Threadripper 2950X - Need help choosing high speed ~3600Mhz ram for gaming performance

[hitme987]

Hi All,

Looking to upgrade my RAM keeping gaming performance as priority. My PC Specs are as follows:

 

CPU : AMD Ryzen Threadripper 2950x with CM Wraithripper cooler  
Mobo: MSI X399 Gaming pro Carbon AC
GPU : Nvidia Zotac RTX 2080 Ti
Memory: 2x16GB DDR4 G.Skill 3200Mhz Ripjaw V-series CL16 Hynix
PSU: Coolermaster V1000 Vanguard 1000W 80+Gold Modular

 

I am currently running dual channel 2x 16GB @ 3133 Mhz CL16 timings, but the gaming performance isn't that great. 20-30 FPS Drops at times with Hitman 2 & Witcher 3 at Ultra and wanted to upgrade as ryzen performs better with quad channel. My Ripjaws isn't B-Die so not able to push over 3133. 

 

I have read Threadripper/ryzen prefers speed over cas and I am looking at 3600 MHz or more Samsung B-die chips. Worst case I can still configure at 3466 Mhz .

 

I have gone through recommendations for threadripper all over the internet and came to a conclusion with the below options. Wanted to have a second opinion on what to buy (made a mistake going for ripjaws earlier). Budget is upto 600$ and want a quad channel 4x8GB ram combo.

 

Option 1:
G.SKILL TridentZ (4 x 8GB) 3600 Mhz CL16 Model F4-3600C16Q-32GTZKK (16-16-16-36) @ 399$

https://www.newegg.com/Product/Product.aspx?Item=N82E16820232584

Findings:
- Samsung K4A8G085WB-BCPB B-die as per this review:   https://www.tweaktown.com/reviews/8268/skill-tridentz-rgb-ddr4-3600-32gb-memory-kit-review/index.html
- As per a 3D mark benchmark, someone was successfully do 3600Mhz on a threadripper 1950x & ASRock X399 Taichi on this ram (not sure whether my MSI x399 carbon pro ac may get the same love). : https://www.3dmark.com/spy/2340426
- Not in my motherboard QVL but Samsung B-die and with above benchmark is it enough for me to go ahead? Not sure if it can reach 3600Mhz.

 

Option 2:

2 kits of below:
G.SKILL Trident Z RGB (For AMD) 16GB (2 x 8GB) 3200 Mhz CL14 Model F4-3200C14D-16GTZRX (14-14-14-34) @  460$ (2x229$)

https://www.newegg.com/Product/Product.aspx?Item=N82E16820232650&ignorebbr=1

 

Findings:
- Samsung B-Die, Has AMD compatibility and will work at 3200Mhz (Might overclock to 3333/3466).
- Does 2 kits of 2x8 GB will be able to do Quad channel @ 3200Mhz or more on a ryzen?
- Seems overpriced (as it is 3200Mhz only)

 

 

Option 3:

 

2 kits of below:
G.SKILL TridentZ Series 16GB (2 x 8GB) 3600 CL15 Model F4-3600C15D-16GTZ (15-15-15-35) @ 440$ (220$x2)

https://www.newegg.com/Product/Product.aspx?Item=N82E16820232306

Findings:
- Samsung B-die but not in my motherboard QVL.  Not sure if it can reach 3600Mhz and also if 2 x 2 x 8GB would be fine.

 

Option 4:

 

G.SKILL TridentZ RGB Series 32GB (4 x 8GB) 3200 CL16  Model F4-3200C16Q-32GTZRX @300$

https://www.newegg.com/Product/Product.aspx?Item=N82E16820232753

Findings:
- Most affordable but not sure if it has Samsung B-Die and can overclock to 3466/3600

 

I am leaning towards option 1 or option 3 as its 3600Mhz ryzen seems to favor gaming performance but concerned if it would be compatible.

 

Let me know if you guys have a better suggestion/alternative.

 

Thanks for the help
Regards.

Edited February 9, 2019 by hitme987

[Stormseeker9]

Corsair vengeance 

[nick name]

TridentZ 3600CL15 is some of the best Samsung b-die.  And it's cheaper now than when I bought it.  It's about $217 on Newegg right now.  

 

Edit:

 

And for around the price I paid when I bought my kit you can get 4000CL17 kits now too.  Though don't expect to get 4000MHz speeds.  I've only heard of one person doing that and it was with an expensive Corsair Dominator Special Edition kit.  

[hitme987]

Thanks nick name. Even as I researched more on this, understood that TridentZ 3600CL15 seems a better kit as 3600CL15 > 3200CL14 in terms of performance. My only concern is whether running it in quad channel 4x8 might reduce speeds to 2933 Mhz because of ryzen/threadripper compatibility.

