Post your Zen 2/2+ memory speeds!

Nightfire

2[H]4U
Joined
Sep 7, 2017
Messages
3,279
Thought it would be a good idea to start a centralized thread as a sequel to the thread by bobzdar.
https://hardforum.com/threads/post-your-ryzen-memory-speeds.1927366/

It would be good to see what setups do the best for benches, real world gaming, etc.

The more info the better, but try to keep it organized for easier comparisons in the future:

Hardware Used (cpu, ram, MB, etc)
CPU clock speed and voltages/settings
F clock settings and speeds
Bios version
Other software (windows version)

Ram speed and voltage
Primary timings
Other timings or a screenshot.

Benchmarks (Aida 64, Cinebench, games, etc)

Let the hours of tweaking commence!
(but not me, I will have to wait a few months :p)
 
Last edited:
Not worth a new thread, bit G.skill has really set the bar high with their TridentZ neo 3800c14
https://www.techpowerup.com/257844/...4-memory-kit-for-amd-ryzen-3000-x570-platform

66.3 ns memory latency
9.8 ns L3 cache
Both numbers are 10% better than Techspot running 3200c14

Read speeds of 58k MB/s and write of 56k MB/s were even bigger gains than the 47k/46k that Techspot saw.
https://www.techpowerup.com/257844/...4-memory-kit-for-amd-ryzen-3000-x570-platform

Real world advantages are not really significant, but it is cool to see Ryzen 3000 push those kind of numbers.

edit: 1.5v - well you knew it wasn't going to be easy.
 
could be worse lol i have never had a crash so there is that
Capture.PNG
 
Running a set of Hyper X Fury Black 32GB (2x16) 2666MHz CL16, got lucky and got a set with Samsung b-die chips and running @ 3533MHz 18/21/21/21/41/1t at 1.25v on a 3600x with ASRock Steel Legend X570 board. I was really conservative on timings to do some initial testing, but will be messing around more to find tighter timings at 3533. The memory is capable of higher speeds, but I got a pretty poorly binned 3600x, so it won't boot if Infinity Fabric is set higher than 3533MHz.

I tried running the memory briefly at 16/19/19/19/37/1t @ 1.35v, but the transfer speeds didn't increase and the latency only went from 71ns to 70ns, so it didn't feel worth the .1v increase. If I can get some better all around timings at 1.35v and get a worthy feeling increase on transfer speeds and latency, I may stick with it.
 
I some how ended up with two different sets of G-Skill 3200Mhz [ 2x4Gb ] so all 4 slots have memory in them as one has Red and the other has Black Heatsinks and rated timings are 16-18-18-38-56-1 @1.35v

So , all I did was set it to XMP profile one in bios and rebooted and Infinity Fabric linked 1 to 1

https://valid.x86.fr/f0lfv1

and how it performs in gaming with a very stock RX 570 if you want to know for that type of build .

 
Last edited:
ram2.PNG


This RAM (32GB Sniper X 3600MHZ 19-20-20-40) is coming out next week for a slightly better set (32GB Trident Z Neo 16-19-19-38). Between the two sets I could have bought one of the really good stuff for less, but it's too late for that now! I'll probably sell the Sniper X for cheap since I only paid $125 for it.
 
I some how ended up with two different sets of G-Skill 3200Mhz [ 2x4Gb ] so all 4 slots have memory in them as one has Red and the other has Black Heatsinks and rated timings are 16-18-18-38-56-1 @1.35v

So , all I did was set it to XMP profile one in bios and rebooted and Infinity Fabric linked 1 to 1

https://valid.x86.fr/f0lfv1

and how it performs in gaming with a very stock RX 570 if you want to know for that type of build .


Nice to see, as I saw several posters suggesting (without evidence) Zen 2 wouldn't work at higher speeds, with 4 dimm slots occupied.
 
Nice to see, as I saw several posters suggesting (without evidence) Zen 2 wouldn't work at higher speeds, with 4 dimm slots occupied.


