*** Official Ryzen Owners Thread ***

[hudson86]

Whats a good memory latency score on Ryzen 2700x / Asrock X470 Taichi , 2.0 Bios

Tried Agesa 0.0.7.2 on bios(Ryzen 3000 bios) 3.10 but reverted back to 2.0 after poor PBO clocks/ lower Cinebench 20 scores.

I have this 16gb kit
https://www.overclockers.co.uk/team...nnel-kit-black-red-tdprd416g32-my-075-tg.html

Its Samsung B-die, but CL16, i've seen a few comments on reddit that refer to it as Samsung B-Die trash bucket .. lol . I can get it to CL14 using Ryzen Calc but occasional crashes. Suspect if i put more voltage through it and do some longer stability tests might get there.

For the mo i'm running it on Ryzen Calc profile V2 , Fast 3200 CL16, my Aida scores are memory read 50016, read , 65.1ns, cache 0.9ns ,



What should / could i get to for 24/7 usage.

Ta muchly

 

[Orange Nexus]

Whats a good memory latency score on Ryzen 2700x / Asrock X470 Taichi , 2.0 Bios

Tried Agesa 0.0.7.2 on bios(Ryzen 3000 bios) 3.10 but reverted back to 2.0 after poor PBO clocks/ lower Cinebench 20 scores.

I have this 16gb kit
https://www.overclockers.co.uk/team...nnel-kit-black-red-tdprd416g32-my-075-tg.html

Its Samsung B-die, but CL16, i've seen a few comments on reddit that refer to it as Samsung B-Die trash bucket .. lol . I can get it to CL14 using Ryzen Calc but occasional crashes. Suspect if i put more voltage through it and do some longer stability tests might get there.

For the mo i'm running it on Ryzen Calc profile V2 , Fast 3200 CL16, my Aida scores are memory read 50016, read , 65.1ns, cache 0.9ns ,



What should / could i get to for 24/7 usage.

Ta muchly

That right there is good for that speed. I don't think you can get any better result for 3200MHz. A 3200 Cl14 will get quite same result.

 

[chrcoluk]

sub timings can be more important than main timings, did you use optimised ryzen calculator values?

I posted on reddit asking which is more important on ryzen, latency or memory bandwidth for ram, as I had to choose between 3000CL14 and 3200CL16, the winning vote was 3200CL16. I then made an adjustment in the bios someone suggested related to geardown mode and it made the 3200CL16 latency almost as good as I had on 3000CL14 which kind of sealed the deal.

One thing to be clear on tho, I would never run ram in a "almost stable" config, all the configs I was choosing from had 0 errors on ram stability testing, 3200CL14 errored hence was not considered.

I kind of want to tweak my intel ram the same way, but there is no intel ram calculator, since I dropped my ram in my intel rig from 3200CL14 to 3000mhz on exact same timings, there is very likely the room to tighten the timings, but withot a calculator I wouldnt know where to start.

 

[keenan]

A quick question to anyone that has done some extensive testing, which configuration would be better for real world usage? All are at 1.45v..

3200Mhz 12-15-12-12-21
3466Mhz 14-17-14-10-21
3600Mhz 14-20-14-14-28

??

 

[hudson86]

That right there is good for that speed. I don't think you can get any better result for 3200MHz. A 3200 Cl14 will get quite same result.

Right cheers, whats the absolute best NS i can aim for between 3200mhz and perhaps 3600mhz ( with probably better memory ) with the 2700x ... question aimed at anyone in general !

Interestingly i paid £160 for that memory early December and it was that price-ish till quite recently, its now £98 , so DDR4 memory prices are finally dropping somewhat .

 

[keenan]

Right cheers, whats the absolute best NS i can aim for between 3200mhz and perhaps 3600mhz ( with probably better memory ) with the 2700x ... question aimed at anyone in general !

Interestingly i paid £160 for that memory early December and it was that price-ish till quite recently, its now £98 , so DDR4 memory prices are finally dropping somewhat .

What DRAM voltage are you running at now?

Have you checked with taiphoon burner if they are indeed B-dies? (sorry just checked your history and they are indeed Samsung B-die)

Before thinking of getting other memory, see if you can run these at 3600Mhz Cl14. On Samsung B-die you are perfectly safe upto 1.5v for daily 24/7 use, but you should be able to achieve that at 1.45v.

 

[hudson86]

What DRAM voltage are you running at now?

Have you checked with taiphoon burner if they are indeed B-dies? (sorry just checked your history and they are indeed Samsung B-die)

Before thinking of getting other memory, see if you can run these at 3600Mhz Cl14. On Samsung B-die you are perfectly safe upto 1.5v for daily 24/7 use, but you should be able to achieve that at 1.45v.

