Buying Ryzen? Confused about ram? Read here

[He_162]

Keep in mind the money saved on getting cheaper ram is rewarded by buying a better GPU, you get much better frames with a more powerful GPU than you do worse ram, so ram really isn't important for budget Ryzen builds. Skip it and put that towards a GPU.

If you already have a GPU, consider saving money from ram and putting that towards a better cooler, because a 4ghz overclock (or close) is better than improved ram speeds for gaming performance.

As for getting fast ram, it's not worth it just for a unnoticeable bump in performance, and if you want 240+hz, Ryzen can do it, but you should get a i7-7700k instead, or at the very least an i5-7600k.

When i5's are 6 cores, and i7's are 6 cores (Coffee Lake), Ryzen will have no place, so AMD better do good with Zen 2, or a Ryzen refresh with higher clocks. Whichever route they go.

Intel's gonna be pushing to improve as much as they can before 2020 to keep 7nm Zen 3 chips out of the lead.

[Botlike]

In case that's a reply to my questions: I already made my decision to buy Ryzen. I'm not building an entire new rig, just upgrading my current one. I already have a GPU and a CPU cooler I can still use.

I'm not on a fixed budget or something. Just wanna have the best possible performance and don't wanna spend a fortune on DIMMs which may get outperformed by cheaper ones.  

[stealth80]

2 hours ago, pajpaj said:

Is this a Samsung B die ? F4-3200C16D-16GVKB, thanks.

Doubtful as it is CAS 16 3200mhz

 

23 minutes ago, Botlike said:

 

1. G.Skill uses different dies for their DIMMs. For best performance, I still should go for something with Samsung B-Dies.
2. Using only 2 DIMMs puts less pressure to the IMC, so I should go for 2x16 instead of 4x8.
3. 16GB sticks are dual ranked - giving better performance than 8GB single ranked would do. (Which results also in less pressure to IMC but still better performance when only comparing those 2 scenarios with 32GB)
4. I have a few kits to choose from, priced almost the same: 3200 CL14, 3466 CL16, 3600 CL17 - I should go for the highest clock and do not care much about timings in the first place.
5. Do I really have to care about speeds and timings anyways since I have to fine-tune? Or should I just go for the cheapest one with B-Dies and tune myself?

1. Pretty much, although some specific versions of corsair LPX use b die also

 

3200 Samsung b die

 

https://pcpartpicker.com/products/memory/#s=403200&X=0,127520&L=0,140

 

3600 Samsung B die

 

https://pcpartpicker.com/products/memory/#X=0,127520&L=0,160&s=403600

 

2. I would agree yes

 

3. I don't agree with this theory. All I have seen thus far is that people using double sided ram tend to have issues getting above 2933mhz - I have only seen 1 german website (as linked earlier) with any testing done to back up this theory - if it is true/I see more results to back it up then I will change the guide to reflect it. There seems to be quite a strong belief in this on reddit (which isn't something I use tbf) and some of tests they link to do show a slight improvement of say 2400 DR against 2667 SR - however these tests were done some time ago prior to AGESA 1006 and were always behind 3200 SR - Now if you can get and run DR 3200 CAS 14 / DR 3600 CAS 16 you "may" get it to run at higher than those test speeds and get some additional performance - but this is at the risk of been stuck at 2667mhz/2933mhz.

 

4. Of the 3 kits listed I would only be sure that the CL14 3200 is B die - however, this isn't all too important now if you are willing to play with voltages and timings. If you just want to have the lowest hassle get the C14 3200. If you want to have the headroom to go higher and don't mind playing with settings, get the fastest ram (3600 C17)

 

5. see 4 

 

 

[Botlike]

According to the thread on reddit you mentioned earlier, the 3600 CL17 (F4-3600C17D-32GTZR) I mentioned before is b-die as well (no information about the 3466 CL16). But since RAM clock is more important than latency, I'll go with the 3600 and try to get it as high as I can.

 

Thank you so much for this thread and your reply. It helped a lot. 

[pajpaj]

1 hour ago, stealth80 said:

Doubtful as it is CAS 16 3200mhz

But will it work? Though it is not mentioned in the QVL of Asrock AB350M PRO 4.

[stealth80]

2 hours ago, pajpaj said:

But will it work? Though it is not mentioned in the QVL of Asrock AB350M PRO 4.

Should do, check the guide I covered non qvl

[stealth80]

3 hours ago, Botlike said:

According to the thread on reddit you mentioned earlier, the 3600 CL17 (F4-3600C17D-32GTZR) I mentioned before is b-die as well (no information about the 3466 CL16). But since RAM clock is more important than latency, I'll go with the 3600 and try to get it as high as I can.