 

Not able to find a good 4000 CL17 in my budget of 600$. Will look around a bit, mostly may go for TridentZ 3600CL15

[nick name]

14 minutes ago, hitme987 said:

Thanks nick name. Even as I researched more on this, understood that TridentZ 3600CL15 seems a better kit as 3600CL15 > 3200CL14 in terms of performance. My only concern is whether running it in quad channel 4x8 might reduce speeds to 2933 Mhz because of ryzen/threadripper compatibility.

 

Not able to find a good 4000 CL17 in my budget of 600$. Will look around a bit, mostly may go for TridentZ 3600CL15

Well that guy that was running 4000MHz was running 4 sticks, but again he is an outlier.  Also, definitely wasn't running CL15.  

 

But your concerns are warranted as running four sticks will tax the IMC more and also motherboard RAM topology will play a larger role.  

 

Edit:

Level1Techs has videos where they successfully achieved speeds faster than 2933MHz so don't give up hope.   

[hitme987]

14 hours ago, nick name said:

Well that guy that was running 4000MHz was running 4 sticks, but again he is an outlier.  Also, definitely wasn't running CL15.  

 

But your concerns are warranted as running four sticks will tax the IMC more and also motherboard RAM topology will play a larger role.  

 

Edit:

Level1Techs has videos where they successfully achieved speeds faster than 2933MHz so don't give up hope.   

Hmm gotta rely on my luck. I have ordered CL15 3600 one and gonna be testing in a few days. Will update the results. Thanks again for the support

[timbergus]

Hi @hitme987. I have a similar doubt here. I'm trying to improve my current computer with faster RAM. I have a four years old G.Skill Ripjaws F4-2400C15Q-32GRK (8x8), and I have bought two different types to replace it. I have benchmarked them using 3DMark, Cinebench, and CPU-Z, and I'm pretty disappointed with the results. The first one is a G.Skill F4-3600C18D-16GTZRX (8x8), compatible with AMD, and the second one is a G.Skill F4-3600C18D-16GTRS (8x8), for Intel processors. The AMD special one gives worse performance than my old 2400MHz, and the other one, almost the same as the 2400MHz one. I'm pretty confused. I thought that improving the performance was only a matter of speed, but it seems that there is something more. I'm not sure if it is the latency (C18D vs C15Q), the motherboard (MSI Meg X399 Creation) or the processor, but it is pretty weird that memory so fast give me results so close or even worse than the slower one.

[hitme987]

hey @timbergus are you running a XMP profile with 3200Mhz or more? Stock setting will run at 2133Mhz.  In My X399 Gaming Carbon AC for a 3200Mhz Ram it gives XMP1 as 2933 and XMP2 as 3200. Though for x399, you should manually tune the timings to achieve best settings. I was able to run 3133Mhz with 16-18-18-38 manually with my Hynix ripjaw 3200Mhz chip. 

And from what I have read, max speed people could run with 8 Rams in threadripper is 2933 Mhz (Correct me anyone if I am wrong). With 2 or 4 rams they could achieve upto 3600Mhz which is what I am looking at.

 

Also, your RAM series is 18D, which is rated as a dual channel RAM and you are mixing 4 kits, so it is 50/50 chance if it can run over 2933Mhz. (Which was my concern as well with F4-3600C15D-16GTZ as it is a dual channel RAM and not sure how high it can go with Quad channel). You can read more here: Don't mix Kits

Coming to latency, lesser response time is better performing in multi-core applications and in benchmarks as well, but in gaming its random as per what I've seen (depends on game). Though low latency is always preferable.

 

With these calculations the best timings for ryzen/threadripper is: 3600CL15 > 3200CL14 > 3600CL16

 

Response time you can calculate as below:

CL / frequency x 1000 = ns (response time)

some examples

• cl14 / 3200 x 1000 = 4.375ns

• cl15 / 3466 x 1000 = 4.328ns

• cl18 / 3600 x 1000 = 5ns

• cl19 / 4000 x 1000 = 4.75ns

 

[timbergus]

Thanks a lot for this info @hitme987. I will try to put my ideas in order next weekend and check the links you have passed. I have been talking with AMD and G.Skill, and I realized that the RAM I'm testing is not in MSI QVL list. And now, you tell me that is not even quad channel! What I was thinking :(

 

Right now, and following G.Skill instructions I'm doing some benchmarks of the modules in pairs, and I will do a comparison between my F4-2400C15Q-32GRK and the F4-3600C18D-16GTRS. The first one at 2400 MHz and the second one in a frequency range from 2400 MHz to 3600 MHz. I'm just looking for the best frequencies to include in the range.

 

For now, the F4-2400C15Q-32GRK is much, much better...

