We need AMD Ryzen Benchmarks! Share your toughts!

[djotacon 190]

4 hours ago, Mahatma Gandhi said:

@djotacon

 

Nice! Did you work on the subtimings of the RAM also?

 

Not right now.

 

First I need to fix some issues with my config:

 

* Change my old mobo ( 350 to 450 chipset)

* Fix the constant problem of irregular fps: based on my own research this can be easily fixed ( see the video irregular fps).

* Fix the fps spikes ( video min 1.52-1.56 ) I think the power source isnt very good.

* Fix the heat ( Madrid right now 35º-40º in my room)

 

etc,etc..

 

When I fix all this problems I try to make a good memory tunning to improve even more the fps.

[Mahatma Gandhi 29]

19 hours ago, djotacon said:

When I fix all this problems I try to make a good memory tunning to improve even more the fps.

 

Let me know, I can probably help tuning. I have quite some experience with RAM tuning on my Ryzen build.

 

Here's some helpful list of settings and results: 

 

https://docs.google.com/spreadsheets/d/1HKPVfDcFO-aieAOXHFQZp15rwWadbPTVDNgO8vtyDCM/edit#gid=0 

[djotacon 190]

I make a test related with memory latencies using intel memory latency checker:

Intel(R) Memory Latency Checker - v3.5
Measuring idle latencies (in ns)...
Memory node Socket       0     0    64.1

 

64.1 I think is an good result using only the XMP profile, I dont see any improvement adding more tunning.

 

I think this link can be useful to low latency hunters like us:

 

http://www.qdpma.com/ServerSystems/MemoryLatencyII.html

[Mahatma Gandhi 29]

Ryzen 2700x @4.2GHz
ASUS Prime X470-Pro
2x8gb DDR4 @3466MHz
Aorus 1080ti
Standard Preset - Standard View Distance
57.4 FPS
https://ibb.co/eXnLkU

 

 

[Groove_C 267]

@Mahatma Gandhi

[Groove_C 267]

Standard Preset - Standard View Distance

 

[Groove_C 267]

Ryzen is a very good option for those, who are still sitting on their 2/3 gen Intels, in gaming. And for all those, who are not just gaming.

My next cpu will probably be a Ryzen 3xxx.

[oldbear 390]

Still looking for the real playable "Minimum" requirements I had done tests on an A8-9600 APU, a Ryzen 3 1300X and a Ryzen 3 2200G based rigs using a RX550 2GB

ATM, I am still doing tests on the R3 2200G using a RX 460 4GB.

 

Results with the A8-9600 where rather low but the Ryzen 3 1300X and Ryzen 3 2200G with a basic OC and 3000 MHz RAM are doing well.

 

 

I must add that playing  integrated Vega on the Ryzen 3 2200G is not giving a satisfying game experience due to FPS drops under the 20 FPS level.

 

 

But using a small RX 550 2GB the experience is good .

 

 

 

 

Edited October 2, 2018 by oldbear
I am not really a bear, but feeling old sometimes

[telejunky 0]

2700x @ Stock with 3200CL14 and RX480 Red Devil gives me around 51FPS in YAAB with standard settings incl. press s @1080p

(Same settings with 1800x were only ~42 FPS)

[Mahatma Gandhi 29]

On 10/19/2018 at 8:39 PM, telejunky said:

2700x @ Stock with 3200CL14 and RX480 Red Devil gives me around 51FPS in YAAB with standard settings incl. press s @1080p

(Same settings with 1800x were only ~42 FPS)

 

You can easily get another 10%-15% if you tune your RAM Timings.

 

See --> https://www.overclock.net/forum/13-amd-general/1640919-ryzen-dram-calculator-1-1-0-beta-2-overclocking-dram-am4.html 

 

50 FPS, I get with an old Ryzen 1700 @ 3.8GHz with these timings (Same RAM as you have @ 1.4V vRAM; Radeon RX470 4 GB):

 

 

 

Let me know if you have questions regarding this...

 

[Valken 622]

Been a while but I was wondering has anyone ever tested with any of the Thread Ripper CPUs with QUAD CHANNEL RAM? Would love to see the FPS scaling with more RAM bandwidth. Those TR and Ryzen CPUs are incredible value these days.

[oldbear 390]

From what I have seen, FPS in Arma3 are related to CPU efficiency and frequency so it will probably more interesting to look for a 2700X with some OC and fastest RAM authorized by MoBo.

