Ryzen 3900X - Initial benchmarks (yes, including DCS!)

The test is according to your instructions, I was surprised too.

Perform the benchmark according to the DackSter94.

Won't mind a lower resolution.

 

benchmark.trk https://forums.eagle.ru/attachment.php?attachmentid=214130&d=1563798700

Sorry, still can't. I don't have SU25 or FC3.

This is the downside of this test.

My order is not thru yet damn, still out of stock.

My order is not thru yet damn, still out of stock.

I got super lucky. On the morning of 7/7 I put in a pre-order for both the 3800X and 3900X with my local store, fully intending to actually get the 3800X but willing to take whichever showed up first. They got a small number of 3900X in the first week and I managed to get one, but they’ve had none since. The 3800X still hasn’t arrived there.

Truthfully, until we get an updated AGESA BIOS 1.0.0.4, the chip is not performing as it should anyway. I’m only getting single core boosts of 4250, nowhere near the 4600 I should get. So, you’re not really missing THAT much, just the early adopter growing pains!

I got super lucky. On the morning of 7/7 I put in a pre-order for both the 3800X and 3900X with my local store, fully intending to actually get the 3800X but willing to take whichever showed up first. They got a small number of 3900X in the first week and I managed to get one, but they’ve had none since. The 3800X still hasn’t arrived there.

 

Truthfully, until we get an updated AGESA BIOS 1.0.0.4, the chip is not performing as it should anyway. I’m only getting single core boosts of 4250, nowhere near the 4600 I should get. So, you’re not really missing THAT much, just the early adopter growing pains!

From what I have read and watched ,the Ryzen CPU wont boost to 4.6 on 1 core if that ONE core is doing anything but LIGHT work.

move your mouse around without marking/highlightimng anything, that should allow max boost. While you do it, watch the Volts, thats when the Volts are allowed to spike too. Just when you ask for serious workloads, it will need more Ampere and thus, in order to not disobey the preprogrammed heat/load/vcore triangel it has to clock down.

Buildzoid has a fantastic video showing this. Forme, tbh,it has a bad taste as Lisa did not underline this "limitation", al they did say was " Turbo is xy high on that version of Ryzen blabla" Imho, that is not lying, but it is also not the full truth.

Still, I like those Ryzens

From what I have read and watched ,the Ryzen CPU wont boost to 4.6 on 1 core if that ONE core is doing anything but LIGHT work.

 

move your mouse around without marking/highlightimng anything, that should allow max boost. While you do it, watch the Volts, thats when the Volts are allowed to spike too. Just when you ask for serious workloads, it will need more Ampere and thus, in order to not disobey the preprogrammed heat/load/vcore triangel it has to clock down.

 

Buildzoid has a fantastic video showing this. Forme, tbh,it has a bad taste as Lisa did not underline this "limitation", al they did say was " Turbo is xy high on that version of Ryzen blabla" Imho, that is not lying, but it is also not the full truth.

 

Still, I like those Ryzens

I know some people are saying that, but I don’t think it’s true. It think immature BIOS is to blame and it will eventually get sorted. Mine doesn’t exhibit that behaviour, I never see 4.6ghz.

Let's hope you are right and my info is based on immature Bios versions.

Looking at the numbers tho, that is hard to believe.

Let's hope you are right and my info is based on immature Bios versions.

 

Looking at the numbers tho, that is hard to believe.

I hope so too, I'd hate to think I paid as much as I did to max out at 4.2Ghz when I could have paid less than half for a 3600 that boosts to the same single core frequency.

Indeed, I may even consider starting an RMA process if the BIOS doesn't fix it.

https://www.pcworld.com/article/3432124/pre-binned-ryzen-3000-cpu-listings-reveal-the-limits-of-amd-overclocking-potential.html

the important part:

quote:

Take the monstrous 12-core, 24-thread Ryzen 7 3900X. AMD specifies it as having a base clock of 3.8GHz, and a maximum boost clock of 4.6GHz. According to Silicon Lottery’s stats, every single 3900X it tested can hit an all-core overclock of 4.0GHz. Bump that up to 4.05GHz and 87 percent pass, and just 68 percent manage to hit 4.1GHz on all cores. From there, the survival rate drops precipitously, with just 35 percent of 3900X samples able to crack 4.15GHz, and a mere six percent hitting an all-core overclock of 4.2GHz.

At Case King, der8auer managed to get a few samples to 4.25GHz and 4.3GHz, but that site doesn’t list survivor rates like Silicon Lottery.

OK guys, as requested, here it is:

 

-Using the attached track file (free flight over Tbilisi),

-all non-VR benchmarks at 1080p (sorry, I don't have a 1440p monitor like Dackster, so I cannot duplicate his settings).

