MiSTer FPGA Forum

if im reading the comments/discord messages right, the testing results from VGE was that 1-2 frames are desynced over a couple hours, when comparing a Classic MiSTer to Taki's Clone. which was later confirmed even console/arcade board to board hardware comparisons would have the same variance, and is within tolerance. and that desync was not universal, VGE only saw it for a couple cores during testing.

the Taki Clones are a boon to the FPGA community. i already have a couple Classics, but i didnt have the v3.0 SDRAM, so i ordered a couple of those from Taki. both of those SDRAMs are cheaper than 1 stick i bought years ago. and a complete MiSTer setup (without a case) for $160 is a no brainer when compared to the ~$350 people have been paying for the past couple of years.

i am very much looking forward to the Taki FPGA Handhelds to see how those turn out. wild times for sure

neogeo81 wrote: ↑Tue Sep 10, 2024 6:38 am

gregory003 wrote: ↑Sun Sep 08, 2024 9:24 pm

…

i watched all videos and didn't see any problems. could you elaborate?

I'm talking about differences in framerate/frame pacing/cores speed:
https://www.dropbox.com/scl/fi/xgxxqpem ... xtefw&dl=0

https://www.dropbox.com/scl/fi/u000qjj0 ... 7tiqb&dl=0

https://www.dropbox.com/scl/fi/nlc2a6od ... zavpn&dl=0

https://www.dropbox.com/scl/fi/cxfoz0v4 ... hyslj&dl=0

https://www.dropbox.com/scl/fi/3purkzfv ... 8et0d&dl=0

https://www.dropbox.com/scl/fi/u3j119mh ... cpouw&dl=0

PikWik wrote: ↑Tue Sep 10, 2024 2:53 pm

if im reading the comments/discord messages right, the testing results from VGE was that 1-2 frames are desynced over a couple hours, when comparing a Classic MiSTer to Taki's Clone. which was later confirmed even console/arcade board to board hardware comparisons would have the same variance, and is within tolerance. and that desync was not universal, VGE only saw it for a couple cores during testing.

Yes, but Mister, DE10-Nano and Taki's clone are not arcade boards or consoles, and "confirmations" of desynced arcade boards don't work for them. They are fpga developers boards which only emulate arcade boards or consoles. And are build around high precision fpga chips, used in medical equipment or ballistic missiles, so they shouldn't "desynced" just that, themselves. And I'm doubt if such selves-desync is in any mean "tolerable".

Given that the FPGA is exactly the same (literally), I'd chalk any desyncs up to potential differences in compile outputs, unless the very same binary cores were being used. Even then, I'm not quite sure I see how that would be possible...if RAM can't keep up, it won't slow down, it'll just produce invalid results.

rhester72 wrote: ↑Wed Sep 11, 2024 12:17 am

Given that the FPGA is exactly the same (literally), I'd chalk any desyncs up to potential differences in compile outputs, unless the very same binary cores were being used. Even then, I'm not quite sure I see how that would be possible...if RAM can't keep up, it won't slow down, it'll just produce invalid results.

I read/heard somewhere that a likely culprit could be 3:2 pull down? The people discussing it are way more knowledgeable than me. I can't find it in my youtube browser history. Maybe in a Lu's Retro Source video or RetroRGB?

dickhardpill wrote: ↑Wed Sep 11, 2024 12:32 am

rhester72 wrote: ↑Wed Sep 11, 2024 12:17 am

Given that the FPGA is exactly the same (literally), I'd chalk any desyncs up to potential differences in compile outputs, unless the very same binary cores were being used. Even then, I'm not quite sure I see how that would be possible...if RAM can't keep up, it won't slow down, it'll just produce invalid results.

I read/heard somewhere that a likely culprit could be 3:2 pull down? The people discussing it are way more knowledgeable than me. I can't find it in my youtube browser history. Maybe in a Lu's Retro Source video or RetroRGB?

Were that the case, you wouldn't see it lagging behind, it would just drop a frame and remain in sync.

rhester72 wrote: ↑Wed Sep 11, 2024 2:27 am

Were that the case, you wouldn't see it lagging behind, it would just drop a frame and remain in sync.

I'm probably not interpreting what they said correctly, like I said, these people know a lot more than me. All i know is they came to the conclusion that both alternatives are a success.

rhester72 wrote: ↑Wed Sep 11, 2024 2:27 am

Were that the case, you wouldn't see it lagging behind, it would just drop a frame and remain in sync.

I did watch the videos and i can´t agree that it was lagging behind.
The reason is that one or two systems where 1-2 frames behind sometimes and sometimes NOT and it came and went during the test.
AND it was not only the taki that did it the dual memory de-10 did it too, and sometime the single memory de-10 did it.

