AMD RYZEN Builders Thread

[richierich1212]

This means my F4-3200C15D-16GTZ with 15-15-15-35 timings is not samsung b-die?
This RAM kits seems to be working well for me though at 3200 MT/s.

No, you have Samsung B-Die sticks. Which is probably why you're able to run them at their rated 3200 speeds.

 

[DrMrLordX]

I'm just unsure which RAM to go with it for good performance any recommendations. RGB would be nice or options of colours.

Trident-Z if you can afford it. I can't guarantee these kits but if you are looking for 2x8GB:

http://pcpartpicker.com/products/me...4000,404133,404200,404266&Z=16384002&sort=a10

Start with those. Anything DDR4-3400 and higher SHOULD be Samsung B-die single-rank. After that it's just a matter of finding how much voltage you need to get where you want to go. Bear in mind that if you are buying G.Skill that the Trident-Z kits of the same speed/timings tend to be better than the equivalent Ripjaws. You might find some dual-rank in the Ripjaws even at these speeds.

 

[Markfw]

I still have to see any samsung b-die OVER cas 14 that will do 3200 in any motherboard, please correct me if I am wrong. I have 2 sets of the cas 14, in 2 different motherboards, that both do 3200.

 

[IEC]

Trident-Z if you can afford it. I can't guarantee these kits but if you are looking for 2x8GB:

http://pcpartpicker.com/products/me...4000,404133,404200,404266&Z=16384002&sort=a10

Start with those. Anything DDR4-3400 and higher SHOULD be Samsung B-die single-rank. After that it's just a matter of finding how much voltage you need to get where you want to go. Bear in mind that if you are buying G.Skill that the Trident-Z kits of the same speed/timings tend to be better than the equivalent Ripjaws. You might find some dual-rank in the Ripjaws even at these speeds.

This is not true. I have a set of 2x8GB DDR4-3400 CL16 TridentZ which is not B-die, and is dual rank. 3200 CL14 is guaranteed B-die, as it is the only IC capable of that speed/latency combination. 3600 with 16-16-16-16-3x timings should also be B-die. Anything with asymmetric timings such as 16-18-18-18-39 is NOT B-die.

 

[Malogeek]

Anything with asymmetric timings such as 16-18-18-18-39 is NOT B-die.

I don't think there are any rules whatsoever in what G.skill are Samsung b-die

We have internally observed good results from 2933, 3200, and 3500 MT/s rates with 16GB kits based on Samsung “B-die” memory chips. Potential kits include:

• G.Skill Trident Z - F4-3200C16D-16GTZR [16-18-18-36 @ 1.35v]

 

[DrMrLordX]

This is not true. I have a set of 2x8GB DDR4-3400 CL16 TridentZ which is not B-die, and is dual rank. 3200 CL14 is guaranteed B-die, as it is the only IC capable of that speed/latency combination. 3600 with 16-16-16-16-3x timings should also be B-die. Anything with asymmetric timings such as 16-18-18-18-39 is NOT B-die.

Didn't know they were shipping any 3400 non-B-die, but my DDR4-3733 17-19-19-42 is asymmetric and it is B-die.

 

[IEC]

Didn't know they were shipping any 3400 non-B-die, but my DDR4-3733 17-19-19-42 is asymmetric and it is B-die.

Well that throws a wrinkle into things, then. I suppose the only way to know for sure is to remove the heatspreader (e.g. my DDR4-3400 sticks are "E" die)

 

[EXCellR8]

looking at the replacement board's QVL and I feel like half of the memory isn't even on the market yet...

 

[DrMrLordX]

Well that throws a wrinkle into things, then. I suppose the only way to know for sure is to remove the heatspreader (e.g. my DDR4-3400 sticks are "E" die)

Or use Thaiphoon Burner. Here's the output for one of my DIMMs:

https://www.dropbox.com/s/g6q6uzwvg3az3yt/thaiphoon.png?dl=0

 

[CuriousMike]

I still have to see any samsung b-die OVER cas 14 that will do 3200 in any motherboard, please correct me if I am wrong. I have 2 sets of the cas 14, in 2 different motherboards, that both do 3200.

Then that might explain why I can't hit 3200.

 

[Markfw]

Then that might explain why I can't hit 3200.

Yes, it appears at the moment with current motherboards and BIOS, the Ryzen CPU is very picky on memory. It will do rated speed only if the CPU "likes" that memory, and this is the only one I read that it really "likes".

I could be wrong, as I said, I just have not read anything other than 3200 samsung b-die, cas 14 that actually does 3200.