I wanted to go back and visit Hawaii with the 3600 as it's set up , it has the latest chipset and driver 19.7.5 with all bios and windows updates and Gen 4 PCI Express is selected in the bios and noting is overclocked . Radeon Free Sync and Anti Lag is in play .. One thing I noticed with Fire Strike is that it likes Hawaii in combine test as to be 12% faster then my RX 570 and I think Ryzen wants that 512mb memory bus Hawaii offers over the standard 256mb ring bus .

I jumped into a game at high levels and took a bit to get going after getting my ass kicked.. but after the windows updates for DX 12 it my first look at Hawaii running it and looks very rich in detail and sharp to me .

 
Last edited:
Nice to see, as I saw several posters suggesting (without evidence) Zen 2 wouldn't work at higher speeds, with 4 dimm slots occupied.

It's really more than two banks (sides) per channel. We should still expect having four double-sided DIMMs to be more of a challenge.
 
Are people paying for a licence on Aida 64? its like $60 where i live and that seems like alot to pay just to run a few benches...
 
but here is the numbers i get with the free version anyways:

3600 fast timings - Copy.PNG



Using these settings:

20190802_175317.jpg
 
I wanted to go back and visit Hawaii with the 3600 as it's set up , it has the latest chipset and driver 19.7.5 with all bios and windows updates and Gen 4 PCI Express is selected in the bios and noting is overclocked . Radeon Free Sync and Anti Lag is in play .. One thing I noticed with Fire Strike is that it likes Hawaii in combine test as to be 12% faster then my RX 570 and I think Ryzen wants that 512mb memory bus Hawaii offers over the standard 256mb ring bus .

I jumped into a game at high levels and took a bit to get going after getting my ass kicked.. but after the windows updates for DX 12 it my first look at Hawaii running it and looks very rich in detail and sharp to me .



Strange Brigade will run on a vacuum cleaner. I can run it at 100fps on a block of wood lol
 
Sometimes I feel like Steve from Techspot monitors my [H] posts and then does incredible work covering it:
https://www.techspot.com/review/1891-ryzen-memory-performance-scaling/

The real world gains in gaming with the tighter timings is WAY better than I expected. The 1% low performance was especially better with the tighter timing even when running cheaper ddr4 3000.

It would be interesting to see if the 9900k sees these kinds of gains with tighter timings or if it will be left in Ryzen's dust...
 
Damn! Your read speeds and latency match the uber G.skill neo sticks while smoking them in write speed. Those are some insane timings. What voltage did you need to hit that.

Also, 4.425 ghz is no less of a feat. An R20 score more than double that of a gen 1 8core. Very nice.

1.45 volts on ram
 
1.45 volts on ram


I have the same kit but with dual 16GB double sided DIMMs running at 3400 14-15-14-28-1T...I can get the same 3600 pandora's box is getting, but I am not 100% sure it is stable yet. I am probably 98% of the way there.

It was really hot this weekend and with my loop torn down I was reaching the limits of my what a case with hardly no airflow being able to tolerate. I didn't want the DIMMs to get toasty since I was already hitting 75C on the CPU during 16 thread handbrake/real bench testing.


I think once I have my loop up and with the 3x120mm top fans blowing down cool outside air on the VRM/DiMM area I will get 3600 stable. If that fails I will try 3533 straight 14, and failing that 3400 c14 straight.
 
My results with Crucial Ballistix LT (Micron e-die) 2x8GB kit with XMP 3200 16-18-18. Motherboard is MSI x570 Gaming Plus. CPU is 3600X.

1T max was 3600 14-19-16-36 @1.41v (I added 0.01v for 24/7 after validating 1.41v overnight.)
2T 3800 15-20-17-38 @ 1.40v (I added 0.01v for 24/7 after validating 1.40v overnight.)
Here are the screenshots with Ryzen Master showing all timings / subtimings & AIDA 64 Read / Copy / Latency. These have also been validated HCI memtest overnight ~1000% each thread for both:
3600 - and 3800 -

Ran Shadow of the Tomb Raider as a benchmark, since this was in a system with a crappy video card during performance optimization I do for my new CPUs before putting them into my main system, and I will never be able to get a CPU bottleneck. SoTR benchmark reports a supposed "If there was a CPU bottleneck it would be at" FPS. While this may or may not be valid, the trend should be valid enough to prove the point of how memory will affect a CPU limited scenario.

