Well, precompiling only makes real sense on a multicore system. When the same core needs to both compile code and execute it, there's not much of an advantage. One more or less requires the other.Peter wrote: Sun Aug 10, 2025 9:36 pmI think the massive speed advantage mostly comes from the JIT compiler, only to a lesser degree from the multicore support.tofro wrote: Sun Aug 10, 2025 4:10 pm UAE is in fact very fast as it uses more than one CPU core to pre-compile 68k code into x86 code in parallel with running the actual 68k emulation. I don't think QPC2 is actually using more than one CPU core.
QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
ʎɐqǝ ɯoɹɟ ǝq oʇ ƃuᴉoƃ ʇou sᴉ pɹɐoqʎǝʞ ʇxǝu ʎɯ 'ɹɐǝp ɥO
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
There is a large advantage even with a single thread since each piece of code is only compiled once, but can run potentially millions of times. Most programs have loops where they spend the majority of time.tofro wrote: Mon Aug 11, 2025 9:10 amWell, precompiling only makes real sense on a multicore system. When the same core needs to both compile code and execute it, there's not much of an advantage. One more or less requires the other.Peter wrote: Sun Aug 10, 2025 9:36 pmI think the massive speed advantage mostly comes from the JIT compiler, only to a lesser degree from the multicore support.tofro wrote: Sun Aug 10, 2025 4:10 pm UAE is in fact very fast as it uses more than one CPU core to pre-compile 68k code into x86 code in parallel with running the actual 68k emulation. I don't think QPC2 is actually using more than one CPU core.
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
Surely not. I tried UAE with JIT compilation in the early 2000s when my PC was still single core and remember that emulation was roughly 3-4 times faster.tofro wrote: Mon Aug 11, 2025 9:10 am Well, precompiling only makes real sense on a multicore system. When the same core needs to both compile code and execute it, there's not much of an advantage. One more or less requires the other.
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
The Q-Liberator malaise again... I just saw that I still have "Qemulator Fast" on my PC, but unfortunately, it does not seem to like my current license.M68008 wrote: Mon Aug 11, 2025 1:03 am Yes, in 2001-2002 I wrote a prototype version of Q-emuLator that compiled basic blocks of 68K code to x86. I only made it available for a short time and dropped it since it was not compatible with some self-modifying QL programs. Interpreters are also already pretty fast on modern PCs.
Last year I took another look at it, but noticed that QLE doesn't even start anymore on it, probably because it runs a number of Q-Liberated programs and Q-Liberator seems to produce some self-modifying code.
Would be cool to compare with UAE.
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
A low-end i5 with 8 GB RAM, somehwat slower than Per's even older box, according to the factors he mentioned.
The emulated Amiga is set to A4000, 68030, 2MB chip RAM, 8 MB fast RAM, fastest possible CPU emulation.
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
More that fast enough. I just ran the benchmark out of academical interest.
QDOS Classic is often overlooked, so I was tempted to demonstrate the performance.
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
Might be easier to actually use UAE and define a QL-like "Amiga" hardware model.XorA wrote: Sun Aug 10, 2025 10:30 pm I did look at it, but incorporating the UAE JIT is not a simple task as its quite tightly coupled to UAE!
But I think emulation speed has become too irrelevant.
Maybe except 68K Linux and compiling kernels/GCC natively.
- Mark Swift
- Bent Pin Expansion Port
- Posts: 93
- Joined: Fri Jul 18, 2014 9:13 am
- Location: Blackpool, Lancs, UK
- Contact:
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
Hi Peter,
I do have a little info on the early QDOS Classic JSßs UAE ROM that should be mentioned.
The original Amiga QDOS Classic clock routine made use of hardware timers to emulate the QL clock.
In UAE, these hardware timers are emulated; so as a result in early UAE ROMs time itself is emulated.
This is not good when measuring benchmarks.
Later Amiga QDOS Classic versions read the Amiga RTC to get the time.
In UAE, the system time is read from the host machine whenever the Amiga RTC is read; so as a result time is accurate in later builds.
With this in mind I re-ran the tests.