[Mahatma Gandhi 29]

@Valken

 

I have no data for threadripper but e.g. the x299 Plattform (also QuadChannel) does not scale well. The ArmA3 performance is more latency dependent than throughput. That beeing said, ArmA3 of course runs well on X299 or TR4 Plattforms as well.

 

@oldbear

 

We recently tested that on a 2700x from a friend:

 

Ryzen 2700x
ASUS Prime X470-Pro
2x8gb DDR4 @3466MHz
Standard Preset - Standard View Distance
57.4 FPS
 

https://ibb.co/eXnLkU 

 

This is absolutely as is, no startup parameters like e xtra threads etc.

[Groove_C 267]

@Mahatma Gandhi

We need to check if more even/consistent frametimes with Ryzen 5/7 are due to more cores/threads or simply because of 16 MB L3 cache vs. 6/9 MB for i5 or 8 MB for i7.

 

Desactivate enough cores to have just 4 + 4 threads and check frametimes.

 

If frametimes will remain that even/consistent even with less cores/threads then it's just because of more cache.

 

But i don't know if all 16 MB L3 cache are accessible even to fewer cores or if it's 2 MB per core.

[Mahatma Gandhi 29]

@Groove_C

@oldbear

 

I have just switched the system to a 2600x which allowed me to clock my RAMs to 3600MHz. The upgrade made a huge difference. I got another + 20%

 

The YAAB result with the CPU at stock and the RAM @ 3600MHz in 1080p and Standard preset is:

 

 

Rest of the System is still identical:

Now Ryzen 5 2600x (stock) (before Ryzen 7 1700 @3.8GHz)

Radeon RX 470 4gb

Asrock X370 mITX

G.Skill FlareX 3200MHz CL14 (now overclocked to 3600MHz CL15)

Samsung SSD

Enermax Triathlor 300W

 

[Groove_C 267]

@Mahatma Gandhi

Please test frametimes and upload a graph.

Then disable 2 cores, run YAAB and also make frametimes graph.

This way we will see if frametimes will be same or if there is a difference.

If no difference with 2 cores less, then it's just because of 16 MB L3 cache.

[Mahatma Gandhi 29]

8 hours ago, Groove_C said:

Please test frametimes and upload a graph.

Then disable 2 cores, run YAAB and also make frametimes graph.

This way we will see if frametimes will be same or if there is a difference.

If no difference with 2 cores less, then it's just because of 16 MB L3 cache.

 

Regarding the particularly smooth frame delivery on Ryzen. I can try doing this, but my concern is that YAAB is not reproducible enough for frame time measurements. Any suggestions for a handy tool to measure frame times?

 

On 1/9/2019 at 11:42 PM, Groove_C said:

But i don't know if all 16 MB L3 cache are accessible even to fewer cores or if it's 2 MB per core.

 

The L3 is fully available to all cores, however it is split in two blocks (2x8mb). Each block of 4 Cores (AMD calls that a CCX) have local fast access to 8mbL3 and somewhat slower access to the 8mb of the other CCX. If I disable cores, I can do that in pairs only (8, 6, 4, 2) because they get disabled symmetrically (at least this is true for my Board). That however means that distant access might become more frequent and bias the result. Not sure that is a trivial question to solve this way.

 

But sure, if there is a handy tool to measure frame times, I can give it a shot.

 

[Valken 622]

@Mahatma Gandhi You are pulling between 10 to 15+ FPS more than I do at 4.6 GHZ/4GHZ Core/2400CL10 DDR3 RAM with a 1060GTX @ standard... I am unsure if its because you have more cores or the RAM scaling is so good but fantastic system!

 

Edit for reference - only goes to show how GOOD Ryzen really is... my next upgrade system for sure...

 

 

Intel Haswell i7-4790k CPU ~ 4.5 GHZ,  Cache 41x, CPU Multi 46x, 98MHZ bus speed due to OC RAM: 

G.Skill DDR3-2540 MHZ CL10 OC'ed, 32GB!!! 1.65 volts stock.

1060 3GB GTX

 

 

Intel Haswell i7-4790k CPU ~4.6 GHZ/ 4.0 GHZ cache / 2400CL10 32GB / 100 MHZ FSB

1060 3GB GTX

 

[Mahatma Gandhi 29]

@Valken

 

I am very happy with the system and I can absolutely recommend this platform for ArmA3/in general. The price/performance tag is pretty outstanding.