-FPS measured with FRAPS.

-Computer specs in sig, 1080ti slightly overclocked.

-3900X is set to game mode in Ryzen master (disables one of the CCX for higher single core boost).

 

Low preset: Min: 1 Max: 189 Avg: 177 (I don't know where the 1 min comes from...)

Med preset: Min: 1 Max: 200 Avg: 179

High preset: Min: 88 Max: 188 Avg: 126 (High preset, plus 4 x MSAA, 16 x AF, all sliders maxed)

Rift S, VR preset: Min: 0 Max: 81 Avg: 79

Rift S, High preset (as above): Min: 1 Max: 81 Avg: 45 (ASW was off, but it still seems to like to hang around 40 or 80, though occasionally it will hit FPS between 80 and 40)

Thank you for the benchmark. Because I'm trying to figure out if switching to Ryzen 3000-series is worth it, I tried it out on my current machine and these were the results:

CPU: i7 4790K @ 4.4GHz
RAM: 16GB DDR3-2400 (CL 11-13-13-31)
GPU: Gainward GTX 1080 GLH

Average FPS in low and medium presets did not differ much from yours, despite the slower GPU (my 1080 vs your 1080 Ti). Maybe that's hitting a CPU bottleneck here?

In high preset, the slower GPU falls back a bit and in VR it finally gets crushed. You did not specify your pixel density used in DCS VR settings - I had mine set up to 1.3, which is my usual setting in VR gameplay.

https://www.pcworld.com/article/3432124/pre-binned-ryzen-3000-cpu-listings-reveal-the-limits-of-amd-overclocking-potential.html

 

 

the important part:

quote:

Take the monstrous 12-core, 24-thread Ryzen 7 3900X. AMD specifies it as having a base clock of 3.8GHz, and a maximum boost clock of 4.6GHz. According to Silicon Lottery’s stats, every single 3900X it tested can hit an all-core overclock of 4.0GHz. Bump that up to 4.05GHz and 87 percent pass, and just 68 percent manage to hit 4.1GHz on all cores. From there, the survival rate drops precipitously, with just 35 percent of 3900X samples able to crack 4.15GHz, and a mere six percent hitting an all-core overclock of 4.2GHz.

At Case King, der8auer managed to get a few samples to 4.25GHz and 4.3GHz, but that site doesn’t list survivor rates like Silicon Lottery.

There’s very little manual overclock room in 3000 series. They almost seem to operate more like GPUs, clocking themselves as high as possible within power and thermal limits. Indeed, mine actually runs at 4.1 all cores during stress testing, and that’s with all settings at auto. You don’t get better performance with manual overclocking on these chips. Now if my single core boost would start hitting somewhere near where it should (4.6ghz), that would be nice! Hopefully AGESA 1.0.0.4 will bring improvements.

Thank you for the benchmark. Because I'm trying to figure out if switching to Ryzen 3000-series is worth it, I tried it out on my current machine and these were the results:

 

 

CPU: i7 4790K @ 4.4GHz
RAM: 16GB DDR3-2400 (CL 11-13-13-31)
GPU: Gainward GTX 1080 GLH

 

 

Average FPS in low and medium presets did not differ much from yours, despite the slower GPU (my 1080 vs your 1080 Ti). Maybe that's hitting a CPU bottleneck here?

 

 

In high preset, the slower GPU falls back a bit and in VR it finally gets crushed. You did not specify your pixel density used in DCS VR settings - I had mine set up to 1.3, which is my usual setting in VR gameplay.

Low and medium settings are definitely CPU bound. I’m running 1.2 for PD in VR.

What program do you use to benchmark? Does it create the fancy graph for you?

I got news and I should get my 3900X tuesday to test in the weekend. :)

Also, just for FYI some reviewers mentioned the 2 chiplets on the 3900X are from different bins and you better off not overclocking locked all core because your limited to the lowest one. The secret to optimize performance for 3000 series is to get the highest memory speed with the lowest timings at the same time the infinity fabric is synchronized for same clock.

What program do you use to benchmark? Does it create the fancy graph for you?

No, sadly not. I used MSI afterburner's benchmarking mode to get the numbers and plugged these into LibreOffice Calc to create the graph.

https://www.pcworld.com/article/3432124/pre-binned-ryzen-3000-cpu-listings-reveal-the-limits-of-amd-overclocking-potential.html

 

 

the important part:

quote:

Take the monstrous 12-core, 24-thread Ryzen 7 3900X. AMD specifies it as having a base clock of 3.8GHz, and a maximum boost clock of 4.6GHz. According to Silicon Lottery’s stats, every single 3900X it tested can hit an all-core overclock of 4.0GHz. Bump that up to 4.05GHz and 87 percent pass, and just 68 percent manage to hit 4.1GHz on all cores. From there, the survival rate drops precipitously, with just 35 percent of 3900X samples able to crack 4.15GHz, and a mere six percent hitting an all-core overclock of 4.2GHz.

