AMD Ryzen 5000 Builders Thread

[dr1337]

I have my infinity fabric at 2100MHz so it's 1:1.

so you're running 4200mhz ram? Not sure you're actually in 1:1 mode, but if you are then its definitely not stable. My 3600 can do 1900fclk but I lose performance because either my ram or the CPU isn't actually stable. Stable enough for basic tasks and benchmarking sure, but it definitely scores lower than at 1800fclk.

 

[Gideon]

BTW, people here are getting great results with this memory and it's actually in stock over here for decent prices (130€ after 20% VAT):

https://pcpartpicker.com/product/vVVD4D/patriot-viper-steel-16-gb-2-x-8-gb-ddr4-4400-cl19-memory-pvs416g440c9k

This is relatively cheap for that quick B-die. According to buildzoid with custom timings it can do even 3800 CL12. At the very least 4000 CL16 shouldn't be a problem, preferably with 4 DIMMs(to get 4-ranks)

 

[DrMrLordX]

@Gideon

That's awfully cheap for b-die. Is it something new, or just the same old same old?

 

[Gideon]

@Gideon

That's awfully cheap for b-die. Is it something new, or just the same old same old?

Don't know really. I'm just proxying info I received. If I had to build a Zen 3 rig from scratch I would probably buy 2 kits of these.

The model number is: PVS416G440C9K

EDIT: yeah it certainly isn't new, but is B-die and seems to work well:

Patriot Viper Steel DDR4 issues (1T)

I was wondering if someone could help me as I have been trying to get my system stable for 2 days. AMD Ryzen 3700x Gigabyte x570 Elite Motherboard...

www.overclockers.co.uk

The final solution is that gear down mode must be set to ON, now the memory is super stable at 3733 CL14 and 62.6 ns latency benchmark in Aida 64.

That's on a 3700X. On the 5xxx series it should go down to the 50-ies.

 

[Avalon]

Finally got my rig up and running last night. Hit a bit of a bump when I discovered that my motherboard didn't come pre-installed with a BIOS that could support my CPU, and I couldn't pop in my old CPU (Ryzen 5 1600) because it was also unsupported. Went through the Q-Flash+ process and while a bit nerve wracking because I couldn't actually see anything being done, it did work and update the BIOS to the current one, and I was good after that. Wasted a couple hours on assembling, disassembling, and re-assembling my machine multiple times due to that. What a pain, but at least it's over with.

I'm trying to run some benchmarks now to compare to my old R5 1600. Getting some interesting results from AIDA64 in the memory benchmarks, wondering if this is normal?

R5 1600 w/G.Skill Sniper X F4-3600C19D-32GSXKB running @ 2666Mhz 14-15-15-36-50-1T (best the memory controller would do):
Read - 41,271MB/s
Write - 41,069MB/s
Copy - 40,484MB/s
Latency - 85.9ns

R5 5600X w/same memory and settings just for baseline comparison purposes:
Read - 38,994MB/s
Write - 21,281MB/s
Copy - 37,859MB/s
Latency - 74.7ns

The read and copy seem to be noticeably lower, and the write speed is almost half, while latency is showing a marked improvement. I read elsewhere that the write bandwidth is so low because my old R5 1600 actually had two CCDs and thus two links to the memory controller, while my R5 5600X only has a single CCD. The read and copy bandwidth still seems a bit low, though. Any cause for concern? I'll be cranking the memory speed up soon.

 

[cytg111]

MSI MEG X570 UNIFY
Solid board with 3xNVMe and good VRM.

Is this a solid recommendation for a 5900 build?

 

[cytg111]

I went with this kit myself after watching the review as well as the comment about Wendel's testing. The performance difference was minimal between speeds over 3200 and figured that CL14 would give me overhead.

F4-3200C14D-32GTZN-G.SKILL International Enterprise Co., Ltd.

G.SKILL

www.gskill.com

What is this sweetspot and what is it worth in relation to Zen3

AMD Zen 2 Memory Performance Scaling with Ryzen 9 3900X

We take a close look at memory scaling on AMD's new Zen 2 Ryzen 3900X, testing both application and gaming performance at seven different memory speed and timing combinations ranging from 2400 MHz all the way up to 4000 MHz.

www.techpowerup.com

“Unfortunately, we were not able to test the DDR4-3733 "sweet spot" configuration recommended by AMD because our CPU simply was not able to run at the required 1867 MHz Infinity Fabric—1800 MHz was the highest, which we're covering with the DDR4-3600 test.”

 

[KentState]

What is this sweetspot and what is it worth in relation to Zen3

AMD Zen 2 Memory Performance Scaling with Ryzen 9 3900X

We take a close look at memory scaling on AMD's new Zen 2 Ryzen 3900X, testing both application and gaming performance at seven different memory speed and timing combinations ranging from 2400 MHz all the way up to 4000 MHz.

www.techpowerup.com

“Unfortunately, we were not able to test the DDR4-3733 "sweet spot" configuration recommended by AMD because our CPU simply was not able to run at the required 1867 MHz Infinity Fabric—1800 MHz was the highest, which we're covering with the DDR4-3600 test.”

