We review what can be considered some very beautiful DDR4 memory intended for the Ryzen platform, of course from G.Skill. It's available in multiple frequencies and timings, we test the 3600 MHz kit.... Review: G.Skill TridentZ NEO DDR4 3600 MHz
So they renamed the B-die sammies NEO? This is the same as my Trident Z kit I have 16-16-16-36 3600. Your 1080P numbers are much higher that the numbers in the 3700x review. Whats the story on that?
what is real life performance hit from going to 16-16-16-36 to 16-19-19-39 F4-3600C16D-32GTZN vs F4-3600C16D-32GTZNC
@Roboionatorr Should not be an issue. I got a corsair kit as it was on sale (cheapest 3600 mhz 16gb with heatsink), and with almost identical read/write etc (since same clock), even that my kit is 18-22-22-42/1T@1.35v. I dont mind spending (more) for better performance/what i like, but not sure if a 50% higher price (over my kit 100$) is really worth it to drop timings/voltage just a bit. And i haven't messed with timings/voltage yet, some getting 16-17-17 to work..
These are Hynix C-die and are supposed to be really good. Got a pair myself but there is definitely work to be done with the timings like someone said earlier.
well seems i can totally stick with my 3200 cl14 flare-x? the only good upgrade here seems the cpu from 2700x to 3800x?
3200cl14 still the best? Could you try four modules of 3200 and 3600 to see if it is true that they go better than two modules?
I don't get the 7zip results in this review... It shows 3700X on MEG Godlike having 49MB/s compression and 3900X having 62MB/s compression. Yet, my peasant 3700X does 61-62MB/s in this test. (ABBA bios, just released yesterday for my board, which is an "old" X370) How is it possible that I get a result 25% faster than Guru3D ? (yes, my memory is slightly overclocked, 3600->3733 but that should have a maximum 4% effect, not 25%) The other results seem to be in the same ballpark, so this is just an issue with the 7zip page. https://www.guru3d.com/articles-pages/g-skill-tridentz-neo-ddr4-3600-mhz-review,12.html
What happens though with these 3600 XMP kits, when an individual CPU infinity fabric is not able to work correctly at 1800MHz? Mine will only stay stable to about 1700MHz. So setting to a 3600 XMP profile would just result in a non-post, reboot, and BIOS goes back to default safe 2400 setting for memory. I mean I don't think AMD guaranteed that fabric will run at maximum 1800 speed did they? So I presume for average Joe Soap, if XMP profile fails they are more or less screwed and would have to return the kit to shop - not everyone likes manually hacking in memory sub-timings!
It already sets to half when over 1666 or so, by "default". So if you just plug in the memory then use DOCP without any other tweaking, a 3600 kit will work at Mclock 1800Mhz with Fclock at 900Mhz. Not ideal, and it's better to just reduce the memory speed than losing so much of IF speed.
It doesn't. Im running 3600 ram, and IF is at 1800. Don't use xmp, or try both profiles (1 is for comp, 2 is for perf), but i would use manual clock/voltage for ram, rest auto.
No... it only sets fabric to half over 1866 (not 1666). That's why all these new Ryzen kits are limited to 3600Mhz - anything over this and fabric speed gets cut in half. What actually happens is, if I put in an XMP 3600 kit in my 3900X it will fail to post and reboot to default 2400 - because it can't do fabric at 1800. And AMD never guaranteed that every one could. What happens in this case? Consumer returns kit to seller? Or manually lower speed - which may not be easy for average Joe.
3200 cl14 still the sweet spot , anything higher seems to bring almost no increase in performance , even loss in some cases .
@geogan someone unable to go to the bios and change ram speed from 3600/+ to lets say 3533 or less, should not build their own rig. @ZXRaziel maybe for you. going from 3200 to 3600 gives me an increase of more than 10% for read/write/copy and the increase in latency from 70 to 78 ns is similar, but i also improve on IF speed by maxing the clock to 1800. so the gains on are easily making up for the "loss" (latency). at least with 18-22-22-42, as i haven't messed with timings yet.