Breakthrough for the ao486 core announced - Cache

[tontonkaloun]

I could be wrong, but I think that is due to using the "Lite" version of Quartus.



I have compiled the DEV branch.

It would probably be be better to wait for official release from the people who know what they are doing, but if you can't...

ao486_Cache.zip

Make sure you update Main_MiSTer or you won't get video.

Thanks

I had compiled it but I got the file "limited time.sof"

I had tried to convert it but with an error message with the same command as you.

I admit, I didn't look any further

I think I am limited with the lite version of quartus, I wanted to take a license today!

If you've got $2,995 to burn, go right ahead
Then again, maybe you're a developer and can actually use the license for more than just compiling up-to-the-minute DEV releases of these MiSTer cores discussed here.
I think I'll wait till the ppl that know what they're actually doing compiles it properly for release before trying to compile myself again. Although I feel I 'know what I'm doing', having gotten this far, it's that 'knowing what I'm ACTUALLY doing' part that gets me

Strange thing is... certain cores [e.g. TG16CD] allow me to compile an RBF using Quartus Lite, not so with this 'ao486' core... why/how, I wonder, does that occur?

No, I did not expect this price!

[macro]

unrelated, but they really need to add a 'hobbyist' option at a reasonable price.

I have other software I have bought because they have had such a 'non profit' hobby user level

(although if someone put an opencores UART in there, then it should be OK with lite anyway)

[NightShadowPT]

More improvements from today (yellow bars). I fear this will not improve more so easy

32bitaccess.png

I love the way you say that after boosting the performance of this core for more than 100%

You've just moved this core to whole new level.

Looking forward to the official release.

Thank you very much for your work.

Cheers,

[lroby74]

With previous (no cache) version, LM60 (Landmark System Speed Test 6.0) gives same speed of a 386 sx 25 mhz...
will try with last new test cached version

[NightShadowPT]

With previous (no cache) version, LM60 (Landmark System Speed Test 6.0) gives same speed of a 386 sx 25 mhz...
will try with last new test cached version

If the graph above is something to go by, you should expect a performance between a 486DX/33 and a 486DX/50.

Looking forward to see the result of your test.

[dshadoff]

With no math coprocessor, it should be compared with a 386DX or 486SX.

[tontonkaloun]

With no math coprocessor, it should be compared with a 386DX or 486SX.

yes

[SuperBabyHix]

There weren't a ton of games from that era that took advantage of an FPU until Quake came out. I think Falcon 3.0 could take advantage and probably a few other sims. On the other hand, if you just love Lotus 1-2-3....

[dshadoff]

The 486DX had the coprocessor integrated, and from that point forward, games started to be built using them. That was 25 years ago, so not everybody knows (or remembers) this today. The 486(DX) came out in 1989, but wasn't common until around 1991-92 because of pricing; for this reason, they came out with the 486SX.
(The Pentium came out in 1993.)

The ao486 core can't be compared with a 486DX because it doesn't have a coprocessor. However, it can probably be compared with a reasonably fast 486SX.

If you try to play a game written after 1993 which states that it requires a 486DX or Pentium of some sort, there is a good chance that it relies on either the coprocessor or Pentium MMX instructions, and may not run on this core.

[bbond007]

This is a test of the latest where FPGAzumSpass has disabled the (faulty) ao486 level 1 cache for an additional boost!

--> https://youtu.be/y9CycqjfOxY

[JUNKER Seed]

That is quite the improvement.

[bbond007]

That is quite the improvement.

[thorr]

Excellent!

[NightShadowPT]

This is a test of the latest where FPGAzumSpass has disabled the (faulty) ao486 level 1 cache for an additional boost!

--> https://youtu.be/y9CycqjfOxY

What does this mean? That it will be possible to implement a more efficient L1 Cache to boost performance even further? Or just that in the FPGA implementation you don't need L1 and L2 caches?