 

[Crono]

OCed the 1600, running stable after stress tests and currently running benchmarks. Idle temp is low 40s, load is 77 (using AIDA64) after a few minutes with spikes up to low 80s before fan curve cranks up the fans (side note: it's a simple thing, but it's really nice to not have to deal with the stupid 20C offset that the 1700X has).

It's possible I can squeeze out more clock speed, but I'll wait for the additional cooling parts to come in. I'm comfortable with this clock speed and voltage if I can't, though, and the current cooling (stock Wraith HSF, 3x Arctic F12 intake, BeQuiet! PureWings 120mm exhaust, Noctua NT-H1 paste) is doing a better job than I thought it would.

Got my 16GB Corsair Vengeance LPX DDR4-3200 running at 2933 MHz, which is perfectly fine by me for now as my R7 system's RAM is at 2666, and it wasn't a crazy expensive kit. I'll wait for the April/May updates before worrying about hitting advertised RAM speed.

 

[DrMrLordX]

I could be wrong, as I said, I just have not read anything other than 3200 samsung b-die, cas 14 that actually does 3200.

Mine will do it, which is interesting since it should not be able to do so well according to its bin. It should be restricted to CL/CAS15 (16 on Ryzen) but it does 14 anyway.

 

[CuriousMike]

Is there any real world benefit to setting my memory from its default 16-18-18-36 to 14-14-14-32?
Changing the timings in the BIOS was simple, and a quick 15 minutes of Prime95 was stable.
(I tried these timings per Marks suggestion that CAS 14 seemed to be a strong indicator for hitting 3200mhz; alas, these timings didn't help me hit 3200)

 

[Markfw]

Is there any real world benefit to setting my memory from its default 16-18-18-36 to 14-14-14-32?
Changing the timings in the BIOS was simple, and a quick 15 minutes of Prime95 was stable.
(I tried these timings per Marks suggestion that CAS 14 seemed to be a strong indicator for hitting 3200mhz; alas, these timings didn't help me hit 3200)

The memory has to be capable of doing those timings. Apparently yours isn't rated for that.

 

[CuriousMike]

It isn't rated for it, but it is running at those timings.

I just don't know if there is a real world performance benefit to running at these lower timings.

 

[imported_jjj]

There are non-Samsung B-die on QVL lists at 3200 and in theory you can push lots of kits to 3200 with sufficiently relaxed timings.

One odd thing is that , for a reason i can't explain, populating all 4 DIMM slots might result in higher perf in actual apps...

 

[imported_jjj]

It isn't rated for it, but it is running at those timings.

I just don't know if there is a real world performance benefit to running at these lower timings.

Check memory latency in AIDA64, the diff is pretty relevant., maybe at about 4ns.

Edit: found an example

Some real world tests with those configs http://www.hardware.fr/articles/958-3/applications-7-zip-winrar.html

 

[Dresdenboy]

There are non-Samsung B-die on QVL lists at 3200 and in theory you can push lots of kits to 3200 with sufficiently relaxed timings.

One odd thing is that , for a reason i can't explain, populating all 4 DIMM slots might result in higher perf in actual apps...

I assume this is related to open mem banks and bank or rank internal roundtrips abd timings.

 

[imported_jjj]

I assume this is related to open mem banks and bank or rank internal roundtrips abd timings.

So an app like AIDA would show a best case scenario but doesn't have much to do with real world perf?
This is worth investigating.

Does seem to be related to latency and apps that use a lot of memory.

 

[Crumpet]

Had a bit of a nightmare with this one.

My motherboard only came in stock Saturday, so i've been sat twiddling my thumbs for a fair while now ready to go.

So when I get the message I practically run up to the store to grab the mobo, and rush home to get cracking.

Strip the old system down, format the SSD's ready for a clean install, and start fiddling about to get the cable management right, I want this to look good after all.
So I mount the fans and the radiator on the front of the case (which is fiddly as all bloody hell) and then decide that I'd prefer the tubing to come in from a different direction..

Rewind, try again, much fiddling, many swearing, a lot of time later and the rad is back in, the fans are on and wired in, can get cracking on the actual pump/block.

And that's when it hits me... NZXT still haven't shipped me the *!%£#$@ AM4 bracket to fit the X62 kraken. Oh you mother.... BREATHE. COFFEE BREAK. BREATHE.

Thankfully I remembered I'd gotten my hands on a Wraith cooler off Ebay (for far too much :S bloody scalpers) so all i'd really lost was time and patience. So I've fitted the Wraith cooler and i've left the X62 Kraken installed, and left the block loose inside my case where it won't move far (with plastic protection on it).