Ta, i'll try that, without dropping out to bios to check , i'm pretty sure 1.38 / Soc at .98

found some results on reddit, one guy hitting 56k / 60ns, so thats something to aim for..

https://www.reddit.com/r/Amd/comments/a36xo3/amd_ryzen_1500x_aida64_memory_benchmark_results/

|:-:|:-:|:-:|:-: Memory|56072 MB/s|55487 MB/s|52342 MB/s|60.2ns L1 Cache|1063.5 GB/s|536.11 GB/s|1063.6 GB/s|0.9 ns L2 Cache|1038.3 GB/s|521.17 GB/s|946.44 GB/s|2.8 ns L3 Cache|438.37 GB/s|442.89 GB/s|443.13 GB/s|9.3 ns
Ryzen 7 2700X 4.325GHz 100.0MHz FSB
DDR4-3600 15-15-15-36 CR1
Crosshair VI Hero BIOS 6201 (AGESA PinnaclePi-AM4 1.0.0.2)

 

[Orange Nexus]

Ta, i'll try that, without dropping out to bios to check , i'm pretty sure 1.38 / Soc at .98

found some results on reddit, one guy hitting 56k / 60ns, so thats something to aim for..

https://www.reddit.com/r/Amd/comments/a36xo3/amd_ryzen_1500x_aida64_memory_benchmark_results/

If you can get lower than 60 like 58, yah, 3600 or higher would be great. Just basing it off mem latency. Sweet spot, imo, is 3466 Cl14. Its needed for high-end gpus using high Hz monitor without putting much stress on the RAM. Except if the RAM is rated for 3600Mhz or higher.

https://i.imgur.com/qRQ6C7U.jpg

 

[hudson86]

If you can get lower than 60 like 58, yah, 3600 or higher would be great. Just basing it off mem latency. Sweet spot, imo, is 3466 Cl14. Its needed for high-end gpus using high Hz monitor without putting much stress on the RAM. Except if the RAM is rated for 3600Mhz or higher.

https://i.imgur.com/qRQ6C7U.jpg

Cheers, good info, I will have a good play later. Am playing at 1440p / 144mhz on Vega 64 Lc, so sounds like where I should be.

I'll start at 1.45 and see what I can get.

 

[humbug]

2133Mhz - 3200Mhz - 4000Mhz side by side Ryzen testing.

The biggest difference is from 2133Mhz to 3200Mhz, as much as 30%, from 3200Mhz to 4000Mhz its mostly only around 5%.

 

[Orange Nexus]

2133Mhz - 3200Mhz - 4000Mhz side by side Ryzen testing.

The biggest difference is from 2133Mhz to 3200Mhz, as much as 30%, from 3200Mhz to 4000Mhz its mostly only around 5%.

Maybe the 4000 speed was not stable? And that timing is quite loose rendering it as good as prolly a RAM set at 3200 Cl14.

Here is another test showing the timing used but the highest speed was only set at 3333MHz.

https://i.imgur.com/eGvnOqw.jpg

Found here . . .

https://www.youtube.com/watch?v=HsQAtsKX3mE

Some games don't respond as much. Here is the timing used . . .

https://i.imgur.com/02mu79t.png

But, like i said, anything over 3200MHz with tight timing, imo, is only necessary for high-end gpus.

 

[Orange Nexus]

I never tested the different RAM speeds i got from my FlareX in gaming, since i only have a low tier gpu. But in Cinebench, the difference from 3200 to 3400 RAM speed is like a 100MHz more of cpu oc. R7 2700 oc'ed to 4.1GHz.

3200 MHz = C15 MT score of 1870
3400 MHz = C15 MT score of 1900

Interestingly, not much difference in ST results.

https://i.imgur.com/42DiGwA.jpg

 

[kitfit1]

Whats a good memory latency score on Ryzen 2700x / Asrock X470 Taichi , 2.0 Bios

Tried Agesa 0.0.7.2 on bios(Ryzen 3000 bios) 3.10 but reverted back to 2.0 after poor PBO clocks/ lower Cinebench 20 scores.

I have this 16gb kit
https://www.overclockers.co.uk/team...nnel-kit-black-red-tdprd416g32-my-075-tg.html

Its Samsung B-die, but CL16, i've seen a few comments on reddit that refer to it as Samsung B-Die trash bucket .. lol . I can get it to CL14 using Ryzen Calc but occasional crashes. Suspect if i put more voltage through it and do some longer stability tests might get there.

For the mo i'm running it on Ryzen Calc profile V2 , Fast 3200 CL16, my Aida scores are memory read 50016, read , 65.1ns, cache 0.9ns ,



What should / could i get to for 24/7 usage.

Ta muchly

I get the following, at Cas14 3533Mhz with 4 sticks of ram. Forget the clockspeed in the screenie, it was a one off, my normal all core clock is 4.25Ghz but the AIDA result is nearly always the same.

