NefMoto

Hi! I'm Mike. I've been around the forum for a couple of years, but I've been mostly quiet. I'm not even sure how to start this thread. Maybe a quick introduction?

I own a 2003 Audi A4 Avant (that's a wagon to the uninitiated (actually, it's a wagon to the initiated, too... it's just a wagon)). I bought the car a while back when I started a YouTube show that was basically centered around showing people how to get into motorsports and racing. The point was to buy literally any car and show people that they could take it to an autocross event (or season), do track days in it, take it to the drag strip... whatever. You don't need to build a full-fledged race car first. Then, if you wanted to, you could build that literal anycar into a full-fledged car when you were ready. I thought I could bring 20 years of racing experience to bear and maybe show people like 18 year old me, who didn't have a clue, how they could get onto the track.

I picked the Audi because it had the 1.8t engine, and I always wanted to own one of those. A friend had a GIAC chipped b5 Passat back when we were kids, and I remember that car being kind of a riot to drive around, so I wanted to relive my youth a bit, too. Hey, lots of reasons. Who cares?

Anyway, I autocrossed the car. I took it to HPDEs. I even rebuilt the engine in my garage. Later on, I gutted it, installed a roll cage, and took it to a race school. It was time to get my race license. The one thing that always mystified me was tuning. I read the community tune thread by thelastleroy several times through, and used the tune to try to learn how to make this stuff work. I even tried it on the track (got my caboose waxed, but it was still fun!).

That was inspirational. But I didn't want to just show up here, take that for free, and go on my way. So, I used the thread to learn how this all comes together; and I used the tune as a base for trying to understand how to work in TunerPro, the NefMoto ECU Flasher, etc. Then I set out to make my own tune so that I could share it with you all here. Call it a way to give back if you will? I hope it's a fair exchange.

It seems like there are a couple of community tune threads for 1+ that might have come and gone without getting off the ground. For this one, I'm actually showing up to document what I already did. It's done. I ran it on a dyno today, and we tweaked it to make it safe and smooth. I'll post the binary here, along with a log. I guess I wanted to document what I did to make it work. I'm always open to comments on it. I don't intend to make any changes until over the winter after our last race of the season.

I don't want this to be a giant wall of text, and it already is one, so I'll probably make a few posts to get this out. I hope that's OK. Details about the car will follow in the next post. It should be obvious, but I guess I have to say it: I'm not responsible if you download this tune, upload it to your car, and it breaks something.

Enjoy!

Details about the car:

2003 Audi A4 Quattro Avant 1.8t. It has the five-speed manual transmission.

Version Information:8E0909518AK-0003

Engine: GT-2871R turbo, front-mounted intercooler, 3" MAF sensor housing (I think it's 3" anyway...), tubular T25 manifold, custom cat-less exhaust. AWM head.

Fuel: Bosch 550cc injectors, bought new from Motoza. Stock fuel pump.

I think that's all that's relevant for the hardware side of things. I'll follow up with my methodology and explain just what, pray tell, the heck I was thinking when I decided to try to force ChatGPT to tune my car.

Know what would be hilarious? Forcing ChatGPT to tune my car. Please understand that this is not what you might call a good idea. It just was and idea, and I wanted to explore it. I've used GPT to write code, and with the advent of GPT 5, I was curious as to what else it could do.

So, I paid my 20 bucks to GPT for the enhanced tool, installed the turbo, and got to work.

Step 1 was to tiptoe into things a bit. The car wouldn't idle off the bat, and I knew why: the new MAF housing was larger than stock, and so when plugged in, the car would lean out and stall. GPT! How do I fix this?

The stock MAF housing was 2.5", and the new one is 3". GPT advised me to edit MLHFM by multiplying all the values by 1.44. Sure, why not? Let's give it a shot. I was, by the way, at the car's debut race with this new hardware. The idea was to get it running real quick in the paddock and see how it ran. But a 1.44 multiple got it to idle, and it might be close enough for now. So we had some success!

After that, I tried uploading various other changes but the uploads kept failing! Oh no! I was concerned about bricking the ECU with the failed attempts, but I discovered something in the aftermath. First, I didn't brick anything. Yeah, we had to push the damn thing onto the trailer because it would not start, but it wasn't bricked and it eventually took another upload. Second--this is important for anybody who experiences a similar issue: you may have to disable antivirus while you're uploading or logging! I was getting all sorts of errors, and I think they centered around these random little disconnects of the USB device. You know the Windows chime for unplugging USB and then plugging it back in? That. The upload would fail. Then I'd get a message about a Tester Present Message.