[hitme987]

@timbergus Don't worry about QVL list. They won't update it with many models. Always make sure to get Samsung B-die chips and they will run at high frequency. Refer to this list and check if your ram is a samsung B-die: https://www.reddit.com/r/Amd/comments/62vp2g/clearing_up_any_samsung_bdie_confusion_eg_on/

 

About your rams, mixing rams still makes it a quad channel and it will work at stock 2133Mhz. But there's 50/50 chance it will run at 3600Mhz. Manufacturers test and put up a 4 or 8 kit if you buy a single kit with 4 or 8 rams and it is guaranteed this way to run at 3600Mhz (again depends on timings and silicon lottery with ram).

 

Let me know your results as well. Try boosting to 3200+ which is a sweet spot for ryzen. I am still waiting for my ram, long wait

[timbergus]

Hi @hitme987. I'm still testing, but right now I'm assuming the following points.

1. RAM channels are handled by the processor. As I understand it, making proper use of the motherboard RAM channels is processors task, and depending on the memory quality, it will have a better performance or not.
2. Independent of the memory type, four DIMMs should work twice as fast in a quad channel capable processor.
3. The faster you make the memory work, and depending on the quality, the noisier could be the processor performance.

I have returned to Amazon the F4-3600C18D-16GTZRX and I kept the F4-3600C18D-16GTRS, because it was the better of both, for the next bunch of tests. For the moment I have the expected results. Comparing it against F4-2400C15Q-32GRK, using the TR2950x, a quad channel capable microprocessor, doubling the RAM in the proper DIMMs, doubles the bandwidth (in terms of microprocessor's performance through the 3DMark benchmarking tool score). I have taken 50 samples per test to avoid system processes noise.

But, when comparing both memories at the same speed, the 3.6GHz one is way much worse than the 2.4GHz one configured both at 2400MHz. I have also tried increasing the frequency on BIOS, and the result was pretty disappointing. The performance increase was ridiculous, and the signal noise grew a lot. I'm not sure if I've messed up with the BIOS configuration, or if this is just a terrible memory (the thick black line is the old memory). It's nice to see that the better signal for stability and performance is at 3200MHz, that matched the best frequency for AMD (the green one).

I did not even test the new memory at 3600MHz because when I managed to make Windows stop launching me blue screens, the test didn't launch. My system went completely nuts at this frequency.

 

Right now my only conclusion is that G.Skill's new memories are terrible 

 

But I'm just starting learning and testing hardware, so I can be wrong.

 

I have been thinking of buying this one: F4-3200C14Q-32GFX. And I'm buying it right now because it is 120€ less per pack in Amazon than the last time I visited  It is a 3200MHz quad channel, AMD C14 specific memory. So even better than my 2400MHz C15, in theory. I think this could be the right one to replace my old one. What do you think about it?

 

This is my last try because I'm afraid of being expelled from Amazon if I continue returning memories (sorry Amazon)  So if this one does not work either, I will stay with the old one and wait.

 

Hope you get your RAM soon @hitme987

[Princess Luna]

On 2/9/2019 at 2:52 PM, Stormseeker9 said:

Corsair vengeance 

You mean Value Select memory with a fancy but useless heat spreaded?

 

Any ways just get Samsung B-Dies from those 3600mhz G.Skill Trident Z low timing sticks.

[TahoeDust]

I have this set I am about to sell...

 

https://www.gskill.com/en/product/f4-3600c16q-32gtzr

 

Samsung B-Die 

[Stormseeker9]

4 hours ago, Princess Cadence said:

You mean Value Select memory with a fancy but useless heat spreaded?

 

Any ways just get Samsung B-Dies from those 3600mhz G.Skill Trident Z low timing sticks.

Vengeance LPX.. I meant. 

 

 

[timbergus]

8 hours ago, Princess Cadence said:

get Samsung B-Dies

Hi @Princess Cadence. Not sure if you know this tool, but seems pretty cool to me: https://benzhaomin.github.io/bdiefinder/. Is this the kind of B-Dies you are recommending? It seems that the F4-3200C14Q-32GFX appears in the list, and also in MSI MEG x399 Creation QVL.

 

 

So regardless of what the tests say when I get it, at least in theory, this seems the right memory.

[hitme987]

1 hour ago, timbergus said:

Hi @Princess Cadence. Not sure if you know this tool, but seems pretty cool to me: https://benzhaomin.github.io/bdiefinder/. Is this the kind of B-Dies you are recommending? It seems that the F4-3200C14Q-32GFX appears in the list, and also in MSI MEG x399 Creation QVL.

 

 

So regardless of what the tests say when I get it, at least in theory, this seems the right memory.

F4-3200C14Q-32GFX is a Samsung B-die. Flare X kits usually are well compatible with x399 platform so you should be good to go. Let us know once you test them!

[timbergus]

Will do. I receive the modules next week, so I will test them during the weekend. I hope to have good news with these ones.