If you go to the 23:40 minute mark of the video, Wendall was quoted that the 2 * DR kit performed best, followed by 4 * SR and then 2 * SR. That together with the game benchmarks from the video show that 3866, 3600 and 3200 where pretty much within a frame or two. Not sure about the sweet spot, but 3200 CL14 2*16GB should perform well and gives me the option to add another 32GB down the road. I didn't want to locked in with a 4*8GB kit.

 

[DrMrLordX]

EDIT: yeah it certainly isn't new, but is B-die and seems to work well:

Interesting! There seem to be a lot of cheap-ish 2x8GB kits at 4000 MHz and up these days, which is really interesting. The price on this RAM at NewEgg is insane:

https://pcpartpicker.com/product/BqWBD3/gskill-ripjaws-v-16-gb-2-x-8-gb-ddr4-4000-memory-f4-4000c15d-16gvk

DDR4-4000 CAS/CL 15 for $170? Unheard of!

 

[Zstream]

R5 5600X w/same memory and settings just for baseline comparison purposes:
Read - 38,994MB/s
Write - 21,281MB/s
Copy - 37,859MB/s
Latency - 74.7ns

That's related 100% to memory. I'm up at 50000MB/s read and have some pretty "ok" ram. 3466 - 16/17/18/36

 

[Avalon]

That's related 100% to memory. I'm up at 50000MB/s read and have some pretty "ok" ram. 3466 - 16/17/18/36

It's the same memory and settings from prior build.

 

[thigobr]

MSI MEG X570 UNIFY
Solid board with 3xNVMe and good VRM.

I have the same board and my experience has been positive so far! I have yet to source a Ryzen 5000 tough. Just concerned with some posts like this one
A lot of people are having issues with MSI boards and the latest BIOS: unable to boot/get stability with memory frequencies greater than 3200MHz

...

...

forum-en.msi.com

 

[lightmanek]

Is this a solid recommendation for a 5900 build?

Very solid, no complains from me. This is the exact same board as MSI MEG X570 ACE Hardware Unboxed tests with, just stripped from all RGB and bling.
One of the best memory overclockers on AM4 platform too, at least from the initial X570 range.
For the price, it is really good pick.

I have the same board and my experience has been positive so far! I have yet to source a Ryzen 5000 tough. Just concerned with some posts like this one
A lot of people are having issues with MSI boards and the latest BIOS: unable to boot/get stability with memory frequencies greater than 3200MHz

...

...

forum-en.msi.com

At least for my G.Skill Trident Z 3600 B-Die modules it works flawlessly! 4x8GB 3800MHz CL18 at 1.37V is more than I was expecting with 4 modules.

 

[ddogg]

5950x came in today and so far very impressed. Looks like 4.6 all core is about the max I can do without having to bump the voltages really high. 11514 is the max I've hit so far running at 4.6 and 3600 CL16 memory

 

[DrMrLordX]

5950x came in today and so far very impressed. Looks like 4.6 all core is about the max I can do without having to bump the voltages really high.

That's really impressive on a 16c chip. What voltage are you running to achieve that score?

 

[ddogg]

That's really impressive on a 16c chip. What voltage are you running to achieve that score?

1.325v for a stable 4.6 all core. Temps peak around 80c even though I'm running a custom loop with a 480mm rad that is exclusively used to cool the CPU. I didn't have too much time to tinker with it yesterday but I may eventually just run this stock due to the high boost clocks. 3-4 cores were boosting to 5Ghz at stock settings.

 

[Gideon]

1.325v for a stable 4.6 all core. Temps peak around 80c even though I'm running a custom loop with a 480mm rad that is exclusively used to cool the CPU. I didn't have too much time to tinker with it yesterday but I may eventually just run this stock due to the high boost clocks. 3-4 cores were boosting to 5Ghz at stock settings.

Don't forget to update your sig

 

[DisEnchantment]

My 5950X arrived, and I ran GB 5.2.5 on it, just for fun.

Gigabyte Technology Co., Ltd. X570 AORUS PRO - Geekbench Browser

Benchmark results for a Gigabyte Technology Co., Ltd. X570 AORUS PRO with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

Different result with CPU governor set to performance

Gigabyte Technology Co., Ltd. X570 AORUS PRO - Geekbench Browser

Benchmark results for a Gigabyte Technology Co., Ltd. X570 AORUS PRO with an AMD Ryzen 9 5950X processor.

browser.geekbench.com

Weird thing is that this benchmark did not even boost the CPU properly in ST mode. I am polling the frequency every 100ms from /proc/cpuinfo and save it to a file, when I check the file, the CPU is not spending time anywhere near 5 GHz.
When I booted to Windows and trying Cinebench or CPU-Z bench I can see it hitting 5GHz and my AIO fans ramp up. Granted, Cinebench for example starts running AVX, but in single core my 5950X will hit 5.015 - 5.05 GHz
Have not tuned memory. I used my old TridentZ CL14 kit (14-14-14-34 3200). I will retest with new TridentZ Neo 3600CL16 4x8GB kit. Or maybe not, since this is a rather pointless test and I am not controlling everything.