I ran 1080P most settings maxed (no HBAO or Motion Blur) in the following speed / timings:
3200 XMP / XMP @ 1900 Infinity Fabric / 3600 14-19-16 / 3800 15-20-17 / 3600 - subtimings tuned / 3800 - subtimings tuned
AVG FPS (multiple runs, averaged, rounded to nearest 0.5FPS):
125.0 / 125.5 / 132.5 / 133.5 / 139.5 / 141.0
5% low FPS:
95.5 / 95.0 / 99.5 / 100.5 / 106.0 / 107.0

During this I had the CPU locked at 4.0 GHz to minimize run to run variation. When CPU was left stock and allowed to boost, results were all over the place. When kept at 4.0GHz, the performance was VERY steady, with sigma ~1 FPS for AVG and under 1 FPS for 5% lows. As an example, for the 16 runs that I did for 3600 & 3800 w/ subtimings tuned (8 each) only 3 runs for 5% lows were not a 106 or 107 (2 105s on 3600 and one 108 on 3800). Even though there's only 1 FPS difference between the 3800 and 3600 tuned subs 5% lows, the variation was low enough that I am confident that the difference is statistically significant.

Overall it demonstrates just how much improvement can be had from subtimings. Sure speed and latency are important, but 3200 16-18-18 --> 3600 14-19-16 offered slightly less improvement than just tightening subtimings at 3600 or 3800. I'm sure 3800 would offer even more improvement if I didn't have to drop to 2T command rate, but that's what it took. 1T wasn't stable even at 3666 20-22-22.

Overall, I'm pretty happy with this from a kit of memory that was 16GB for $80 including tax and shipping. Sure b-die would do better, but it's way more pricey. Considering I don't often game in CPU limited scenarios anyway, expensive memory doesn't make sense. This kit was great with a midrange price, but able to overclock to decidedly higher than midrange performance.
 
Last edited:
My results with Crucial Ballistix LT (Micron e-die) 2x8GB kit with XMP 3200 16-18-18. Motherboard is MSI x570 Gaming Plus. CPU is 3600X.

1T max was 3600 14-19-16-36 @1.41v (I added 0.01v for 24/7 after validating 1.41v overnight.)
2T 3800 15-20-17-38 @ 1.40v (I added 0.01v for 24/7 after validating 1.40v overnight.)
Here are the screenshots with Ryzen Master showing all timings / subtimings & AIDA 64 Read / Copy / Latency. These have also been validated HCI memtest overnight ~1000% each thread for both:
3600 - and 3800 -

Ran Shadow of the Tomb Raider as a benchmark, since this was in a system with a crappy video card during performance optimization I do for my new CPUs before putting them into my main system, and I will never be able to get a CPU bottleneck. SoTR benchmark reports a supposed "If there was a CPU bottleneck it would be at" FPS. While this may or may not be valid, the trend should be valid enough to prove the point of how memory will affect a CPU limited scenario.

I ran 1080P most settings maxed (no HBAO or Motion Blur) in the following speed / timings:
3200 XMP / XMP @ 1900 Infinity Fabric / 3600 14-19-16 / 3800 15-20-17 / 3600 - subtimings tuned / 3800 - subtimings tuned
AVG FPS (multiple runs, averaged, rounded to nearest 0.5FPS):
125.0 / 125.5 / 132.5 / 133.5 / 139.5 / 141.0
5% low FPS:
95.5 / 95.0 / 99.5 / 100.5 / 106.0 / 107.0

During this I had the CPU locked at 4.0 GHz to minimize run to run variation. When CPU was left stock and allowed to boost, results were all over the place. When kept at 4.0GHz, the performance was VERY steady, with sigma ~1 FPS for AVG and under 1 FPS for 5% lows. As an example, for the 16 runs that I did for 3600 & 3800 w/ subtimings tuned (8 each) only 3 runs for 5% lows were not a 106 or 107 (2 105s on 3600 and one 108 on 3800). Even though there's only 1 FPS difference between the 3800 and 3600 tuned subs 5% lows, the variation was low enough that I am confident that the difference is statistically significant.