This showed that the older UAE ROM got the timings wrong by a factor of 1.16
Timings below seem to show that JSQC/UAE/JIT runs approximately 1.7 times the speed of QPCII
However since SuperBasic on HBA is much quicker than on JS, QPCII feels around 2.64 quicker than JSQC/UAE/JIT on the same hardware; as the GladTimings scores below show.
The following tests were run on my wifes Lenovo Laptop with an Intel Core i5-1240P.
My M1 MacBook Pro achieves only a fraction of these results, but is quick enough for me.
Qtop-Index scores
GladTimings_bas scores
It does make me wonder who is running the fastest QL though, and what our 1980s selves would have done with all that power.
The air-time is much welcome, and I've been following this thread with interest.Peter wrote: Mon Aug 11, 2025 1:42 pm QDOS Classic is often overlooked, so I was tempted to demostrate the performance.![]()
I do have a little info on the early QDOS Classic JSßs UAE ROM that should be mentioned.
The original Amiga QDOS Classic clock routine made use of hardware timers to emulate the QL clock.
In UAE, these hardware timers are emulated; so as a result in early UAE ROMs time itself is emulated.
This is not good when measuring benchmarks.
Later Amiga QDOS Classic versions read the Amiga RTC to get the time.
In UAE, the system time is read from the host machine whenever the Amiga RTC is read; so as a result time is accurate in later builds.
With this in mind I re-ran the tests.
This showed that the older UAE ROM got the timings wrong by a factor of 1.16
Timings below seem to show that JSQC/UAE/JIT runs approximately 1.7 times the speed of QPCII
However since SuperBasic on HBA is much quicker than on JS, QPCII feels around 2.64 quicker than JSQC/UAE/JIT on the same hardware; as the GladTimings scores below show.
The following tests were run on my wifes Lenovo Laptop with an Intel Core i5-1240P.
My M1 MacBook Pro achieves only a fraction of these results, but is quick enough for me.
Qtop-Index scores
Code: Select all
On JSßs/UAE/JIT = 4422.975 (skewed timing)
On JSQC/UAE/JIT = 3809.026 (actual timing)
on QPCII = 2240.744
Code: Select all
On JSßs/UAE/JIT = 10776.7 (skewed timing)
On JSQC/UAE/JIT = 6040.5 (actual timing)
on QPCII = 15958.8
I agree that when emulation is running thousands of times quicker than the original hardware; stats become a little academic.
It does make me wonder who is running the fastest QL though, and what our 1980s selves would have done with all that power.
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
I did a comparison in May, for a video, of my Amiga 600 (1MB WB 2.0) to a Lenovo i3 at 2.26 GHz (4GB 2024 Mint Linux). I was looking at job queue management, hitting Linux with over 15,000 jobs to see how it handled them. (Note the Amiga was unusable after 5 jobs, but my 128K JSU QL ran 43 jobs easily which when extrapolated would have far exceeded Linux.)
I ended up estimating the Lenovo at about 3500 times faster than the Amiga using old benchmarks, comparable modern benchmarks, and connecting benchmarks between the old and new. It took a bit of extrapolation but I was trying to get as best as one could do comparing old and new technologies (esp with the difference of single vs multi-core).
In any case, Mark's comment at the end made me think it would be cool to see who could get the fastest (accurate) benchmark of a QL system using whatever means. I like Peter's approach of using WinUAE with QDOS Classic since it's not double emulation and WinUAE may be the fastest 68K platform out there.
I ended up estimating the Lenovo at about 3500 times faster than the Amiga using old benchmarks, comparable modern benchmarks, and connecting benchmarks between the old and new. It took a bit of extrapolation but I was trying to get as best as one could do comparing old and new technologies (esp with the difference of single vs multi-core).
In any case, Mark's comment at the end made me think it would be cool to see who could get the fastest (accurate) benchmark of a QL system using whatever means. I like Peter's approach of using WinUAE with QDOS Classic since it's not double emulation and WinUAE may be the fastest 68K platform out there.
Re: QDT Running ON My Gaming PC Running Windows 10 Home Running QPC 2
Is a UAE "kickstart" ROM available for the later version?Mark Swift wrote: Tue Aug 12, 2025 9:41 pm Later Amiga QDOS Classic versions read the Amiga RTC to get the time.
In UAE, the system time is read from the host machine whenever the Amiga RTC is read; so as a result time is accurate in later builds.