At Case King, der8auer managed to get a few samples to 4.25GHz and 4.3GHz, but that site doesn’t list survivor rates like Silicon Lottery.

"All core" has nothing to do with DCS.

I got news and I should get my 3900X tuesday to test in the weekend. :)

 

Also, just for FYI some reviewers mentioned the 2 chiplets on the 3900X are from different bins and you better off not overclocking locked all core because your limited to the lowest one. The secret to optimize performance for 3000 series is to get the highest memory speed with the lowest timings at the same time the infinity fabric is synchronized for same clock.

I think the magic number is 3800mhz, right?

I’ve yet to get my ram higher than 3200, though I haven’t played with timings much. With first gen I never even got it to it’s rated 3200!

I haven’t been too worried about my low boost, as DCS is now GPU limited at my settings so extra CPU tweaking will have minimal effect anyway.

I think the magic number is 3800mhz, right?

 

I’ve yet to get my ram higher than 3200, though I haven’t played with timings much. With first gen I never even got it to it’s rated 3200!

 

I haven’t been too worried about my low boost, as DCS is now GPU limited at my settings so extra CPU tweaking will have minimal effect anyway.

3600, unless you also oc the IF clock, that may work for 3733 and maybe even on 3800 with top pick dies, but 3600 is the number

going to get my CPU today :)

Tests maybe next weekend.

EDIT: got it! now testing the old CPU then will install the new one.

going to get my CPU today :)

 

Tests maybe next weekend.

 

EDIT: got it! now testing the old CPU then will install the new one.

Looking forward to your results.

BTW..AMD released a brand new chipset driver today :joystick:

Ohh, I have 2700X too. Tell me please, once you swap it, what you did. Most folks say there is no need to reinstall windows. Just upgrade BIOS, uninstall chipset drivers, swap cpu, fire up the pc and install chipset drivers. But some suggest its better to make a fresh install.

My 3900X is supposed to arrive in early September. I kinda envy those in better supplied countries.

I did just swap the BIOS prior to CPU swap but forgot to remove AMD drivers and ran into problems on windows boot. Had to boot in safe mode and uninstall. Then it started normally. Now I'm trying to settle Infinity fabric with memory clocks (1800/3600Mhz).

Temperatures are SCARY. Idle around 50C, haven't dared running benchmarks yet. CPU boost goes to 4500Mhz single core and 4150Mhz all-core. But once in cinebench I believe this will be lower when using AVX.

I did just swap the BIOS prior to CPU swap but forgot to remove AMD drivers and ran into problems on windows boot. Had to boot in safe mode and uninstall. Then it started normally. Now I'm trying to settle Infinity fabric with memory clocks (1800/3600Mhz).

 

Temperatures are SCARY. Idle around 50C, haven't dared running benchmarks yet. CPU boost goes to 4500Mhz single core and 4150Mhz all-core. But once in cinebench I believe this will be lower when using AVX.

Because of the density of the die, temp swings are massive. Starting an all core FFT on prime95, mine jumps immediately from 35 to 90, then holds around 90. There are just so many transistors in such a small package that there’s nowhere for the heat to go until the spreader and heat sink start to soak it up.

I’m glad to hear you are single core boosting well, I’m still not getting past 4.2ghz. What bios are you using for your C6H?

I'm running 3900x on an Asus Crosshair VII Hero. Loading up bios default givs me 4525 ghz but with as tight timings on the memory as I can get it only boost to 4275ghz (2x16 GB dual rank memory @1900/3800, BIOS 2701 AGESA 1.0.0.3 ABB)

preliminary synthetic testing sees about 14-15% single core performance increase from my older 2700X (OC @ 4.2 Ghz all core) compared to the new 3900X (stock+undervolted @ 1.35V). Multicore gets a MASSIVE 66% improvement.

Clocks vary between 3 GHZ and 4.525 Ghz. I also noticed that indeed the 2 chiplets boost differently. One goes up to 4.525 Ghz and the other one stays 4.2-4.3 Ghz.

The increase in FPS in DCS is what interests me the most, cant wait to see that one mate.

As for chiplets binning. If thats the case what're saying, than well that is a shady practice. Their selling high end chip with only half of it being really high end and able to reach maximum clocks, while rest is akin to a mid or low range piece.

3600 CL14 :thumbup:

Great rig Pilotasso !