ST test

Spoiler

Multicore test do load the CPU

Spoiler

Active mitigations for 5950X

Spoiler

Running GIMP on this processor is so fast, ooof. The snappiness is very noticeable.
Trying to compile the kernel now, but I guess it is going to kill it.

One thing I love about this processor is much lower thermals compared to my 3900X. It could be due to being a better binned chip but anyway it is impressive.

 

[Kenmitch]

Has anybody with a 5xxx run the following benchmark?

Post your Cpumark99 scores

download here This is an old single-threaded benchmark utility but I've tracked this over the years and would love to see what people are getting with their processors. e8600@4.2ghz: 605 UPDATE 3/16/12: Please update with what you have. I'd really love to see your Sandy Bridge scores!

forums.anandtech.com

It would be somewhat interesting to see the results. Heavily dominated by Intel for the last couple of decades.

 

[DisEnchantment]

Just 1h of playing with CPU, no memory optimizations, no per core tweaks and voltage not dialed-in yet.
My 3900XT on the charts was ran at 4500 CCX0 and 4300CCX1.
I can do ST 4.9GHz with 1.33V on core 3 CCX0, but haven't tried for 5GHz yet

Fun CPU to play with and Precision Overdrive is quite good too!
I set EDC and TDC to minimum (45W) and ran CB R20. To my shock and disbelief, I scored around 5800 points which is the same as my old, overclocked and tweaked Threadripper 1920X which was consuming more than 180W for that run

Strange that your AIDA Memory benchmark has 68.3ns with 3667CL16, but I get 61.5ns with 3600CL16. G.Skill TridentZ Neo 4x8GB kit.
Also the cache BW is totally different. You have older build which is probably not updated for Vermeer.


I forgot to test it with my CL14 3200 MT/s kit. Not gonna change the RAM again,

 

[Avalon]

Updated AIDA64 numbers for the memory benchmarks w/2x16GB G.Skill Sniper X 3600 C19 modules:

**Ryzen 5 1600**

Memory timings @ 2666Mhz:
CAS - 14
tRCD - 15
tRP - 15
tRAS - 36
tRC - 50
CR - 1T

AIDA64 @ 2666Mhz
Read - 41,271MB/s
Write - 41,069MB/s
Copy - 40,484MB/s
Latency - 85.9ns

**Ryzen 5 5600X**

Memory timings @ 2666Mhz:
CAS - 14
tRCD - 15
tRP - 15
tRAS - 36
tRC - 50
CR - 1T

Memory timings @ 3600Mhz:
CAS - 16
tRCD - 20
tRP - 20
tRAS - 40
tRC - 60
CR - 1T

AIDA64 @ 2666Mhz:
Read - 38,994MB/s
Write - 21,281MB/s
Copy - 37,859MB/s
Latency - 74.7ns

AIDA64 @ 3600Mhz:
Read - 51,182MB/s
Write - 28,734MB/s
Copy - 49,665MB/s
Latency - 64.2ns

Dropping my tRCD or tRP any lower seems to result in system instability, and slightly lower tRC or tRAS seemed to decrease scores a bit, so I think I found my sweet spot. Going to run some additional memtest/burn in, then move on to the CPU

 

[Noid]

Strange that your AIDA Memory benchmark has 68.3ns with 3667CL16, but I get 61.5ns with 3600CL16. G.Skill TridentZ Neo 4x8GB kit.
Also the cache BW is totally different. You have older build which is probably not updated for Vermeer.
View attachment 33481

I forgot to test it with my CL14 3200 MT/s kit. Not gonna change the RAM again,

New AIDA64 with new score evaluations.

Just so everyone is on the SAME version of AIDA64.
You cant compare using AIDA64 anymore.
( unless you use the SAME version )

Also, the new version has not been " boosting " the CPU anymore.
( takes a few " re-runs " to get the previous " boost " score - making this tool questionable for testing )

However, the nanos are picky at each IF setting on my MB.
DDR re-tuning is needed at each IF setting ( or DDR base Mhz )
( taking alot of time to find the ones to modify , his nanos are just a matter of tuning )

 

[Shmee]

Well I finally managed to get a 5800X ordered. Should be here pretty soon. Besides update motherboard to latest UEFI, anything else I should do, like driver or windows related?

 

[Hitman928]

Well I finally managed to get a 5800X ordered. Should be here pretty soon. Besides update motherboard to latest UEFI, anything else I should do, like driver or windows related?

Newest chipset drivers. Other than that, I think it should be good to go. Which store did you find one in stock?

 

[moinmoin]

Also, the new version has not been " boosting " the CPU anymore.

So we should treat its results as "idle score" from this version on?