Overall it demonstrates just how much improvement can be had from subtimings. Sure speed and latency are important, but 3200 16-18-18 --> 3600 14-19-16 offered slightly less improvement than just tightening subtimings at 3600 or 3800. I'm sure 3800 would offer even more improvement if I didn't have to drop to 2T command rate, but that's what it took. 1T wasn't stable even at 3666 20-22-22.

Overall, I'm pretty happy with this from a kit of memory that was 16GB for $80 including tax and shipping. Sure b-die would do better, but it's way more pricey. Considering I don't often game in CPU limited scenarios anyway, expensive memory doesn't make sense. This kit was great with a midrange price, but able to overclock to decidedly higher than midrange performance.



Nice results, especially form an $80 kit. I wish I had known Asus was going to break radio silence and post a new BIOS release on the 31st like they did. I would have waited to order my 32GB B die kit (but it was a good deal at $190 vs $130 for every 16GB b die kit!) In order to give my Micron E Die kit (same as yours but again was a 32GB DR 3200c16 kit) a fair shot at getting stable.

I could run 3200c16 across but could get it stable in memory benchmarks despite passing everything else.
 
Ml
Ran Shadow of the Tomb Raider as a benchmark....
I ran 1080P most settings maxed (no HBAO or Motion Blur) in the following speed / timings:
3200 XMP / XMP @ 1900 Infinity Fabric / 3600 14-19-16 / 3800 15-20-17 / 3600 - subtimings tuned / 3800 - subtimings tuned
AVG FPS (multiple runs, averaged, rounded to nearest 0.5FPS):
125.0 / 125.5 / 132.5 / 133.5 / 139.5 / 141.0
5% low FPS:
95.5 / 95.0 / 99.5 / 100.5 / 106.0 / 107.0
...

This is some seriously Pro work.
5% gains in that game by no change other than tighter subtimings is nothing to sneeze at.

Here are a few other games that saw significant gains:
Screenshot_20190808-232702_YouTube.jpg

The bottom 2 bars are $70 T-Force Dark. The middle 3 are $130 g.skill flareX and the top 2 are $170 g.skill Neo. Both the Dark and Neo were given tweaks. Both saw 5% gains on avg fps and 10-15% boost to 1% lows.

An earlier test showed a stock 3900x at 166/219 fps (similar to 3200c14 above) while the 5 ghz 9900k maxed out at 170/228 fps.

https://www.techspot.com/review/1877-core-i9-9900k-vs-ryzen-9-3900x/

Assasin's Creed was even more impressive and tangible with the lower overall fps marks:
Screenshot_20190808-225638_YouTube.jpg

Although not the same settings, you can easily extrapolate and see how a 3900x could blow past even a 5.0 ghz 9900k in this game:
ACO.png


Full video here:
 
Damn! Your read speeds and latency match the uber G.skill neo sticks while smoking them in write speed. Those are some insane timings. What voltage did you need to hit that.

Also, 4.425 ghz is no less of a feat. An R20 score more than double that of a gen 1 8core. Very nice.

It's hard to get these timings at 3600 unless your B die silicon is really binned. I think mine are just as tight. But I havent aida tested my Dominator platinum.

I just used Ryzen ram tool to get the timings
I'm using the fast settings for my 3600 b die corsair dom. 3600 at cl 14 is nuts fast. True latency I got was like 67ns.
 
Last edited:
It's hard to get these timings at 3600 unless your B die silicon is really binned. I think mine are just as tight. But I havent aida tested my Dominator platinum.