IF it was lagging behind that would have mean that the number of frames it was behind on would have increasing and increasing but they did not!

Quartz tolerances, from what i've read.

Retro-Nerd wrote: ↑Wed Sep 11, 2024 5:31 pm

Quartz tolerances, from what i've read.

Probably - but that being the case, two identical real-world-hardware machines will very slightly drift from one another over a long enough period of time for the same reason, making it a pretty useless test.

Of course the test is pointless. But that's what the Retro Gaming youtuber like to do, for whatever reason. Just comparing things with each other. And then people get it wrong.

Retro-Nerd wrote: ↑Wed Sep 11, 2024 6:01 pm

Of course the test is pointless. But that's what the Retro Gaming youtuber like to do, for whatever reason. Just comparing things with each other. And then people get it wrong.

But such test were made after up to few minutes of running cores. Isn't error at 1-3 frames during period of few minutes slightly to much? No one youtuber made test with core running overnight. They just started core and waited on title screen for intro to start. In one case test was made one hour after starting core.

Is 60 minutes "long enough period" to notice quartz drifting and consider it as tolerable?

Consumer-grade quartz has a tolerance of around 0.0006%.

If your claim can be boiled down (to simplify the math) to an error of 1 frame in 5 minutes, that would represent 0.0056% deviation, or around 10x that of expected. However, if it's 1 frame in an hour, it's actually slightly above average. It will never be zero.

I haven't seen the videos and frankly have no interest, but my point is that however you slice it, the test is basically completely pointless bullshit...but the scale absolutely matters. I've read in this very thread that it's as bad as 3 frames in "a few minutes" to a handful of frames over many hours. That's a big, BIG difference in magnitude between the two.

It heavily depent whatever means "few" for different people. 5? 9? 11?

Still, if this is only quartz tolerance problem it shouldn't be more than 1 frame per 50 hours (with claimed tolerance at 0.0006% for consumer devices).

yes, i am also interested in the differences of the clone vs the classic MiSTer.
i only have an original MiSTer and no clone, so i cant compare the 2 over long periods of time, while comparing the capture footage on a frame by frame analysis. and someone mentioned there may be a slight frame difference between 2 classic MiSTers (with a single RAM setup, and a dual ram setup)?

im very sure someone out there will test these in extreme ways to stress the SDRAM and come to a conclusion.
and it will either be an imperceptible difference, a dealbreaker of frames different with slowdown in random places in a game, or it may very well be identical

There is also a video that compared 3 systems, two was DE10's. Also the De10's was not sync all over the time. So i believe that is more like a factory tolerance with some parts.

oh ok, well there you go

theres even a difference in frames when comparing 2 original DE10s

quoted from PCN's video https://www.youtube.com/watch?v=9E-8Uz9qI3M
with regards to the capture footage used where they see frame differences in playback
...These games were beam chasers based on 15Khz arcade monitors. 60hz is going to duplicate a frame occasionally because these were not locked to 60Hz. Most arcade games are somewhere near 60, but not perfect. It's like converting 23.957 to 30hz (29.97) has to do 3:2 pulldown. I'm fairly sure all you're seeing is an artifact of that on modern 60Hz locked captures (or 50/PAL if you're using that). If it were out of sync, the distance would have compounded over time.... Over an hour having a frame difference, that can be dismissed over signal conversion [and] is essentially parity."

"...60 fps x 60 seconds x 60 minutes = 216000 frames in an hour. The difference is 3 frames, 3/216000 = .00138% difference. Or to put it in crystal oscillator terms, 13.8 parts per million error.

There are at least two different clock circuits on the DE10-nano, Rev C uses a Si5350C, previous revs use a CDCE937. The CDCE937 circuit uses a 24Mhz crystal oscillator; mine is a rev C, so I cannot tell which exact crystal is being used on the older boards. Right now on Digi-key, 24Mhz crystal oscillators have a range of 7ppm - 120ppm frequency tolerance, so what you are seeing here is most likely within the margin of error of the crystal. Original arcade boards also experienced these board-to-board variations."

i think thats enough evidence to say the clones are at least within the norm of variance ...... for now

If anyone can sit and play a game for 1 hour and then spot that 1 frame has dropped since starting, he's one hell of a guy.

hitm4n wrote: ↑Thu Sep 12, 2024 6:00 pm

If anyone can sit and play a game for 1 hour and then spot that 1 frame has dropped since starting, he's one hell of a guy.

Same here mate, tired of folks banging on about milliseconds and latency which seems a common trait across all FPGA projects.
We're all in our 50s.60s now.
It's not Christmas 1983 when you got your spectrum and you could react quick enough playing Hungry Horace or Manic Miner.
Most folks are just happy to be able to play anything, never mind latency etc.