Nevermind, we carry on. Finish the build, put the GPU in and wouldn't you know, dip$%@! forgot to plug the VGA connector into the PSU.. HUNNNNNGHHHH. Okay grab headtorch.. HNNNNGHHHGHHGHGHG headtorch batteries are dead. FINE, grab phone, hold phone in mouth half blinding myself to use it as a torch and get my arm stuffed so far up into this H440 that I feel like i'm violating a cow. Plug it in, move on. BREATHE.

Everything wired up. All good. Hit the power button and... HOLY CRAP IT ACTUALLY BOOTED. Normally something should have gone wrong at this point, but no.. It just works.. It's a miracle.

Aaaaand then I go to test out a game on steam and my brand new Corsair RM1000X has so much coil whine that I can't even hear the game music through my 5.1 surround system.

F. M. L

Going to wait until Vega release to see if the coil whine is just related to this specific GPU as the R9 290 is a thirsty mare. Managed to roughly halve the coil whine through power tweaking and bios updates, so it's livable with at this point, but we'll see...

So that's my rant so far.. Just got to keep waiting...And keep breathing...

 

[Dresdenboy]

So an app like AIDA would show a best case scenario but doesn't have much to do with real world perf?
This is worth investigating.

Does seem to be related to latency and apps that use a lot of memory.

AIDA & Co. just deliver results for specific cases/patterns (often the two extremes: linear and random).

The mentioned apps which use a lot of memory would just trigger some delaying conditions more often. Here's an nice read on the topic:
http://www.anandtech.com/show/3851/...ow-about-sdram-memory-but-were-afraid-to-ask/
or alternatively one of the many lecture slide decks:
http://www.ece.cmu.edu/~ece740/f11/...ures:onur-740-fall11-lecture25-mainmemory.pdf

 

[imported_jjj]

AIDA & Co. just deliver results for specific cases/patterns (often the two extremes: linear and random).

The mentioned apps which use a lot of memory would just trigger some delaying conditions more often. Here's an nice readup on the topic:
http://www.anandtech.com/show/3851/...ow-about-sdram-memory-but-were-afraid-to-ask/
or alternatively one of the many lecture slide decks:
http://www.ece.cmu.edu/~ece740/f11/...ures:onur-740-fall11-lecture25-mainmemory.pdf

Never thought about the matter before at all, sounds stupid but it is what it is.

Now i assume it's mostly about more open pages with dual rank or 2 DIMMs per channel- better chance for a hit. Guess interleaved accesses can boost perf too, assuming that's being done.
There should be some extra complication for the memory controller with dual rank (or 2 DIMMs per chan) but manageable.

Mentioned AIDA because folks look at BW and latency and often that's it. Here's an example, a review that tests with AIDA and nothing more http://www.legitreviews.com/amd-ryzen-single-rank-versus-dual-rank-ddr4-memory-performance_192960
And in this case, the benchmark fails to expose the real world differences so it becomes useless for this particular purpose.
The perf gaps must be investigated and quantified. HardwareFR gets pretty large differences , more than enough to impact purchasing decisions.

 

[guachi]

OCed the 1600, running stable after stress tests and currently running benchmarks. Idle temp is low 40s, load is 77 (using AIDA64) after a few minutes with spikes up to low 80s before fan curve cranks up the fans (side note: it's a simple thing, but it's really nice to not have to deal with the stupid 20C offset that the 1700X has).

Got my 16GB Corsair Vengeance LPX DDR4-3200 running at 2933 MHz, which is perfectly fine by me for now as my R7 system's RAM is at 2666, and it wasn't a crazy expensive kit. I'll wait for the April/May updates before worrying about hitting advertised RAM speed.

Have you tried any other voltage/CPU speeds? I have my 1700 at only 3.7 GHz but it's at 1.25v and I could probably go lower. Ryzen is so sensitive that you could sacrifice 100 MHz and drop temps/watts a ton.

I have the same memory as you also running at 2933 MHz. It's Hynix, for those not aware. I got it because it was so much cheaper than the Samsung b-die 3200 at CAS 14 timings. And I don't really care about the timings as much as I did saving $50. Like you, I'll wait for future updates from AMD for memory support.

So to those out there who get the CAS 18 3200 Corsair, it might be on the QVL (mine was) but it doesn't look like it will run at 3200 so don't waste your time. Set it at a still respectable 2933 and wait a month. You'll save yourself the headache.

 

[Malogeek]

I have the same memory as you also running at 2933 MHz. It's Hynix, for those not aware.

Apparently the specific version 5.39 of the Corsair LPX 3200C16 is Samsung, but the previous versions are Hynix.