2666 VS 3200 on Ryzen 1600X

[Johan45]

I've been having some issues while running this ram at 3200 even though it says it's stable in more than one test so I decided to drop the speed and tighten it a bit. The ram is Team Delta 3000 CL 16. I mostly use this PC for gaming, writing and watching TV so I did a couple of game benchmarks and Cinebench R15 it seems to be the only bench that suffered with the lower ram speed.
Just thought I'd share my results here. This is with a 980 STRIX and both game are fully maxed on quality at 1080p

Ram at 3200 CL 16




Ram at 2666 CL 14

 

[Kenrou]

Not fully stable at 3200 ?

 

[trents]

.7% difference between the two. Hardly anything. Did you run it multiple times to check the variance between one run and another?

 

[Kenrou]

.7% ? Did you notice the difference in max fps ?

 

[trents]

OP stated he ran some game tests as well but Cinebench was the only one that he saw any performance deterioration when running the RAM at the slower frequency and <1% is not even statistically significant in my book.

 

[Kenrou]

i didn't just look at overall, this isn't normal especially on a CPU that supposedly loves high speed memory. I'm agreeing with Johan that this is sign of instability, not arguing

 

[Woomack]

I had some issues with 1 out of 3 kits while running at 3200. All were passing every single stability test but while playing games one kit wasn't stable. It's usually causing performance drops or memory errors ( often just application closes or freezes without error ).
3200 should be faster regardless of timings on this platform so if performance is lower than 2666 then something isn't right with settings or stability overall. Actually this is weird ( and some others noticed that too ) but in some tests and games 3200 CL16 or even CL18 is faster than 3200 CL14. It can be caused by stability or something else.

I has similar experience as Johan but in other games. However I just switched to other memory kit which runs perfectly stable at 3200.

 

[Johan45]

I've actually been chasing this random GFX driver crash. The ram at 3200 in all stability tests was fine 3+ hour P95 with 80%+ ram allocated in custom run and IBTAVX with 13000 MB set for ram. The crash will happen in games and even just in internet browser. It's so random too. Can watch a 2+ hr movie and all is fine then maybe 20 minutes in or an hour and a half the driver will crash 3-4 times in a minute.
So far I have tried
Different GFX cards
Multiple BIOS versions
Fresh install of Windows and drivers
It's really bugging the sh** out of me.

 

[trents]

I wouldn't think instability would affect test scores if the benchmark completes and doesn't crash or freeze up. What would be interesting is to see what differences in test scores there would be if you move down from 2666 to 2300/2400. 2666 could be the sweet spot for the R5 1600/1600X. An it could also be that if you move up to the 1700/1800 R7 series you will see more difference using the higher memory speeds because of the extra cores.

 

[Johan45]

Part of that eveness I thin comes from the fact it's just not very good ram. Hates to be overvolted and trying to tighten the timings down at 3200 was a test of wills. If I compared that to a good set of Sammy "B" you would see a big difference I think. There are a couple of timings that would normally be set at '1' on sammy and this couldn't go below '5'

 

[Kenrou]

3600 vs 3200 vs 2600 vs 2133 :

 

[trents]

As I said, there should be larger differences in performance when using higher speed RAM with the R7 CPUs since they have two more cores and two more threads. But this thread is testing an R5. Higher speed RAM only benefits any CPU when the processor is getting starved for data. It's like putting a bigger carburetor on an engine. Unless the smaller carburetor is bottle-necking the air flow into the engine it won't help.

 

[Kenrou]

From what i read about the infinity circuit it works faster the more cores you have so it affects all Ryzen to a certain degree. But from this video it seems that the GPU is also a massive factor when picking high speed memory, he says at the end that 2666mhz seems to be the sweet spot with 1070's and below ?

 

[Johan45]

Or that memory I was testing is at it's performance limit . Those videos can all be taken with a grain of salt. OK it's hard to get 2133 to perform like 3600. But I have seen with ryzen that it really depends on what you're benching. CB15 and CB11.5 to a greater extent always respond better to ram speed. Gaming typically prefers latency. I didn't hear in that video (at work) if those were purchased kits at those speeds or self tuned which is another ball of string.
Yes memory speed matters even on the 6 core but it really depends on how you get that speed and the sticks. Here's a good example I'm ~ 200 MHz slower but faster ram.

5216 MHz 3300 ram CB11.5 20.19



5437 MHz 3100 ram CB11.5 20.21

 

[Johan45]

Here's another that IMO shows how little ram affects gaming at least on my set-up
Switched in the Samsung, I wasn't saying my ram was unstable either just that theres something going on with my system. Dropping to 2666 made no difference with my naggy driver crashes either. I had bot the 3200 profile and 2666 stability tested. This one isn't but I have a pretty good idea of where to set things.
As you'll see the only real difference was with CB15 again but more noticeable this time as I said Samsung id a lot easier to tune for performance I set the ram to 1.4V since I'm still testing might not need it.

 

[trents]

Johan, can you run some file compuression/decompression tests (like 7 zip) with the two frequencies? That might be an even better test than Cinebench to explore the impact of memory bandwidth on performance with Ryzen.

 

[Woomack]

In some computations when error happens then program is calculating again without crash, then it takes longer to finish some task and results are worse.
Modern web browsers are loading a lot of small data into the RAM and are caching a lot of stuff. The same some games. When RAM is unstable then I also see web browsers and games crashing much faster than stability tests. Looks like all stability tests have their own patterns and are not using so many random writes/reads.

Anyway what Johan described is happening about the same on my rig with Ryzen 1700X.

I try not to base on results from the internet just because it's really easy to manipulate with all results when you know how. Some sites are showing big improvements with higher frequency memory, some are showing barely any difference. There are many variables which are affecting performance but if you make everything right ( or motherboard let you ) then higher frequency memory is helping in everything on Ryzen just because everything works on the same bus and is directly connected to memory controller.

 

[Johan45]

I just wanted to follow up on this. The "instability" I thought I had was nothing to do with the PC but my AVR on it's way out. It finally died about a month ago.