[timbergus]

Well, finally I received the new RAM (G.Skill Flare X 3200MHz - F4-3200C14Q-32GFX) and did a few benchmarks. My best result has increased the processor performance an 11.19% over my old memory (G.Skill Ripjaws4 2400MHz - F4-2400C15Q-32GRK). The improvement over 2933MHz is pretty small, so not sure if buying a more expensive memory over 3000MHz worth it. I think I'm OK with 3200MHz. And 64GB over 32GB? Well, it improves, but it does not worth the price. Here is a plot with 32GB vs. 64GB and the different frequencies of the RAM. More memory, more stable in mid frequencies, but at full power, more or less the same. P means with the latency profile set in BIOS, and auto, as you have it in BIOS without configuration, all compared with the old memory (0.00% point). Hope you find this useful

 

[hitme987]

Same here. I received my  F4-3600C16Q-32GTZKK 4x8GB kit today and the results are very good with quad channel. Able to run at 3333Mhz CL14 14-14-34 stable. Was booting with 3466Mhz but had some BSOD's. 3600Mhz wasn't booting at all.

 

I haven't benchmarked it with any specific tool, but I have captured gaming benchmarks with Hitman2 and Shadow of the tomb raider. I am running everything at ultra and 3440x1440 resolution and for SOTTR I am using MSAAx4 Anti-Aliasing.

 

With Dual Channel 2x16GB Ripjaws v-series at 3133Mhz CL16:

 

1) Hitman2 Benchmark:  Min of 6.73 FPS, Max of 88.24 FPS, Avg: 51.37 FPS. Score: 53.13FPS.

Screenshot: https://ibb.co/D76vDFQ

 

2) SOTTR Benchmark: Frames rendered: 8456, 99% GPU Bound, Avg fps: 53

Screenshot: https://ibb.co/qWZgxLP

 

 

 

With Quad Channel 4x8GB Tridentz F4-3600C16Q-32GTZKK at 3333Mhz CL14:

 

1) Hitman2 Benchmark:  Min of 17.78 FPS, Max of 119.65 FPS, Avg: 73.41 FPS. Score: 72.06FPS

Screenshot: https://ibb.co/9Z5v7qz

 

2) SOTTR Benchmark: Frames rendered: 8731, 100% GPU Bound, Avg fps: 55

Screenshot: https://ibb.co/vdP3pjB

 

Though the benchmarks only show one side of the story. In gaming, I am seeing no stuttering now in Hitman2. Min FPS was 32 in crowded Mumbai areas, now Min FPS is 45 and very much playable. Avg FPS increased over 20-30fps and around 60fps most of the times. Max FPS crossing 100 most of the times, was max of 80FPS or so before.  

Also observed better frame rates with creator mode than game mode with quad channel. Although gaming isn't threadripper piece of cake, it is still a very competent gaming CPU. I have seen similar benchmarks with 8700k as well for hitman2.

 

Very much satisfied with my threadripper gaming performance now. 

 

 

 

[nick name]

Damn, that helped a lot.  Well done.  

[hitme987]

22 hours ago, nick name said:

Damn, that helped a lot.  Well done.  

Thanks. Appreciate your initial inputs as well. I initially was going for 3600Mhz CL15 ram but it was dual sticks. Didn't want to mix up 2 kits and hassle of returns if it doesn't work out.

Also, I had a Geekbench4 benchmark result of my previous setup, did a run with my new ram and below are the results.

 

With Dual Channel 2x16GB Ripjaws v-series at 3133Mhz CL16:

Single-Core Score: 4670 & Multi-Core Score: 27515

https://browser.geekbench.com/v4/cpu/11849701

With Quad Channel 4x8GB Tridentz F4-3600C16Q-32GTZKK at 3333Mhz CL14:

Single-Core Score: 4973 & Multi-Core Score: 39108

https://browser.geekbench.com/v4/cpu/12478578

 

[nick name]

1 hour ago, hitme987 said:

Thanks. Appreciate your initial inputs as well. I initially was going for 3600Mhz CL15 ram but it was dual sticks. Didn't want to mix up 2 kits and hassle of returns if it doesn't work out.

Also, I had a Geekbench4 benchmark result of my previous setup, did a run with my new ram and below are the results.

 

With Dual Channel 2x16GB Ripjaws v-series at 3133Mhz CL16:

Single-Core Score: 4670 & Multi-Core Score: 27515

https://browser.geekbench.com/v4/cpu/11849701

With Quad Channel 4x8GB Tridentz F4-3600C16Q-32GTZKK at 3333Mhz CL14:

Single-Core Score: 4973 & Multi-Core Score: 39108

https://browser.geekbench.com/v4/cpu/12478578

 

Someone mentioned a 3600CL15 32GB kit for sale in Europe, but I haven't seen any available in the US, but I think you probably would have ended up where you are or damn close anyway.