 

Thank you so much for this thread and your reply. It helped a lot. 

No problem 

[pajpaj]

1 hour ago, stealth80 said:

Should do, check the guide I covered non qvl

Okay, will re-read it, thank you.

[Droidbot]

On 04/06/2017 at 1:50 PM, tim11111111 said:

 

Micron  die chips are also getting worked on.

On some of the cheapest Crucial 2133mhz sticks, some people have Overlocked the memory(tweaking timing, voltage..) to 2933mhz which is a really good speed for ryzen albeit not the best (3666-4000mhz).

Here's and example of b350 tomahawk with ~30$ 4gb crucial ram OCed to 2933 on an april realeased  BIOS https://forum-en.msi.com/index.php?topic=285450.0

The ram is this one @ newegg https://www.newegg.com/Product/Product.aspx?Item=9SIAD7H5EX2061

Whats your take with other RAM manufacturer other than Samsung? Do you think New agesa will allow good support to OC cheap ram like this without fancy "gamer" bling?

This board (asrock pro4) also also List crucial ECC memory on the QVL which is interesting to see on B350's.

Also is there any diminishing returns in your experience after going past 3200mhz?

Crucial and Hynix chips hit 2933-3200 max. 

 

Samsung non-B die hit 2933 max. 

 

Samsung B-Die hit 3600 max. 

 

More ram speed is good, no matter how much. 

[WhiteWulfe]

https://pcpartpicker.com/user/WhiteWulfe/saved/

 

That's what I'll be getting for my build. Does anyone know if that MOBO supports 3200mhz RAM atm? Can't find anything on it. One thing about the build too....Still not sure on the damn cpu cooler, so any suggestions would help, that don't break the bank. Thanks.

[stealth80]

9 hours ago, WhiteWulfe said:

https://pcpartpicker.com/user/WhiteWulfe/saved/

 

That's what I'll be getting for my build. Does anyone know if that MOBO supports 3200mhz RAM atm? Can't find anything on it. One thing about the build too....Still not sure on the damn cpu cooler, so any suggestions would help, that don't break the bank. Thanks.

https://www.msi.com/Motherboard/support/X370-GAMING-PLUS.html#support-mem-2

 

pick one that is tested to 3200 from that list to have hassle free building. You basically need Samsung B die like the guide suggests. For the cooler, I wouldn't grab that, i'd spend the extra on AIO on get a Noctua

[WhiteWulfe]

7 hours ago, stealth80 said:

https://www.msi.com/Motherboard/support/X370-GAMING-PLUS.html#support-mem-2

 

pick one that is tested to 3200 from that list to have hassle free building. You basically need Samsung B die like the guide suggests. For the cooler, I wouldn't grab that, i'd spend the extra on AIO on get a Noctua

Im not understanding the Samsung B die? Little confused by the term...

[stealth80]

4 hours ago, WhiteWulfe said:

Im not understanding the Samsung B die? Little confused by the term...

read the guide 

[Carclis]

You should probably update the main post with some of the new info (not very new anymore) just to help inform people of what is most important with regards to choosing a DDR4 kit or optimising it for performance.

Take some of the info from this post by AMD.

Quote

Conclusion #1: Dual rank DIMMs (yellow) offered the best performance amongst “set and forget” (light blue, orange, yellow) memory configured automatically by XMP profiles.

Conclusion #1a: But the increased overclocking headroom of single rank modules was more than enough to overpower the benefits of rank interleaving, so manually-tuned single rank DDR4-3200 and 3466 won the day (dark blue and green).

Conclusion #2: BankGroupSwap should likely be disabled for users that want the best PC gaming performance. As always, test your specific use case.

Conclusion #3: Chasing the highest possible clockspeed required timings so relaxed that real world performance suffered versus lower frequencies with tighter timings. This is a fine balance, however, so testing on your platform is always helpful.

Conclusion #4: Geardown Mode should likely be disabled if your overclock is stable with a 1T command rate. As always, test your specific use case.

So basically Ryzen performance is linked to DDR4 latency which is exactly the case on Intel as well, although Ryzen benefits much more greatly. This means that CAS latency and timings in general are equally as important as frequency and should not be sacrificed for tiny gains in frequency.

[stealth80]

2 hours ago, Carclis said:

You should probably update the main post with some of the new info (not very new anymore) just to help inform people of what is most important with regards to choosing a DDR4 kit or optimising it for performance.