I'm starting to think that some people just don't want others to have paid less to get a MiSTer by trying to justify why their more expensive setup was worth it.

Waifu4Life wrote: ↑Fri Sep 13, 2024 8:55 pm

I'm starting to think that some people just don't want others to have paid less to get a MiSTer by trying to justify why their more expensive setup was worth it.

Unfortunately I find myself agreeing

It's sad that with over 81 posts in this topic not one of them is about first hand experience with the new MiSTer Pi board. That would be actually interesting and useful information.

For myself, I've had my own MiSTer stack for over 4 years and it was money well spent. It's also great that other manufacturers besides Terasic are now producing MiSTer compatible boards.

The irony that we’ve moved past scrutinizing emulation cycle accuracy vs. real hardware, but now demand new emulators to exactly replicate an existing emulator… Just goes to show how great MiSTer is. What a time to be alive

The price thing is a possibility, I paid about 500 quid for now a few months later I can buy for 150 quid.
I'm just hoping, if we just one competent dev onboard, out of the thousands of new taki owners, it'll be worth it in the long run.

gregory003 wrote: ↑Wed Sep 11, 2024 7:07 pm

Isn't error at 1-3 frames during period of few minutes slightly to much?

No, it's not too much. You can't really determine that there is any inaccuracy from this kind of test.

ToothbrushThreepwood wrote: ↑Sat Sep 14, 2024 8:10 am

The irony that we’ve moved past scrutinizing emulation cycle accuracy vs. real hardware, but now demand new emulators to exactly replicate an existing emulator… Just goes to show how great MiSTer is. What a time to be alive

The irony is that, we've had cheap emulation boxes like Raspberry Pi, with pretty good accuracy and not-so-bad input lag and not-so-cheap Mister with pretty good accuracy and awesome input lag. Now we had new fpga toy, with messed accuracy but still, for whatever reasons, it is a few times more expensive than Raspberry. And for what? Because "fpga" or "hardware emulation" is a new buzzword in retro scene and you think that fpga emulation is some kind of magic and just want to new cheaper, but not cheap, toy just because it is fpga-based?

gregory003 wrote: ↑Sat Sep 14, 2024 5:16 pm

The irony is that, we've had cheap emulation boxes like Raspberry Pi, with pretty good accuracy and not-so-bad input lag and not-so-cheap Mister with pretty good accuracy and awesome input lag. Now we had new fpga toy, with messed accuracy but still, for whatever reasons, it is a few times more expensive than Raspberry. And for what? Because "fpga" or "hardware emulation" is a new buzzword in retro scene and you think that fpga emulation is some kind of magic and just want to new cheaper, but not cheap, toy just because it is fpga-based?

Rasberry Pi has ARM Soc and the emulators are optimized to get full speed on that, not the best possible accurary (for that you still need a a decent PC) or low input lag. Please stop this discussions.

niallquinn wrote: ↑Fri Sep 13, 2024 7:30 pm

hitm4n wrote: ↑Thu Sep 12, 2024 6:00 pm

If anyone can sit and play a game for 1 hour and then spot that 1 frame has dropped since starting, he's one hell of a guy.

Same here mate, tired of folks banging on about milliseconds and latency which seems a common trait across all FPGA projects.
We're all in our 50s.60s now.
It's not Christmas 1983 when you got your spectrum and you could react quick enough playing Hungry Horace or Manic Miner.
Most folks are just happy to be able to play anything, never mind latency etc.

YOU are in YOUR 60s. I'm not you. And if YOU don't give a crap about accuracy, latency, etc. and YOU will be happy from whatever crap that just playing old games, why YOU just don't buy Raspberry Pi or some Chinese box with less than 40$, instead YOU opted to 160$ emulation box, no better than any others? For me, higher prices should mean higher quality and the ability to fulfill higher expectations. I can't and don't want to tell you what to do with your money - I'm just wondering why you want to make such spending?

As a person not “in their 60s”, I would suggest calming down the rhetoric a bit, and realize that these are the same chips as the DE-10 nano uses, and they have been proven to deliver within a statistical margin of error in every possible way.

Enjoy that the MiSTer is now accessible to the people that would normally advise you to “just buy a Raspberry Pi” instead of trying to start an argument that will lead nowhere. These new boards are absolutely capable of delivering what the DE-10 nano does.

These are not “emulation boxes”, they are the same Altera FPGAs as the original MiSTer setup uses, and they perform equally well.

For anyone interested-
You can click the wrench/spanner and unsubscribe from this topic. Have a great day gaming on whatever platform you choose and welcome to the MiSTer FPGA forum...