I just used Ryzen ram tool to get the timings
I'm using the fast settings for my 3600 b die corsair dom. 3600 at cl 14 is nuts fast. True latency I got was like 67ns.


Yeah I agree about the binning. What voltage are you using for SoC/Vddr? I am at 1.1V and 1.45V respectively, but that was to ensure stability first. I haven't had time to try to lower them down due to cooling issues.

I went to reassemble my loop and something told me to check my CPU block. Opened it up and it was full of crap. Checked the tubing and it was as well. I made the mistake of letting the loop sit with some water in it during my back surgery and we had some insane humidity with it regularly getting up to 70%+ indoors. It's 68f and 71% as I type this with my AC set to 70F.

I had to order a new CPU block since I couldn't get it clean and all new tubing and have to flush my rads. Never had a single issue with this stuff in the past until moving to the PNW and this darn rainforest.

/Rant
 
Yeah I agree about the binning. What voltage are you using for SoC/Vddr? I am at 1.1V and 1.45V respectively, but that was to ensure stability first. I haven't had time to try to lower them down due to cooling issues.

I went to reassemble my loop and something told me to check my CPU block. Opened it up and it was full of crap. Checked the tubing and it was as well. I made the mistake of letting the loop sit with some water in it during my back surgery and we had some insane humidity with it regularly getting up to 70%+ indoors. It's 68f and 71% as I type this with my AC set to 70F.

I had to order a new CPU block since I couldn't get it clean and all new tubing and have to flush my rads. Never had a single issue with this stuff in the past until moving to the PNW and this darn rainforest.

/Rant

I live in GA and its probably more humid than the pacific nw. If your indoor RH is 71% with the AC on your AC is probably too small for your house. You might want to put a dehumidifier in your computer room or upsize your central HVAC unit.

Im running a ton too large in my home and it gets my RH down to 35% when the outside can be 100%. My water pump is constantly running every 15 mins full tank under my AC unit downstairs.

Of course the house came with this oversized AC when I bought it so I am not talking as if I am an authority on it. Just what I have noticed with this really large unit I have.
 
I live in GA and its probably more humid than the pacific nw. If your indoor RH is 71% with the AC on your AC is probably too small for your house. You might want to put a dehumidifier in your computer room or upsize your central HVAC unit.

Im running a ton too large in my home and it gets my RH down to 35% when the outside can be 100%. My water pump is constantly running every 15 mins full tank under my AC unit downstairs.

Of course the house came with this oversized AC when I bought it so I am not talking as if I am an authority on it. Just what I have noticed with this really large unit I have.


I moved from the mid Atlantic, so 100% humidity year round was the norm for the first 34 years of my life.

Here in the PNW nothing, even new construction here has central AC (at least on the coast like we are). I have a 9000 BTU portable AC that I use in the living room, but if the temperature is within 3F of the "set temperature" then it mainly runs on fan mode since they use a small high efficiency compressor and a very small drain tank since they normally dump the water vapor back into he cold coils to increase efficiency even more.

Last summer was fine, it rarely got above 65-68C with 35-50% humidity (without the AC running) but this year it has been extremely humid here with outdoor humidity reaching 85-95%...

I truly miss central AC. I had considered buying a whole house humidifier but the problem is one of storage. We have a 1100 sq ft apartment and the way it is laid out makes it hard since the back of the house (living room/dining room/kitchen) gets full sun (when we get it lol) during the morning and the front of the building (where both bedrooms are) get blasted by sun for the last 10 hours of the day making them super hot).

We have another one of the portable ACs in the bedroom but I do not run that during the day, just usually turn it on about 30 mins before bed and it runs as needed at night.
 