Take some of the info from this post by AMD.

Yes I will include that, keep in mind that this post was to aid people with ram buying and getting it to run, I didn't really want to start making it into an optimisation thread as peoples results will vary, which brings me onto my next point:

 

2 hours ago, Carclis said:

So basically Ryzen performance is linked to DDR4 latency which is exactly the case on Intel as well, although Ryzen benefits much more greatly. This means that CAS latency and timings in general are equally as important as frequency and should not be sacrificed for tiny gains in frequency.

Be VERY careful with statements like this - whilst you are correct, you are also wrong. At 3200mhz and above, there is very little to choose between latency and frequency, however between say 2133mhz and 2933mhz, the jump is insane and frequency should be chosen in that range above all. Infact in those very tests, the best results seems to be the 3466mhz + 1T running CAS 14 rather than 3200mhz CAS 12, which goes against what you stated. See below, the biggest effect seems to "bank group swapping and 1T" rather than CAS

 

 

[MageTank]

13 hours ago, stealth80 said:

Yes I will include that, keep in mind that this post was to aid people with ram buying and getting it to run, I didn't really want to start making it into an optimisation thread as peoples results will vary, which brings me onto my next point:

 

Be VERY careful with statements like this - whilst you are correct, you are also wrong. At 3200mhz and above, there is very little to choose between latency and frequency, however between say 2133mhz and 2933mhz, the jump is insane and frequency should be chosen in that range above all. Infact in those very tests, the best results seems to be the 3466mhz + 1T running CAS 14 rather than 3200mhz CAS 12, which goes against what you stated. See below, the biggest effect seems to "bank group swapping and 1T" rather than CAS

 

 

 

 

He is not wrong, you are wrong in thinking that CAS is the only factor in latency. It's not. Plenty of tertiary timings govern latency, some to a far greater extent. The correct attribution to memory performance is not frequency or latency, but rather, bandwidth efficiency. Bandwidth efficiency is what dictates how much of your peak theoretical bandwidth is actually achieved under any given circumstance. I have tested this extensively on Ryzen, and those tests above from AMD also showcase exactly what I am talking about when you see "tweaked sub-timings" in their results. Essentially, just dialing in clocks + primary timings mean absolutely nothing if you are only getting 80% of your bandwidth instead of say, 95% of it.

 

Here is some quick math. DDR4 3200 has a peak theoretical memory bandwidth of 51.2GB/s. If your bandwidth efficiency is only 80%, you will only achieve 80% of that peak theoretical number, which is 40.96GB/s. If you tweak your tertiary timings, and achieve say, 95% efficiency, you would getting 48.64GB/s. Quite a significant difference in bandwidth, even though primary timings and frequency are identical.

 

Now, latency. Primary timings offer a pretty decent reduction in latency as they are tightened, but that is only part of the equation. Your tREFI (a timing you do not have access to on Ryzen), tRFC (a timing you do have access to), along with tertiary timings, make up just as much an impact on overall memory latency in the grand scheme of things. Remember, latency is measured in clock cycles. The faster you complete that cycle, the lower your overall latency. By improving your bandwidth efficiency, you are completing the clock cycles faster, your latency is being reduced as a result. 

 

The "best" results, are not black and white. It will differ per IC used, per rank/bank count, and even per IMC. You have you experiment with different kits, CPU's and boards to determine what is best with your particular setup. What is best for you, may not be best for someone else either, as quality of IC's, IMC's and even board trace topology differs dramatically, even on the exact same pieces of hardware.

 

I would say, up to about 3600, you can achieve decent bandwidth efficiency without compromising on latency to any great deal. Anything beyond that however, is likely the point of diminished returns. The few 3866 results I've seen, were actually slower than the 3600 results, even with nearly identical primary timings. This is because as you increase frequency, you increase the stress on the IMC. In order to reduce that stress, people often sacrifice tertiary timings, which in turn, sacrifices that very efficiency I spoke of earlier. 3600 C16 at 95% efficiency is faster than 3866 C16 at say, 85%. 

[stealth80]

4 minutes ago, MageTank said:

snip

As I said, I didn't want it to turn into an optimisation thread, it is to clear up issues with getting ram to work at it's rated speed on ryzen. I also stated that he is right and wrong, by saying what he did, he could in fact mislead people. If someone was to think that 2133 at cas 14 is better than 2933 at cas 16 (for example) they would be making a huge mistake on ryzen. I'm not disagreeing with what you are saying reference variables in all ram timings,  it is in response to that thread, the best result was in fact 3466mhz using sr, and 1t with cas 14 rather than cas 12. They didn't test all the other timings available, I'm guessing cause it would take an absolute eternity!