You know who helped me fix that? ChatGPT. Unfortunately not in time for the race weekend. At this point, I was the one at the keyboard making changes in TunerPro. But I was curious... what would happen if I allowed ChatGPT to modify the binaries, then I installed them, log it, and then allow it to use the log file as feedback for another iteration? Sound like a terrible idea? It was! But it was also interesting! Do not do that thing I did, unless you are not afraid of trashing an engine. More on that next.

OK. Time to establish test procedures.

1.) Approach ChatGPT and tell it what I wanted to do.
2.) Feed it a stock binary (yes, always start with a stock binary)
3.) Feed it an XDF
4.) It creates a binary
5.) I upload and log it
6.) Upload the log to ChatGPT
7.) ChatGPT iterates on the binary

I edited the logs so that only the 3rd gear pull exists, and all other data is deleted. If I didn't do that, then ChatGPT struggled to find the actual pull.

I started off having it remake the MAF housing change. Interestingly, it didn't know where the tables were at first. I got to learn how we define table sizes, starting addresses, etc. OK, fine, so I gave it what it asked for and... lo and behold, it actually changed the right table. I did a byte comparison to make sure that was all it changed, and hey, wouldn't you know it, that's all it changed!

I knew by this point that I had to disable antivirus and wifi (I don't know for sure if wifi needs to be disabled, but with antivirus disabled, I sure didn't want wifi on anyway). I flashed the ECU and the car fired up. Cool. It didn't drive very well, though. Notably, at this time, I had chosen to forgo the larger 550cc Bosch injectors, and I remained with the stock units. I was just trying to get the car to run and drive normally while throwing the fewest number of variables at it as I could. The car was overboosting, though, so I threw the big Bosches in there and I think that's when I asked ChatGPT to adjust KRKTE. Actually, for fun, I asked it what I should adjust next (or maybe I told it I installed the big Bosch injectors), and it told me "KRKTE". I let it adjust that, uploaded it, logged it, and sent the log to GPT for feedback. This was getting fun. But it started making mistakes. That's next.

Before I uploaded any files that ChatGPT gave me, I always verified them to ensure that it wasn't doing anything weird. For a while anyway.

Eventually, I got complacent. I was having fun and the car seemed happy. Until it wasn't. I pulled out onto main street to give it a little rip and it stalled. I couldn't get it restarted either. That was scary, especially since it was dark. But, I got it fired up and went straight back home. I looked at the KFMIRL table, because that's what we were working on, and I found that from 50% to 100% requested load, in the 1,000 RPM column, that it had just decided that the value should be 3. Not 30, or 300... just 3. Well, 3.4, really, but let's not split hairs.

Why did it do that? Not a clue. I've included that binary for you to peruse if you wish. Please do not install or try to run it. It's just here for the sake of a little fun, but also as a lesson--if you go down this path like a masochist, please verify the stuff it gives you.

It was at this point that I decided to change my methodology. ChatGPT would no longer be the keyboard warrior. I would do keyboard inputs, and GPT would inform me about what to do. Also, this is a race car--I am not supposed to be driving it on the street, so my real test plan was to do a lot of this stuff at the drag strip. You can get into a test and tune day at 8 AM for $40. Run the car down the track all day, log it all, and go make iterative changes. No, I wasn't planning on launching the thing--really, just leave the line, get it to 3rd gear, and lay on it at 3,000 RPM. Is that the fast way down the strip? Nope. But it's the best way to get my data.

Sadly, I didn't get the car finished before the test and tune event that preceded my next road racing weekend at Pittsburgh International Raceway. I guess more on that next.

Pitt Race!

My dad and I loaded up my 2000 Fleetwood motorhome in--I don't know--August? Something. Then we threw the Audi onto the trailer and drug the thing HUNDREDS of miles through the mountains of Pennsylvania to go race it. I had something on the car that worked, but worked very poorly. I was overboosting like crazy, but I figured I could have a day to hang out with my friends and tap on the keyboard a bit in the sunshine, and we'd get the thing reigned in. Saturday was fun, but I wasn't competitive. The car would overboost, then the throttle plate would close and it would be in limp mode. But what the hell, I ran an entire race like that. It felt like a stock K03 and a stock tune honestly, with maybe no low-end torque.