Nice to see, as I saw several posters suggesting (without evidence) Zen 2 wouldn't work at higher speeds, with 4 dimm slots occupied.

yeah but he had 4gb sticks prob single rank. i'm guessing you would have more issues trying to OC 4 X 16 GB dual rank sticks
 
Threadripper 2950x, 4x8gb Corsair Vengeance RGB at 3333 CAS 16

The kit is rated for 3466 CAS 16 (verified working on Intel) but this MSI X399 board hates running above 3333 (same issue with my previous 1920x). Oh well, good enough.
 
yeah but he had 4gb sticks prob single rank. i'm guessing you would have more issues trying to OC 4 X 16 GB dual rank sticks


Zen2 is limited to 2933 with all 4 DIMMs filled by SR modules..If you go to 4 mixed SR/DR offical speeds drop to 2666.

This is the reason I went for 16GB dual rank sticks this time around. It allowed the best mix of capacity (32GB) with the best possible speed using 2 DIMMs.

Even I can't end up getting 3600c14 stable, 3400c14 is wicked fast.
 
Swapped out my Sniper X for some Trident Z Neo. In case anyone is interested how the mid-grade Neo stuff works (2 x 16GB 3600MHz 16-19-19-39 geardown on) with the default XMP profile, here's a benchmark:

ramnew.PNG


In comparison to the cheaper Sniper X:

ram2.PNG


Not sure why it's showing unknown for my motherboard in AIDA now, but the rest of the system remains the same. XMP Profile 2 for both sets of ram.

Edit: I might have changed chipset drivers between the runs, but I can't remember.
 
This is the reason I went for 16GB dual rank sticks this time around. It allowed the best mix of capacity (32GB) with the best possible speed using 2 DIMMs.

Even I can't end up getting 3600c14 stable, 3400c14 is wicked fast.

me too. what sticks did you get? i picked up the micron TUF 2X16 3200 c16 @1.35 (to run on ASUS TUF X570 when i get the $$) hoping to get at least what you're getting. figured they have a little oc headroom being they are 1.35v dimms

did you try 3600 c16? supposedly faster memory = faster infinity fabric = better performance??
 
me too. what sticks did you get? i picked up the micron TUF 2X16 3200 c16 @1.35 (to run on ASUS TUF X570 when i get the $$) hoping to get at least what you're getting. figured they have a little oc headroom being they are 1.35v dimms

did you try 3600 c16? supposedly faster memory = faster infinity fabric = better performance??

I have a G.Skill 3000c14 B die kit...I can run 3600c14 but I haven't gotten it 100% stable yet. I have my loop all torn apart due to a mess up on my end resulting in stupid growth in there since I let it sit idle with fluid in it and the PrimoChill Liquid Utopia wasn't enough to keep it away with no circulation.

I was going to test 3600c14 more but I was reaching the limits of air cooling with intake fans due to them being attached to my rads so I stopped.

I hope to all it all back together by late Monday or Tuesday and will do more testing then.

I had a 3200c16 Micron E die kit that ran 3600c16 and fine with normal use but the launch bios was iffy and I couldn't get it completely stable.

I may have lost the IMC lottery, but even if so I'm not that bummed. 3400c14 is super fast, my AIDA64 results are almost as good as pandora's box .
 
Not sure why my Memory Bandwidth and Latency are so bad compared to others considering I am at DDR4 3533 and 1730 mem bus with CL14 ram. Any tips? this is just XMP settings and the CPU is a 3600 running at about 3.9GHZ all core load and about 4.0-4.1GHZ single core boost. Current AGESA is 1.0.0.2 (Asus B450i ROG board) I think PBO is broken with the current AGESA/Bios combo. I am bit disappointed in the current build... but I knew this being an early adopter.

Current Ram Setup.PNG
 
Not sure why my Memory Bandwidth and Latency are so bad compared to others considering I am at DDR4 3533 and 1730 mem bus with CL14 ram. Any tips? this is just XMP settings and the CPU is a 3600 running at about 3.9GHZ all core load and about 4.0-4.1GHZ single core boost. Current AGESA is 1.0.0.2 (Asus B450i ROG board) I think PBO is broken with the current AGESA/Bios combo. I am bit disappointed in the current build... but I knew this being an early adopter.

Your tRFC is 606. Cut that in half. It is the single most important number for Ryzen memory performance.
 
Back
Top