It isn't something I want to be used in that thread, it is a drop point for people who are buying ryzen and want to know how to get their ram to work. If they want to tweak at those levels, they wouldn't need the guide in the first place. ?

 

What have your experiences been with trc on ryzen? I have a strange issue where my ram is completely unstable if I set it to 54 (16 + 38) and can only make it stable at around 62. Interestingly the board always wants to make it load at 75, with the other timings as 16-18-18-18 which doesn't make much sense

[MageTank]

18 minutes ago, stealth80 said:

As I said, I didn't want it to turn into an optimisation thread, it is to clear up issues with getting ram to work at it's rated speed on ryzen. I also stated that he is right and wrong, by saying what he did, he could in fact mislead people. If someone was to think that 2133 at cas 14 is better than 2933 at cas 16 (for example) they would be making a huge mistake on ryzen. I'm not disagreeing with what you are saying reference variables in all ram timings,  it is in response to that thread, the best result was in fact 3466mhz using sr, and 1t with cas 14 rather than cas 12. They didn't test all the other timings available, I'm guessing cause it would take an absolute eternity!

It isn't something I want to be used in that thread, it is a drop point for people who are buying ryzen and want to know how to get their ram to work. If they want to tweak at those levels, they wouldn't need the guide in the first place. ?

 

What have your experiences been with trc on ryzen? I have a strange issue where my ram is completely unstable if I set it to 54 (16 + 38) and can only make it stable at around 62. Interestingly the board always wants to make it load at 75, with the other timings as 16-18-18-18 which doesn't make much sense

Hmm, odd that you are running into instability at 54. You shouldn't be having any issues as you are not closing the row before you get the data you need (as designated by tRC = tRAS + tRP). Are you certain this is the only timing causing potential instability? Stability testing for ram is a bit complicated, as it is difficult to stress every address at every column. Memtest86 isn't a stress test, and you need a blend of several different stress tests at various ram sizes to actually determine stability. I would leave it loose, and double check to make sure that your ram is indeed stable in it's current configuration.

 

That being said, you won't really find any noticeable performance boost by tightening tRC. If anyone is claiming any boost in gaming from tightening tRC, they are lying to you. Even in synthetic tests, you will be hard pressed to find a boost from it. 

[stealth80]

4 minutes ago, MageTank said:

Hmm, odd that you are running into instability at 54. You shouldn't be having any issues as you are not closing the row before you get the data you need (as designated by tRC = tRAS + tRP). Are you certain this is the only timing causing potential instability? Stability testing for ram is a bit complicated, as it is difficult to stress every address at every column. Memtest86 isn't a stress test, and you need a blend of several different stress tests at various ram sizes to actually determine stability. I would leave it loose, and double check to make sure that your ram is indeed stable in it's current configuration.

 

That being said, you won't really find any noticeable performance boost by tightening tRC. If anyone is claiming any boost in gaming from tightening tRC, they are lying to you. Even in synthetic tests, you will be hard pressed to find a boost from it. 

My timings right now are 16-18-18-18-62-1t, ddrv 1.41v xmp disabled, gear down and bank swap disabled. This has got me the best fps thus far, if I lower trc I get a boot loop, this also my second set of ram I've had issues with. Non b die 

[You_are_a_cunt]

Right, so here's a bit of an ear scratcher from me:

When I built my Ryzen system I got a single Corsair XMP kit (2x4gb) sku: CMK8GX4M2B3000C15

Right now it's damn near impossible to get my hands on one and wherever they're available to order they're about 100 EUR (25 EUR more than I paid for it a couple months ago). I'm not really willing to pay that much.

Here's my question: Can I safely mix and match with another stick or kit running the same CAS and the same timings? DDR4 is fairly new to me and I have little to no experience with mixing manufacturers on it.

The closest I could find in therms of price I'm willing to pay and specs is the Kingston HyperX Predator (sku: HX430C15PB3/8), which is an 8 gig stick (as opposed to the 2x4 I have). I ran a similar config on FX and I know it ran on that, but how will Ryzen (specifically A-XMP) handle running 8+4+4 from different manufacturers at 2933MHz?

 

Specs for each of the SKUs:

https://www.kingston.com/dataSheets/HX430C15PB3_8.pdf

http://www.corsair.com/en-gb/vengeance-lpx-8gb-2x4gb-ddr4-dram-3000mhz-c15-memory-kit-black-cmk8gx4m2b3000c15

 

From what I can tell, XMP timings for 3000 (2933) are both CL15-17-17 (with Corsair giving the 4th value of 35)

 

The price difference is about 25 EUR, with the HyperX being the cheaper.

