VOGONS


First post, by DeepestBlueDragon

User metadata
Rank Newbie
Rank
Newbie

Hello folks,

I'm seeking some assistance with understanding differences in performance between 86Box and PCem.

Some basic information. My host PC is an i7-8086K, 64GB RAM, SSD boot and VM storage drives, 3060Ti, W10.

I have tried to set up both 86Box and PCem to emulate a PII-400, 256MB RAM, 16GB IDE HDD, S3 Virge DX 4MB, SB or compatible, no parallel/serial, Realtek NIC, running Windows 98 SE. OS installed the same way using the same media. No updates, just the base OS with built in drivers.

I've noticed that despite setting the two emulators to be as similar as possible, they are performing very differently. 86Box takes a LONG time to get into Windows (several minutes), whereas PCem boots in within a minute or two. However disc access in 86Box is significantly faster. I downloaded the DOSBENCH suite and ran it on both emulators. The results are attached in the screenshot and are also below.

It would seem that the way the emulators work seem to mean certain tests perform better on one than the other.

86Box performs faster in these tests:
- Landmark's CPU test (33% faster)
- Speedsys (15% faster)

PCem performs faster in these tests:
- Topbench (217% faster)
- Landmark Video test (231% faster)
- PC Player (26%)
- Chris's 3D Bench (24%)
- 3D Bench Superscape (23%)

On top of these, I booted up MHDD and did a read test. 86Box reported 914.8Mb/s, vs 62.2Mb/s in PCem. I suspect the difference here is because 86Box has an option for "Generic RAM Disk (Max Speed)" option in its HDD settings, whereas PCem doesn't.

So my questions here are:

1. Am I right in thinking the differences in performance are just down to how the two emulators work?
I love the customisations in 86Box, but PCem seems to perform better overall. Is there any way to
2. ...boost PCem's HDD speed?
3. ... increase 86Box's overall performance?

Thanks for the questions, and apologies if they're rather n00b-like; this kind of emulation is new to me. I'm more used to Proxmox and VirtualBox!

Reply 1 of 12, by DeepestBlueDragon

User metadata
Rank Newbie
Rank
Newbie

(Just noticed the MHDD line shows KB/S, should be Mb/s)

Reply 2 of 12, by jh80

User metadata
Rank Member
Rank
Member

What versions of 86Box and PCem? And more importantly, which dynamic recompiler (or interpreter?) in 86Box? I think the new dynarec is similar to (or same as) PCem's, while the old dynarec is retained from before PCem updated.

Because (by default) both emulators have the dynamic recompiler enabled, what those CPU benchmarks are measuring is more about how efficient the dynarecs are in processing certain code. One emulator might be better in one context but worse in another depending on the code that is being processed by the dynarec. That's also why those CPU benchmarks also don't reflect necessarily how each emulator will handle a specific game. Essentially, each and every game will likely have slightly different performance.

Reply 3 of 12, by Battler

User metadata
Rank Member
Rank
Member

Also, 86Box and PCem have entirely different Pentium II timings - 86Box's were written by Plant based on the K6 timings and tweaks using real Pentium II benchmarks as a reference.

PCem also defaults its video timings to 8-bit ISA, which may yield increased % at the expense of much slower video memory affected - setting them to the option named Default (yes, it's confusing!) should bring them on par.

Reply 4 of 12, by dosmax

User metadata
Rank Member
Rank
Member

My practical approach to the matter is: stick to the latest PCEm when performance counts and it doesn't hurt, i.e. for playing demanding games.

IIRC 86box said they favor correctness over speed, which is a valid choice. PCEm though is still faster in almost every demanding game. So unless you don't need anything only 86box offers, and your app/game runs with PCem, there isn't really a good reason not use it whenever performance counts.

A shame that the PCEm dev got alienated away by some fights/events over the 86box fork a few years back. A wizard (or sorceress IIRC) like that is hard to replace.

Reply 5 of 12, by DeepestBlueDragon

User metadata
Rank Newbie
Rank
Newbie

Appreciate the explanations, thanks. So it looks like I need to choose which one I use will depend on application. That's a shame, I was so hoping to have just one that does it all!

I'm using 86Box v6.0 (build 9001) with "old dynarec", and PCEm v17 with "Dynamic recompiler enabled" (check box, which is enabled and greyed out). If there's a way of getting more detailed info that's needed, do let me know and I'll get it!

Reply 6 of 12, by jh80

User metadata
Rank Member
Rank
Member
DeepestBlueDragon wrote on 2026-10-09, 05:39:

Appreciate the explanations, thanks. So it looks like I need to choose which one I use will depend on application. That's a shame, I was so hoping to have just one that does it all!

I'm using 86Box v6.0 (build 9001) with "old dynarec", and PCEm v17 with "Dynamic recompiler enabled" (check box, which is enabled and greyed out). If there's a way of getting more detailed info that's needed, do let me know and I'll get it!

It probably only matters if you're trying to maximize which CPU you can emulate. I just go for a more modest approach that covers my needs.

You can try the 86box new dynamic recompiler ("NDR" version on Github, also here: https://86box.net/builds). It should run faster.

As I recall, anything Pentium or higher will require the dynarec.

