent % chance of getting the Memory bus up to 1800Mhz/3600xfr reliably. The gains after that, while certainly noticeable in benchmarks, start to really fall off (and this includes changing timings to keep latency somewhat consistent). So, in my experience, the goals are: first, get to DDR-3200 (if you can even manage it, some sticks just won't get there no matter how loose the timings). Second, find your best timings at that speed. Benchmark. Try for the next speed grade i.e. 3266, then 3333, then 3400, then 3466, then 3533, then 3600. At each of those points, find your best timings, then benchmark. It does take a while, and a good DRAM calculator is very helpful. Pick the one that gives you the best overall benchmark results, stress test it. If it fails, try the second best, then the third best. If the three best fail, back off on your timings for each a bit, bench again, then stress test again.

As a reference, if you do intend on investing in better RAM, my son has the ATX version of your board, the B450-Pro4. With a late run 2700x and an "inexpensive" pair of Micron E-die sticks (I think they were just about $60 on sale), his system can pass memory stress tests at 3600 CL16 with modest secondary timings. The 1600 AF is essentially a slightly down clocked 2600, and I'm almost certain that it's your individual RAM sticks that are holding you back. We benchmarked his system when he was building it at the DDR4-2400 speeds his system initially configured with and then at his final settings at 3600. The difference was noticeable in games and the heavier benchmarks, but the general responsiveness of the system in day to day stuff like web browsing and doing schoolwork wasn't even perceptible. If you're not trying to game at really high frame rates for FPS type games, it's probably not worth putting in too much effort or money to get better sticks if what you have works.
 
  • Like
Reactions: VirtualLarry

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
Just wanted to clarify here since thor suggested I actually tighten timings at a lower ram speed.

From what I've read regarding Ryzen would it not be more beneficial for me to run a loser timing at a higher speed if the effective latency is the same because it also increases the infinity fabric speed?

For example from the ram latency chart I'm viewing on reddit my ram stock at stock 2133 @ Cas 15 has a latency of 14.06, but 3200 at Cas 20 has a latency of 12.5.

So, 3200 even at such crap latency would be superior?

There is a double edged sword in there somewhere. Lower cas latency has a big effect on gaming performance. Not major elsewhere but games love low latency. So you want to find a middle ground. If it was between 1-2 CAS then I'd go speed with Ryzen and CAS with Intel. But anything more than that and I don't think the tradeoff is that great, original inter-CXX worries in 2017 were overblown. Not that it doesn't help but most of the issues were scheduling issues that MS has mostly rectified.