Clean install of DOSBox 0.74-3 with cycles=max on my i7-6700K@4.4GHz.

D1:
Microseconds 1 loop: 2.20
Dhrystones / second: 455192
VAX MIPS rating: 259.07

D2:
Microseconds 1 loop: 2.20
Dhrystones / second: 455192
VAX MIPS rating: 259.07

Both results used to be compareble, so I guess something has affected your second result

Hmm, interesting, I run that benchmark on my ThinkPad T470 with i5-7300U, and got these (comparable) results.

D1:
Microseconds 1 loop: 3.52
Dhrystones / second: 284444
VAX MIPS rating: 161.89

D2:
Microseconds 1 loop: 3.60
Dhrystones / second: 277657
VAX MIPS rating: 158.03

I'll try that once again on my main PC at home, but I run that D2 test multiple times yesterday, and always got the same results.

maybe check power saving options
or leave dosbox running for a while to give some time to warm cpu up 😉
check cpu stats in task manager

Alright, I don't know why, but I rerun that benchmark, and I got comparable result... Or, exactly the same one 🤣... I don't know, what shenanigans possessed my PC yesterday 😁

don't you happen to have "max limit ..."?

On both machines I run benchmarks with cycles=max.

Using my notebook Ryzen 7 2700U
D1:
Microseconds 1 loop: 5.41
Dhrystones / second: 184971
VAX MIPS rating: 105.28

D2:
Microseconds 1 loop: 5.36
Dhrystones / second: 186698
VAX MIPS rating: 106.26

DOSBox is made for games. It's more interesting to test it in them. Other tests are additional to that.
Quake and doom were used by me in the signature's theme. Such results can be compared with stats of wide range of real machines.

This one is testing cpu speed, the main problem of dosbox cause there are already dosbox versions with virtualised/accelerated 3D GFX (Direct 3D/OpenGL/Glide)
It is also easy to use and rather reliable and can be performed on real PCs.

Not that there was much reason to test it. More powerful core means more power in dosbox. I just started this thread to get some latest desktop Ryzen (5/7) 3XXX results that are said to be more powerful than intel right now.

I will be putting together my B450+Ryzen5/2600 PC tonight, I'll be contributing to this thread pretty soon I suppose 😀

Ryzen 5 2400G, DOSBox 0.74-3 (from Fedora repo) on Fedora 29 x86_64

D1:
Microseconds 1 loop: 5.58
Dhrystones / second: 179272
VAX MIPS rating: 102.03

D2:
Microseconds 1 loop: 5.45
Dhrystones / second: 183644
VAX MIPS rating: 104.52

DOSBox SVNr4267 (which includes jmarsh's 64-bit dynamic_x86 patch), built with gcc-8.3.1:

D1:
Microseconds 1 loop: 2.87
Dhrystones / second: 347826
VAX MIPS rating: 197.97

D2:
Microseconds 1 loop: 2.92
Dhrystones / second: 342704
VAX MIPS rating: 195.05

i5 2500k @ 4.2GHz

Dynrec

Microseconds 1 loop: 5.53
Dhrystones / second: 180791
VAX MIPS rating: 102.90

Microseconds 1 loop: 5.41
Dhrystones / second: 184971
VAX MIPS rating: 105.28

Dynamic (jmarsh patch for 64 bit)

Microseconds 1 loop: 3.76
Dhrystones / second: 266112
VAX MIPS rating: 151.46

Microseconds 1 loop: 3.46
Dhrystones / second: 289266
VAX MIPS rating: 164.64

Ryzen 7 3700x

32-bit 0.74-3 (Windows)

D1:
Microseconds 1 loop: 3.24
Dhrystones / second: 308806
VAX MIPS rating: 175.76

D2:
Microseconds 1 loop: 3.45
Dhrystones / second: 289593
VAX MIPS rating: 164.82

Not exactly spectacular, and there seems to be a fair amount of variance between runs. However...

64-bit build of SVN r4271

D1:
Microseconds 1 loop: 2.04
Dhrystones / second: 490046
VAX MIPS rating: 278.91

D2:
Microseconds 1 loop: 2.04
Dhrystones / second: 489297
VAX MIPS rating: 278.48

Much better! My timedemo results with Quake were also significantly better with the new 64-bit build.

Silanda
it would be good if you could type "core" inside your 32-bit 0.74-3 (Windows)
to verify your core setting

It's dynamic; I just retested with core explicitly set to dynamic and experienced similar results. The 32-bit version simply doesn't seem to perform brilliantly on Ryzen. I was experiencing the same thing with my own builds.

Thing is though, I noticed with Quake that I could push the cycles significantly higher than max seemed to be setting them judging by timedemo results. I tried this by increasing cycles until the audio started to skip and then backed off.

always use
cycles=max 105%
(it is actually 95%)
max is just 90%
10% was left unused for safety in the past when auto cycles where not working as good as now

also just in case check if your power profile is set to high performance

None of those things have much if any impact. Watching the core speeds in Ryzen Master as the benchmark runs suggest that load is being split between two cores, but neither core is approaching being maxed out. I suppose the Windows scheduler could be playing a role, but setting affinity to two threads on the same core results in the same performance even though the core in use does ramp up to a higher clockspeed than it would otherwise.