 

Will this work properly at 2933MHz or am I looking at spending the extra dime for a matching kit?

[stealth80]

42 minutes ago, revsilverspine said:

Right, so here's a bit of an ear scratcher from me:

When I built my Ryzen system I got a single Corsair XMP kit (2x4gb) sku: CMK8GX4M2B3000C15

Right now it's damn near impossible to get my hands on one and wherever they're available to order they're about 100 EUR (25 EUR more than I paid for it a couple months ago). I'm not really willing to pay that much.

Here's my question: Can I safely mix and match with another stick or kit running the same CAS and the same timings? DDR4 is fairly new to me and I have little to no experience with mixing manufacturers on it.

The closest I could find in therms of price I'm willing to pay and specs is the Kingston HyperX Predator (sku: HX430C15PB3/8), which is an 8 gig stick (as opposed to the 2x4 I have). I ran a similar config on FX and I know it ran on that, but how will Ryzen (specifically A-XMP) handle running 8+4+4 from different manufacturers at 2933MHz?

 

Specs for each of the SKUs:

https://www.kingston.com/dataSheets/HX430C15PB3_8.pdf

http://www.corsair.com/en-gb/vengeance-lpx-8gb-2x4gb-ddr4-dram-3000mhz-c15-memory-kit-black-cmk8gx4m2b3000c15

 

From what I can tell, XMP timings for 3000 (2933) are both CL15-17-17 (with Corsair giving the 4th value of 35)

 

The price difference is about 25 EUR, with the HyperX being the cheaper.

 

Will this work properly at 2933MHz or am I looking at spending the extra dime for a matching kit?

It will work, but to what degree I can't say without testing myself. The price increase has happened to all Ram, including GDDR for GPU's. The price increases are going to keep going until the supply situation is resolved. If it was me, I would sell your 2x4gb sticks and buy a 16gb set - This way you will have no issues with compatibility, and you will be ables to run the sticks at their rated speeds - 4 sticks causes issues on Ryzen - although, my board just got another "increased DDR compatibility Bios" so it may get better. 

[You_are_a_cunt]

29 minutes ago, stealth80 said:

It will work, but to what degree I can't say without testing myself. The price increase has happened to all Ram, including GDDR for GPU's. The price increases are going to keep going until the supply situation is resolved. If it was me, I would sell your 2x4gb sticks and buy a 16gb set - This way you will have no issues with compatibility, and you will be ables to run the sticks at their rated speeds - 4 sticks causes issues on Ryzen - although, my board just got another "increased DDR compatibility Bios" so it may get better. 

Yeah, I thought about getting the single 8 gig stick, selling my 2x4 kit and then buying another 8 gig stick.

Right now I'm still in the debating phase of buying more ram. Do I really need it? Do I really want to spend another ~100 eur just in case I ever need more than 8 gigs of ram (which I might not, given my current usage).

It's mostly a haze but I think I'd be better off just waiting for prices to go back down to almost normal.

[GReM]

Searching all day for info on ram in general, and that overclocks with ryzen, you are one helpful bro. I appreciate your efforts buddy

[stealth80]

3 hours ago, GReM said:

Searching all day for info on ram in general, and that overclocks with ryzen, you are one helpful bro. I appreciate your efforts buddy

No problem, this is exactly why I did the post 

[JohnStone]

Hello there!
I'm about to get my new pc with ryzen 5 1600 and ga-ab350 gaming 3 motherboard,but there's still something comfuses me about ram..
Firstly i was about to buy this one HX424C15FB2/8 ( single sided on 2400mhz ) which it was clearly a bad decision after i found out that ryzen could work properly on 2667mhz.
After that,i've searched to buy some single sided of 2667mhz but i found out that on this motherboard there are only 4gb sticks compatible..
So,the question now is,if i would buy dual sided of 1x8gb on 2667mhz (native 2667),would it still work or not at all since the memory configuration graphic from amd doesn't say something about it?
If yes,would i be able to install 2x4gb singe sided and 2x8gb dual sided on 2667mhz (native 2667) and work perfectly? ( which i doubt about it,it's more likely both of dual sided not working at all since it's native is on a higher frequency of the standard ).
And if not,what is the next best option for me? 1x8gb of 2400mhz on dual sided? ( by the way,i'm not thinking about to overclock it at all ).
You really doing great job here!!!!