[Lodovik]

Incredible gain in performance! Like upgrading to a new PC.

Thanks FPGAzumSpass!

[NightShadowPT]

With no math coprocessor, it should be compared with a 386DX or 486SX.

With a 2.5x to 3x performance improvement from the current core? (roughly equivalent to a 386SX 25Mhz)?

Seems to be too drastic an improvement not to put the new core at a 486DX performance level.

[SuperBabyHix]

The DX vs SX differentiation really is not significant for games. Off the top of my head I can think of maybe three games (prior to Quake coming out in 1996) that made use for a math co-processor (Falcon 3.0, MS Flight Sim 5, and I believe Sim City.) The Doom engine, for example, used integer math exclusively.

You can check against the very massive Phils VGA Benchmark Database to see this core is performing around the 486 SX/DX 33 level.
https://docs.google.com/spreadsheets/d/ ... edit#gid=0

[lroby74]

However, if VGA will not gain VBE Bios compatibility and support, a lot of games will not works, like for example NFS, just tried yesterday..
It doen't works on 486 core while it's perfect on PCem and on VMWare Workstation too..
we really need an improvement on VGA support otherwise a very fast 486 cpu will be less useful

[Cebion]

I think FPGAzumSpass has licked some blood
Awesome!

[dshadoff]

With no math coprocessor, it should be compared with a 386DX or 486SX.

With a 2.5x to 3x performance improvement from the current core? (roughly equivalent to a 386SX 25Mhz)?

Seems to be too drastic an improvement not to put the new core at a 486DX performance level.

Strictly speaking, the core is similar or slower than a 386 running at the same clock speed (note that the clock speed of this core is 90MHz). However, I believe that it may implement the new 486 instructions (minus the coprocessor). You may not remember 386DX-40 or 486SX2/50, but they existed and would be suitable comparisons for performance (and would likely outmatch the core in its current state). But honestly, it doesn't much matter since there were so many other factors speeding up or slowing down machines of that era.

I see the cache-related performance increases as releasing the original potential of the core, but I see other people in this thread wishing for more and more CPU-thirsty games, which eventually will be beyond the reach of this core.

I think it might be interesting to send the core into a new direction like to pursue the NEC PC-98 series of PC-like computers, which might actually be easier than reworking the CPU's pipelining and execution units.

[NightShadowPT]

The DX vs SX differentiation really is not significant for games. Off the top of my head I can think of maybe three games (prior to Quake coming out in 1996) that made use for a math co-processor (Falcon 3.0, MS Flight Sim 5, and I believe Sim City.) The Doom engine, for example, used integer math exclusively.

You can check against the very massive Phils VGA Benchmark Database to see this core is performing around the 486 SX/DX 33 level.
https://docs.google.com/spreadsheets/d/ ... edit#gid=0

This table is great and I think it really helps estimating the results of these improvements.

This new Core version will fall squarely on 486DX33 territory (comparing PCPBench FPS, Doom FPS and Realticks scores), an AMAZING improvement!

Strictly speaking, the core is similar or slower than a 386 running at the same clock speed (note that the clock speed of this core is 90MHz). However, I believe that it may implement the new 486 instructions (minus the coprocessor). You may not remember 386DX-40 or 486SX2/50, but they existed and would be suitable comparisons for performance (and would likely outmatch the core in its current state). But honestly, it doesn't much matter since there were so many other factors speeding up or slowing down machines of that era.

I do remember them, I was quite active on the PC during the 386sx to 486DX4 era, and I am quite familiar with their performance. Great times

I think it is clear that the ao486 core is far from optimized, one of the main reasons we cannot compare Mhz to Mhz. That's why we need to look at benchmark scores, to give us a ballpark performance that you can compare to real hardware.

I see the cache-related performance increases as releasing the original potential of the core, but I see other people in this thread wishing for more and more CPU-thirsty games, which eventually will be beyond the reach of this core.

