Discussion in 'AMD x86 CPUs and chipsets' started by luke o, Dec 27, 2018.
OK I guess I should have watched the whole video.... oops...
Buildziod had things to say on the subject of X570 Boards on Gamers Nexus:
If you want to go for a drive I will help you out with testing. Although Ballarat to my place (3169) is a hike.
I have a 3600X running in my Crosshair VI Hero: https://hwbot.org/search/submissions/permalink?userId=10871&cpuId=5872&cpuBatchId=3604 and I also have a Strix B450-F Gaming here as well.
Weird problems but they could be pointing to a DOA CPU. Are there any other BIOS versions on MSI's website?
They only put up 720p benchmarks with 2080TI class cards so numpties on the internet can point and say "but the 9900K is 25% faster in games". It may be technically correct* however it has no bearing on the real world where GPUs of that class will invariably be used to game at 1440p minimum with all of the eye candy. Stupid example is stupid.
*the best kind of correct
Feeling much better about my storage choice..
The exact same controller I have the hp version wins over these pci4 ssd. So probably worth waiting for newer ssds. With larger faster ram, controllers and flash.
What's with all the local shops charging 50% more for the only x570 mitx board. Should be $300 But you would struggle to find it for $400.
thanks for the link Phido - looks like no real world difference at all which is surprising. SSD speed was really the only reason I was considering an X570 motherboard, as I tend to do a whole upgrade rather than parts over time. Looks like an X470 is a cheaper, just as good option.
edit: though it looks like a second m.2 in most x470 will be speed limited at x2.
I'm not surprised (as I stated several pages back) PCIe 4 will have very little real world impact ssds sustained sequential write speed is all it will affect and this is not what makes a ssd feel fast
Which is why you can't tell the difference between a sata ssd and a M. 2 of silimar grades.
So some testing with voltages.
I set a -0.05 offset, which made it 1.406v in BIOS (1.425v when on Auto and has been confirmed normal), the one thing I noticed is my Cinebench scores went up 30 points.
Now from what I know, well saw in a lot of video's was Ryzen when starved for voltage doesn't crash like Intel, it just performs worst, correct me if I'm wrong though.
I ran tests with Ryzen Master open and the voltages maxed out at 1.39v, but temps were a little higher, 74c up from 70c when I was using 1.32v, core speed the same.
PBO is set to Auto, so I'm guessing it's not on.
I lowered the voltages with -0.125 offset, enabled PBO and got my biggest Cinebench score.
But in Ryzen master the voltage hit 1.42v, so I'm guessing PBO is ignoring the set offset...maybe, it shows 1.42v but the temps are 70c, the same temp I was getting with the 1.32v before PBO being enabled, PBO is only giving me 15 - 20Mhz extra all core boost anyway.
I am trying to play it safe being new to Ryzen, but it does work a little different to Intel, but I'll get it..
Click to view full size!
Click to view full size!
Intel is same. If you set vcore a bit too low it will score less. Why I belt the volts up for the benchmark run. Then default it after.
Quick & done
Gamers Nexus are currently live streaming 3900x LN2 overclocking, if anyone is interested.
There's never been a new pci/agp/pcie/sata/ata/usb/other standard that immediately showed prior standards were redundant. 3-4yrs later though, yes.
What are the stock voltages in these things? 1.4 seems way too high for a 7nm chip but... I don't even know what I'm basing that on.
partybear Can you drop your RAM frequency down lower for testing purposes? Even to 2133 to be sure.
vcore seems fairly high, but unless it's tapping the CPU out it should make it less stable.
After spending a bit of time now memory testing on the CH7 and 3900x the sweet spot for my setup seems to be 3733c14-15-14-14-28 1T.
It's a fair bit easier to work with than 3800Mhz.
Vdimm 1.4v, soc 1.1v. Tested 1Hr gsat and 5000% RamTest
Tested at 4200Mhz with higher cas but Aida64 shows a fair drop in performance.
Used TridentZ 4266 sticks which are B die.
A few performance results for the 3900x at 4.45Ghz. Mem 3733c14
Fire Strike Physics. 32 598
Time Spy. CPU 13 977
Time Spy Extreme CPU 7159
Geekbench 4 Single. 6154. Multi. 49024
My idea of stable voltages is 3 hours of OCCT Small data set no AVX offset, that was for my 24/7 overclock...lol...
Now I'm back from my morning walk I'm going to play more.
I forgot to mention all my benchmarks are just with the stock ram XMP of 3200Mhzz 16-18-18-38, so the Infinity fabric is only running at 1600Mhz.
I'm just trying to figure out how the voltages a PBO work, haven't touched LLC yet.
I'm building up the confidence to try overclocking that high, it's just that Steve's (hardware unboxed) dead 3900x is always in the back of my mind...
in the absense of any official die shots, - Some IR die shots have been done from the taleneted mr Fritz
I/O die https://www.flickr.com/photos/130561288@N04/48268745401/in/photostream/
CCX chip: https://www.flickr.com/photos/130561288@N04/48268822607/in/photostream/
He's apparently going to do some high res colour shots shortly.
I don't know why but die shots always make me stop and pause to consider just how amazing technology is.
Yes I saw his cpu died.
I used manual voltage of 1.385v in bios.
I have a lot of vdroop as no llc settings used yet.
My idea of stable is x1 minute of XTU. Prime 95 built into it ... it really is shit.
The other 2hr59min can be doing benchmarks adjusting Vcore on the fly
Corsair lpx? Should be good for c14 cr1
GN did a 3900x overclocking live stream.
Out of the box on the Gigabyte motherboard the 3900x with around 1.475v stock, no PBO etc.
Nah it's G-Skill Trident Z RGB (F4-3200C16D-32GTZR)
3600X pointless as per most of the speculation. Wonder what the 3800X is going to be like (hey at least they changed a number as well as adding an X)