 

[chrcoluk]

2133Mhz - 3200Mhz - 4000Mhz side by side Ryzen testing.

The biggest difference is from 2133Mhz to 3200Mhz, as much as 30%, from 3200Mhz to 4000Mhz its mostly only around 5%.

I feel the testing that really should have been to see if latency or bandwidth is more important is things like 3200CL16 vs 3000CL14.

Also I feel it should perhaps have been a 2400 vs 3000 vs top end, as I think even cheapo ddr4 is faster than 2133 given the jedec spec is higher? my ocuk value ddr4 ram I got for my old mining rig was 2400mhz, and the issue with 3200 vs 3000 is you need samsung B die usually for advertised 3200 kit and there is a definite significant price bump for that. But not so much to get to 3000.

 

[chrcoluk]

I never tested the different RAM speeds i got from my FlareX in gaming, since i only have a low tier gpu. But in Cinebench, the difference from 3200 to 3400 RAM speed is like a 100MHz more of cpu oc. R7 2700 oc'ed to 4.1GHz.

3200 MHz = C15 MT score of 1870
3400 MHz = C15 MT score of 1900

Interestingly, not much difference in ST results.

https://i.imgur.com/42DiGwA.jpg

just remember cinebench isnt reflective of real world stuff. Its a convenient test, but dont rely on it to assess real world performance. Even less so for ryzen given it messes up XFR.

Was the 3200 to 3400 keeping same timings?

Based on this table someone generated tho it suggests my 3000CL14 should be better than 32000CL16 as it provides a lower access time.

https://i.imgur.com/6ZSDax7.png

3200CL16 - 10ns access time
3000CL14 - 9.33ns access time
3000CL16 - 10.67ns access time (stock settings of ram)

 

[VortexA1]

I have built a computer with the Ryzen 2600x and I get 10s on Super Pi 1m( I have the memory running at 3000mhz ).Whereas with my other computer which is an i5-2500k overclocked to 4.1ghz I get 9.5 seconds..is that because Super Pi isn't optimized for Ryzen.. I love Ryzen ..just a bit surprised..

I just read that because the program is old it doesn't account for Ryzen.. so that's the answer.

 

[Orange Nexus]

just remember cinebench isnt reflective of real world stuff. Its a convenient test, but dont rely on it to assess real world performance. Even less so for ryzen given it messes up XFR.

Was the 3200 to 3400 keeping same timings?

Based on this table someone generated tho it suggests my 3000CL14 should be better than 32000CL16 as it provides a lower access time.

https://i.imgur.com/6ZSDax7.png

Yes, the timings on both 3200 and 3400 are same. In an earlier post i saw the 3200 Cl16 (happens to be B-die) has same latency as my 3266 Cl14. So, that tells me that it must be the quality of the RAM. Cine might not reflect exactly how RAM performs in games but as shown in that post speed and tight timings do indeed help overcome the issue with infinity fabric. Something that has been discussed a number of time in several forums. So, if we see reviewers test Ryzen 1000 and 2000 series using oc'ed rams but loose timings, we should know that the results won't be the same when tested at same speed but tighter timings. You see it all the time. For example, testing with the RAM Oc'ed to 3466 but using Cl16 - the result would be somewhat gimped. Just recently i saw a reviewer use 2400 MHz RAM kit and oc'ed it to 3333 Cl20. Might as well lv the RAM at default. I've been running my FlareX at 3266 Cl14 since Sept 2017.

 

[Orange Nexus]

I have built a computer with the Ryzen 2600x and I get 10s on Super Pi 1m( I have the memory running at 3000mhz ).Whereas with my other computer which is an i5-2500k overclocked to 4.1ghz I get 9.5 seconds..is that because Super Pi isn't optimized for Ryzen.. I love Ryzen ..just a bit surprised..

I just read that because the program is old it doesn't account for Ryzen.. so that's the answer.

https://mygaming.co.za/news/hardwar...roves-amd-super-pi-benchmark-performance.html

 

[chrcoluk]

for reference I flipped my ryzen 2 rig now to 3000CL14. Since that has the better native latency over 3200CL16.

 

[humbug]

https://mygaming.co.za/news/hardwar...roves-amd-super-pi-benchmark-performance.html

I love the casual nature of this paragraph...

The Bulldozer Conditioner does not permanently fix x87 performance – that is up to motherboard vendors to update their BIOS to fix the problem, The Stilt says. For a slightly more permanent solution, there’s the Intel Compiler patcher, which scans executables for games and programs and disables code that causes a performance drop on non-Intel processors. It goes without saying that MyGaming is not responsible for any mishaps that may arise as a result of your use of the patcher.

Intel's compilers deliberately scan for none Intel CPU's and then puts them on a gimp code to slow their performance.