I think it might be interesting to send the core into a new direction like to pursue the NEC PC-98 series of PC-like computers, which might actually be easier than reworking the CPU's pipelining and execution units.

It is obvious that the core would benefit from having SVGA at some point, but for now I'd rather see it be more accurate with better performance and stability, including less bugs with EMM (there is still something wonky with it's memory management).

Would love to see a NEC PC-98 and/or Tandy cores (branched from AO486), but I see them more as separate cores because they are targeting a more specific kind of hardware.

[bbond007]

Strictly speaking, the core is similar or slower than a 386 running at the same clock speed (note that the clock speed of this core is 90MHz). However, I believe that it may implement the new 486 instructions (minus the coprocessor). You may not remember 386DX-40 or 486SX2/50, but they existed and would be suitable comparisons for performance (and would likely outmatch the core in its current state). But honestly, it doesn't much matter since there were so many other factors speeding up or slowing down machines of that era.

I feel the build I am testing is somewhere between 386DX-40mhz and 486SX-25mhz performance wise which is fantastic!

ao486_SI.png (364.12 KiB) Viewed 11513 times

[Beeble]

Those benchmarks are an indicator, but performance varies n comparrison.

Wing Commander 2 runs on the Mister like a 486 DX50 at least I would say, which doesnt reflect in the synthetic benchmarks.

Impression may vary depending on scenario I assume.

[Neo666]

Just tried ao486_Cache3.zip core that bbond007 posted earlier - great speed improvement over official core - Doom runs a lot better!

However, I'm getting no music from the Sound Blaster for some reason, same VHD has SB music with official core. Anyone else had this issue?

Thanks

[NightShadowPT]

Those benchmarks are an indicator, but performance varies n comparrison.

Wing Commander 2 runs on the Mister like a 486 DX50 at least I would say, which doesnt reflect in the synthetic benchmarks.

Impression may vary depending on scenario I assume.

That's a fair point.

I guess it will depend if the game is dependent on the less well implemented parts for the core or not.

[lroby74]

Just tried ao486_Cache3.zip core that bbond007 posted earlier - great speed improvement over official core - Doom runs a lot better!

However, I'm getting no music from the Sound Blaster for some reason, same VHD has SB music with official core. Anyone else had this issue?

Thanks

Do you have SET BLASTER string in Autoexec.bat?
Something like this : SET BLASTER = A220 I5 D1 H5 P330

[Neo666]

Yes, it's working just fine with the official build, but the same VHD with the same settings has no Souind Blaster music on the test core - sound effects are playing though

[tontonkaloun]

Yes, it's working just fine with the official build, but the same VHD with the same settings has no Souind Blaster music on the test core - sound effects are playing though

Yes, you're right ! no music for example with doom2

[laserdisc]

So for laughs I thought I would take my GOG.com install of Xwing Collector's Edition and place it into it's own VHD (to act as a cdrom). And took cache3 for a test drive.
Intro plays with MUCH smoother than release. I get full audio (setup below) but there's still some slight music playback (fluidsynth) slowdown in places as before otherwise most of the time the music plays fine. The joystick doesn't work as with the release version it does. Don't know what's happening there. Noticeable performance increase to say the least, colour me impressed.

Setup:
MidiLink: Local - Type: FluidSynth (using basic sf-55 soundfont, nothing fancy here) [Default MidiLink.ini (no changes)]
SoftMPU 1.91
Empty-CD : Fake CDROM driver to load MSCDEX to get a drive letter out of it.
MSDOS: ASSIGN.COM : used this to assign D: (2nd VHD) to E: (MSCDEX assigned driver letter)
AUTOEXEC.BAT : "SET BLASTER=A220 I5 D1 P330"
Xwing set to use General Midi at port 330

[PikWik]

so cool to see ao486 getting more attention and progress.
looks like we may be closer to having a proper core for all the classic DOS games of late 80s and 90s !