FRONTPAGE AMD Ryzen 9 3900X and Ryzen 7 3700X CPU Review

[bignazpwns]

I got a $10k thermal camera at work I'll try and get some pictures. We got keysight coming in to calibrate everything next week so I'll wait till after so everything is known to be accurate.

My temps are on water. At 4.3ghz on a 3900x 1.35v I'm loading at 62c that's after my loop is completely heat soaked. So I don't know how a heatsink would do with it.


Also what are you watching for the temps? I found core temp to report the highest temp and hold that temp longer. Ryzen master is some what in the middle. Not the lowest temp reported but definitely not the highest. Idle they are pretty close but during use I seen a 10c higher reading on coretemp then I did with Ryzen master.

 

[Woomack]

Run AIDA64 stability test CPU+FPU tests together (and only these 2) or Prime95 small FFT, last AVX version. Temps in both are similar to me. I'm using hwinfo64 for temps reading but AIDA64 is showing the same CPU temp.
I only noticed that on Gigabyte, CPU+SoC power is about 80-100W (R5 3600) under full load while on ASRock it's showing about 50-60W (R7 3700X). Something is wrong in that and actually in both of these values.

 

[EarthDog]

Also what are you watching for the temps? I found core temp to report the highest temp and hold that temp longer. Ryzen master is some what in the middle. Not the lowest temp reported but definitely not the highest. Idle they are pretty close but during use I seen a 10c higher reading on coretemp then I did with Ryzen master.

Hwinfo64 is what I use...

 

[bignazpwns]

Just sent an email to Laird about getting some Tgon 9XXX samples. The stuff is much better then what is available on the market. So I'll test it.

So had something strange happen. Doing a single core cpuz benchmark. did 588 single core. What is a little under what 5ghz should pull. Im guessing 5ghz should bring in around 600. But its higher then i expected and those temps were higher then what i been seeing. Room had a high ambient but i was in the upper 60's for the most part then bam all this happened and temps shot up. So its definitely not 5ghz. Single core my cpuz boost to 4.65ghz consistently, cpu tops out at 4.725 @ 1.45v. Looks like a reporting error. CPUZ reported the same clocks and bclock. But the score for the single core is whats got me confused to what happened. 588 is high for a 4.65ghz single core. Single core power consumption was as 66w during the run.

No issues since then. I haven't seen this happen again. Nothing really changed to cause it.

This is the first weird thing i seen really in terms of bugs.

 

[Woomack]

One tip how to quickly find CPU, memory or PCIE errors ... simply run hwinfo64 and on the bottom of the sensors window there is an info about errors. Even though it seems stable then when there are 1-2 errors then it's crashing in stability tests in 5-20 mins so when you see an error then there is no point to even test it.

 

[bignazpwns]

I cant wait for the day i can get accurate info from one program and not have to use 3.

Core temp reports the clocks pretty accurate until it decides to do this. I wish i could pull a 178mhz bclk

 

[ShrimpBrime]

Lol. 7739Mhz. Too bad Cpu-z won't do that for you.

I've been using GPU-Z for Cpu temps. Seems accurate to AIDA64 and HWInfo64. I do believe I'm in the 80c when running 1.45v+ yes indeed. lol (2700x)

 

[bignazpwns]

That's legit and look at those temps on water.

It's weird it does it if I walk away. Dosent go into standby or anything like lack. I look and my bclk just keeps going higher.

 

[EarthDog]

I cant wait for the day i can get accurate info from one program and not have to us

more than 1 may mess with reading sensors properly. Hwinfo64 does it all.....

 

[Bustos]

All Ryzen 3000s X chips available! If you’re in my area, that is.

 

[Blaylock]

more than 1 may mess with reading sensors properly. Hwinfo64 does it all.....

FWIW I've found HWMonitor is the most accurate for my build (ASUS ROG STRIX B450i w/ 3700X). All the other softwares seem to display voltage, Temps, or a combination of sensors inaccurately. (using a single application at a time of course)

 

[UltraTaco]

How do you know which one is 'most' accurate? Have you done physical measurements at the spurce?

 

[Blaylock]

That's a valid question taco.

For voltage a look at what I set in BIOS and see what the sensor is reading. Taking in to account droop/drop to see which is closer to what "should be" displayed.

Temperature is a bit tricky but if one program is showing a much different reading than several others it's easy to rule that one as a false or inaccurate reading.

 

[The Coolest]

If you have a debug display on the motherboard, you can probably set it to display CPU temperature.
I don't have a 3rd gen Ryzen yet, but on 1st and 2nd gen, temps in Core Temp matched exactly to what the display on the board was showing.

 

[bignazpwns]

more than 1 may mess with reading sensors properly. Hwinfo64 does it all.....

HWinfo for some reason for my build is extremely off. .867 volts in HWinfo shows as .998v. so it's not taking in account for my voltage offset. Also what I seen is the max clocks after an hour or so all show higher then they should be. Unrealistically high. Happened on my 2700x also. So it's a mobo/hwinfo combo.

 

[bassnut]

What's your thoughts of the HyperX Predator 4000 cleared 17 ..... with the ryzen 3900X. These sticks are considerably cheaper the most 4000 that I have seen @ $259

 

[custom90gt]

I'm running my HyperX predator 4000 @ 3733 17-17-17-36 without any issues, I haven't tried to dial down the timings.

 

[mackerel]

I have tried that ram on Zen 2 systems.

On 3700X + X370 with AGESA Combo-AM4 1.0.0.2, it works fine at XMP3600 settings, but doesn't boot at XMP4000. They have since released new bios with 1.0.0.3AB which I haven't loaded yet.
On 3600 + B450 with AGESA Combo-AM4 1.0.0.3, it works fine at both XMP3600 and XMP4000. Generally higher performance in what I've tested so far at 3600 presumably due to tighter timings and synchronous IF. They have since released 1.0.0.3ABB which I have installed but not tested.

Edit: bios version is important. On the same B450 system with 1.0.0.1, I couldn't run at 3600. I don't know where I posted it, but I did try to tighten the timings at both 3600 and 4000, and saw gains in both. 4000 still lagged 3600 results.

 

[bassnut]

Thanks .... with these timings I am assuming Samsung B dies?

 

[custom90gt]

Thanks .... with these timings I am assuming Samsung B dies?

That they are!