Article: AMD Ryzen DRAM Timings, Frequency and Ranks performance effect in games

Back then I didn't even think about possessing the AMD Ryzen (in the primary PC I had an 8700K, now 9900K), so that can be shown as an excuse

 

Very good article.
We appreciate the work done to see how ram influences ryzen 2.

 

Awesome article! I just think the nomenclature on the graphs are not that good.
3733 mhz with how many GBs SR or DR how many slots etc...

 

If not indicated otherwise - the default is 2 x 8 GB

 

Never really under stood double ranked vs single ranked? still not sure I do.

is DR using 4 slots to get say 16gb and SR using 2 slots to get 16gb?

I know general using all ram slots for say 16gb vs 2 general shows some gains, but the whole SR/DR never really made sense to me

 

I disagree and here is why below , sorry to the two people who may have seen these benches before

Although Hilbert Hagedoorn can not test every configuration out there , to me it looks like he has done a massive amount of ram testing and hit all the sweet spots. People should read the conclusion .

I also suggest before you buy new ram, try tightening all ram timings for best results and put any money towards a new GPU instead.

The memory clock (mclk), the memory controller clock (uclk), and the infinity fabric clock (fclk) set to higher settings is a myth really because if your savvy enough, then a person can get the same PC gaming performance results with tight timings set on there ram and should not feel they are missing out on anything.

Tuned Ram AMD Ryzen Fabric Clock 1467Mhz (DDR4 2933Mhz) vs Tuned Ram AMD Ryzen Fabric Clock 1933Mhz (DDR4 3866Mhz)

My testing with normal ram speed vs highest achievable ram speed to date on any AMD Ryzen system that is synced 1:1:1 this also applies to ram synced at 1:1:2 (EG:Higher than 3866MHZ ram) with tight timings. All timings list at end of videos in Ryzen Master

Battlefield 5 and Grand Theft Auto 5 1920x1080 same results


Red Dead Redemption 2 and Sleeping Dogs 1920x1080 and 2560x1440 same results

 

Awesome article.
I believe I left roughly 10% performance on the table with my discount-bin dual-rank 3000mhz cl-15 32gb memory (2x16gb) on the el-cheapo Asrock x570 PG4.
Considering this is a content creation machine and not a gamer as it's primary intended purposes - not too shabby. I am still quite happy with performance, and not worried one bit about the 'missing' 10%, as it's still plenty good and worlds better than the 4790k it replaced. That 10% isn't going to make or break anything I do on here on the daily. Long before I get to ever needing that last drop of performance, I'll need 64gb of memory anyway. It's dubious at best if this cheap motherboard with it's bargain-basement price tag will even be stable at 3000mhz cl-15 with four sticks of dual rank memory on it anyways.

It's good to know that for those of us who wish to tinker, can expect to see some improvements up to the memory / IF divider cutoff. Sadly, my tinkering days are done, this machine runs at stock settings, as I cannot afford to affect stability in any way on my content creation box.
All the speed in the world does you NO good if you get even one hint of instability with it.
@gerardfraser Good content add with those videos.

 

THX H.H. good read

Im at 1900FLCK at 3800MHz CL16-17-17-15 4x8GB DDR4 using 3700X on Hero VI (Latency ~62-63ns)

Settings on good B-die (How to):

DRAM V. 1.4 up to 1.45v
SOC 1.125v or more if not posting (try 1.150v)
VDDG both at min. 0.975v up to 1.1v

Spoiler

 

You're focusing your energy on the wrong goal. You gave up because you're trying to convince Intel fans how good AMD can be with some tweaking. Did you really think you were ever going to get through to that audience in the first place? They don't care. You really need to ask yourself why it's important to you in the first place.

Maybe ignore their replies and focus the context of your comments on helping AMD users get the most from their system. I'm sure a lot of people will find that useful. The more AMD can do to improve in this area out of the box, the less performance will be left on the table for the average Joe that doesn't know any better (which is the majority of users).

 

Great article safe subtimings vs fast you can have 7-10% more performance
this is my 2700x+gskill flarex 3200(xmp profile enabled) vs 2700x@4.4ghz+3466 flarex(tweaked with fast subtimings)

 

Also i have notice that ryzen 1000 series using ryzen calculator gains the most performance vs 2000 and 3000.

 

Hi OnnA try this gear down mode disable,rtt nom off,rtt wr off,rtt park 5,procodt 34.3 and 24,20,20,24 fast preset(trident z silver red) this gives 62.4 latency

 

Terrific article guys. There is very good information here.

One thing I noticed is World War Z seemed to be the most sensitive game to memory settings with the highest values having a range of about 20% above the minimum. That is about twice the span seen in the others. That might be a game worth focusing on for further articles about fine tuning adjustments since the difference in results will be more apparent. It is good to see a variety of games tested but since the results are consistent among most of them I would prefer to see a wider variety of adjustments tested on fewer games.

 

I m sorry but the fact that you have been lucky with a your ram, or that you could swap ram till you got the good one, does not mean that running samsung bdie at 3733 is super easy or that Good B-die can go 3800 cl12.

Is not super easy and execptional B-die can do 3800 cl 12 maybe.

Maybe you are paritcularly good at it, but saying `is super easy` seems too much.

 

Would Intel CPUs have similar results, I tried to set my 4000mhz, 19cl RAM as 1T (instead of default 2T) but PC would not even boot lol, so for now default settings.

 

Well, there's a goal to pursue when for example the next generation will be in, right?

 

Both cpus have benefits

 

Nice review ! Some +5 fps are quite good ... I had to stop at 3600 CL18 when I ordered mine but next time I will buy something better.
One question for HH / reviewer ... this test was made with best GFX available ... do these gains reproduce with a medium - high gfx ? like let's say 2600 - 2700 RT ?

 

guess my 3600 won the silicon lottery on the memory controller as i run for like half a year hynic cjr at 3800 1:1 cl 16

 

As an average joe, im more than happy to know that a 3600 4x8gb cl 14/16 will work good enough lol

I wont be "wasting" 5 days tweaking memory for a 2% gain (5 fps? lol). If you have fun with it, good for you, enjoy your tweaking.