Reply 7 of 12, by dosmax

User metadata
Rank Member
Rank
Member

Just don't expect too much from 86box's NDR version.

In terms of what is possble to safely run on my rig (Ryzen 9 7950X) it's still a ~100 MHz gap, i.e. a 300MHz CPU on 86box vs. 400MHz CPU on PCem. That's enough to keep PCem around for demanding stuff. Maybe one day 86box find someone capable enough to get their correctness first approach up to PCem's speed, but that day has not come yet.

You mileage may vary depending on your host rig, what you actually run etc, but I don't think 86box is a valid replacement yet at the upper end of the performance spectrum. It's way ahead by now in terms of features and user-friendlyness, but with performance in mind it's still no match to PCEm.

Reply 8 of 12, by Battler

User metadata
Rank Member
Rank
Member

Did you set the video timings in PCem to something that's not the "8-bit ISA" they default to? Because that make you able to emulate a faster CPU at the expense of the emulated graphics being far slower due to far slower video memory access. Also make sure your renderer in 86Box is set to OpenGL and not Qt Software.

Also, which game on which emulated OS is it that shows such a difference? And would you be able to post a configuration file? Actual examples would be nice so we can at least attempt to bisect and see when the regression may have been introduced.

And lastly, are you comparing 86Box to PCem v17 or the v18 development builds? Because from what I hear, the latter are themselves slower than v17, chiefly due to the (S)VGA accuracy improvements committed at the beginning of v18 development that have been ported to 86Box as well, though the compiler updates (v17 was compiled with ancient GCC 4.7.x while v18 is compiled with something more modern) and the corresponding run-time library updates, have contributed as well, which is also true for 86Box which we compile, on x64 Windows, with the latest GCC available in MSys2, which as of writing this post, is 16.2.0-4.

There's been at least one case where run-time library updates have slowed down the emulator. We worked around it by rewriting the relevant portion of x87 emulation in ASM, but there may have been more such library-induced performance regression. Hence my question about what exactly shows the difference so we can attempt to pinpoint the cause and address it.

Reply 9 of 12, by dosmax

User metadata
Rank Member
Rank
Member

I'm running a bunch of tests every once in a while against a PCem version I have from 2020, which according to its readme is v17 with Windows 95/98.

You should be able to verify these results easily e.g. with 3dmark 99, while having at eye on which emulator is able to run it at what CPU frequency and stable 100% emulation rate. Now you may say that's just a benchmark, but for one it's not the only test, and secondly I dont't really care. I check what I check and consider best what works for me at the highest CPU frequency and 100% emulation rate, PCem wins consistently even against 86box NDR.

I appreciate all the work you and all the contributors of 86box do. It has come a long way and is obviously better than PCem in many respects. Just not performance wise, at least not for me (or the OP here as their benchmarks show).

Let's keep it at that. Considering the experience the original PCEm dev made with talking to you back in the day, I won't dive into any further discussion here.

Reply 10 of 12, by DeepestBlueDragon

User metadata
Rank Newbie
Rank
Newbie
Battler wrote on 2026-10-09, 15:31:

Did you set the video timings in PCem to something that's not the "8-bit ISA" they default to?

It's set to "default". Next time I get time to play, I'll see what the various options do.

Also make sure your renderer in 86Box is set to OpenGL and not Qt Software.

Noted!

Also, which game on which emulated OS is it that shows such a difference?

It's the DOSBENCH suite running on Win98 DOS. Booted to DOS, not in a Window under Windows. Fuller details in my OP.

And would you be able to post a configuration file?

From 86Box? I don't know that DOSBENCH has specific config files.

And lastly, are you comparing 86Box to PCem v17 or the v18 development builds?

I'm using 86Box v6.0 (build 9001) with "old dynarec", and PCEm v17 with "Dynamic recompiler enabled" (check box, which is enabled and greyed out). If there's a way of getting more detailed info that's needed, do let me know and I'll get it!

Reply 11 of 12, by Battler

User metadata
Rank Member
Rank
Member

To make the comparison apples to apples, you should be comparing 86Box v6.0 with new dynarec to PCem v18 builds, with the graphics card timings set to "default" in PCem.

From 86Box? I don't know that DOSBENCH has specific config files.

No, I meant 86box.cfg (and PCem's equivalent).

Reply 12 of 12, by Battler

User metadata
Rank Member
Rank
Member
DeepestBlueDragon wrote on 2026-09-30, 07:10:
PCem performs faster in these tests: - Topbench (217% faster) - Landmark Video test (231% faster) - PC Player (26%) - Chris's 3D […]
Show full quote

PCem performs faster in these tests:
- Topbench (217% faster)
- Landmark Video test (231% faster)
- PC Player (26%)
- Chris's 3D Bench (24%)
- 3D Bench Superscape (23%)

Just a question - what graphics card are you emulating in 86Box and what in PCem?

Edit: I saw now:

I have tried to set up both 86Box and PCem to emulate a PII-400, 256MB RAM, 16GB IDE HDD, S3 Virge DX 4MB, SB or compatible, no parallel/serial, Realtek NIC, running Windows 98 SE. OS installed the same way using the same media. No updates, just the base OS with built in drivers.

There may be something in our P6 timings.