 

But yes, RAM is a very contributing factor at least on the AMD Plattform. So if one is thinking about upgrading to the Ryzen platform, don't be cheap on the RAM. One saves a lot on the CPU already. Although expensive, 3200MHz CL14 modules make the cut over 3000-3200MHz CL16 which you cannot clock or further optimize timings. But I am also surprised, because your RAM is pretty competitive, too. With 2400MHz CL10 you have identical latencies to my 3600MHz CL15.

 

I can do some core count scaling over the next week if thats of interest. My feeling is, it matters most for the frame times, in particular if other software runs in the background (e.g. Skype and Relive or my Dropbox sync). The average FPS result is likely stable down to 8 threads, as long as nothing else runs in the background.

[Groove_C 267]

@Mahatma Gandhi
I think it's possible to capture frametimes with FRAPS and then make a graph using Excel.

But I'm not sure about this. Never made this.

I'm most interested in 4 cores (8 threads) frametimes graph with Ryzen 2xxx vs. 6/8 cores (12/16 threads) frametimes graph.

 

Better use Altis Benchmark 0.60 - it's a little more consistant than YAAB.

https://steamcommunity.com/sharedfiles/filedetails/?id=173435011

 

And also kill all possible processes running in the background for better comparable results.

[Mahatma Gandhi 29]

Alright,

 

I did use the Altis Benchmark and benched 3 times each 6C/12T, 6C/6T, 4C/8T and 4C/4T.

 

The result by Fraps for the first 60 seconds is basically identical. Even between Full 6C/12T or 4C/4T

 

 

However, if you look at the frame run, 2/3 runs with 4C/4T or 4C/8T have huge spikes. Here a representative one:

 

 

Heres another one:

 

6C/12T are all smooth

 

 

So it seems that you need more than 8 Threads to avoid stutter entirely, but this might be a wrong conclusion as threads from ArmA3 might jump CCX more frequently and this might be the cause. But you can also see the 1% low Frametimes go up from 4C/8T to 6C/12T

[Groove_C 267]

@Mahatma Gandhi

So 12/16 threads result in smoother gameplay.

 

Imagine the FPS and gameplay smoothness with a Ryzen 3xxx 6C/12T )

[Mahatma Gandhi 29]

@Groove_C

 

Yes, It seems like it. And this was with every background programm closed. For sure I haven't seen any stutter since I have this system but while benching with 4 cores, I have seen noticeable stutter which are in the results above.

 

Not sure Ryzen 3k can be any smoother to be honest. But yeah, I am also looking forward to a YAAB Bench with one of those. My next thing is the GPU, as I don't play yaab all day and in real world, I am entirely GPU bound now.

 

Or I just stay with what I have and enjoy ArmA

[Groove_C 267]

@Mahatma Gandhi

Ryzen 3xxx won't make your gameplay smoother, but will deliver more FPS 100%, because of its higher frequency. And maybe it will as well have a better memory controller for even higher RAM frequencies.

 

If you plan to stay @ 1080p, a Sapphire RX 590 Nitro+ 8 GB for 270 € would be the best option to play on ultra settings.

 

But I would wait for Navi as it's not that far instead of buying "old" tech, since you already have a RX 570 and it's playable for the moment.

[Mahatma Gandhi 29]

Retrospectively, the RX470 4gb I have was a very decent buy (back in 2016). Last weekend, we have been playing the "A long way" Coop mission and I am on average well above 80 FPS, mostly 90-110FPS with the "Very high preset" + HBMO @ strong and View Distance set to 8000m.

 

The GPU shows a constant 100% load, so I am GPU bound, but at  this level... It simply does not matter anymore to me. I am to old meanwhile to see a difference between 80 and 120 min FPS. The most important aspect to me is that it is completely smooth. Fully even frame times matter much more than 15% average FPS. I have always hated these stutters in the OFP/ArmA3 engine and this is the first build I got completely rid of them. So Kudos to AMD for finally bringing an affordable and fast platform with many cores to the marked.

 

If ArmA4 ever comes to market, I might go for the Ryzen 3k if necessary (lets see how the enfusion engine works). Luckily, they announced that it will be a drop in replacement that works on any current AM4 Board and apparently, it will even upgrade the Boards to PCIe Gen4.

 

Nothing else to wish for.

 

By the way, in the other thread, you wrote that the 2600x only gives 35 FPS @ Ultra setting. I checked that yesterday, too. So with Ultra Settings applied and no reduced View Distance (no S-key) but disabled AA (otherwise the RX470 is a hard limit) I end up with 46.8 FPS on average in YAAB but it remains fully smooth.

 

Don't ever be cheap on RAM if you play ArmA ;-) Thats the most common mistake people do when configuring their systems.