Fine.

But I wanted to be faster and run the car without worrying about scattering parts on the track. So I stayed up late with my friends, made some s'mores on the tabletop stove thingy, and reigned in the tune.

One mistake I'm not embarrassed to admit to making was that I made KFMIRL way too high. Way. Too. High. In fact, I didn't realize that these values were one-byte size, so anything greater than 255 was not valid. (My god, what was I even thinking asking for values that high? Well... heh... ChatGPT suggested it. Remember, I was still trying to see if this could be done by GPT.)

I informed GPT that we were way over on this table and that I wanted to reign things in, so we dropped KFMIRL to something closer to stock. I took inspiration from the Stage 1 tune to see how far we should bump things up.

This stuff is really hard. Look, I read all 50 or 60 pages of the stage 1 tune thread. I read the S4 wiki. Keeping everything in your brain is really hard. I would write notes to myself as I read along "Do this, then do this, etc.". Then I'd still do dumb stuff. But the cool thing is that if you just keep hammering away at it, eventually it starts to make some sense.

So, what did I do? I dramatically reduced KFMIRL, I attacked the LDRXN table to lower boost limits, and I uploaded one last tune. And hey, this one worked! For two laps! There was a point where I went to do my 3-4 shift, and I reached down to grab the shifter and it was already in 4th. The thing was scooting! Until the turbo fell off. Of the engine. While I was on the track.  :-[

It didn't fall completely off the car. I used grade 8 fasteners with lock washers to secure the turbo to the manifold. Those of you who know how stupid that is are slapping your foreheads. For everyone else, it was stupid. Thermal expansion rates, clamping loads under high heat, and vibration all work to make securing turbos to engines really, really hard. So I got two laps in on qualifying that morning, and my friends worked feverishly to reattach the turbo with whatever they could find while I went to Dunkin. (I was buying them coffee as well as parts to fix the RV. Did I mention that the RV broke down the minute we got into town? Busted a heater hose pipe and dumped all of its coolant. Fun weekend.) We got the car fixed and ready to roll in time for last call to grid for the race. Two more laps (just to take the green flag for points), and then it fell off again.

What did I learn? I was able to create a tune that worked for the most part. It wasn't very good, and it did overboost, but it was much less dangerous. It was close enough that, when I got the thing reattached with better fasteners, I was pretty sure I was going to get it going right.

Another shameful binary that you should definitely not run is attached so you could see how lame my KFMIRL was. They're date/time stamped. I also included what I created and ran on Sunday. 20250822_1148 is the maxed out dumb one. 20250823_1231 is what I created that night and ran on Sunday until my turbo fell off. I should note, as well, that I was avoiding full throttle much of the time. I knew it was overboosting, and I wanted to try to keep the engine in one piece. Interestingly, because KFMIRL was so out of whack, I had very little control with the throttle. This was not a driveable tune. lol

I'm still here. I've been having trouble getting the site to load in recent days. At least when I am trying to post. I hope to put one or two more messages, but then open it up to see what you guys think. More when I have some time to write!

OK, so. The USTCC East season was getting ready to end with a race weekend at New Jersey Motorsports Park on the Thunderbolt course. I love this track, and I have been looking forward to the race all season, so I wanted to make sure I made as good a showing as I could. The first thing that I set out to do was make sure the turbo wasn't going to just fall off again. So, I reattached the turbo to the exhaust manifold with stainless steel studs that I secured with two-way reversible locking nuts. This type of lock nut uses deformed threads to prevent the nuts from moving, kind of like cross-threading does, but... you know... on purpose.

That worked.

I was also concerned that I might have had too much preload in the waste gate actuator arm, so I adjusted that to loosen it up a bit. The cracking pressure, I believe, is right around 11 psi now if my memory serves me.

It was close to the time that I needed to load the car up onto the trailer and take it over to a dyno in town to try to get the tune smoothed out. I was pretty much finished with the ChatGPT method, so before going over there, I started over from the stock tune and tried to get things dialed in again. I tried to follow the methodology laid out in the S4 wiki (again, I had lots of do this, then do that notes to work from). What I created ended up being a little strange with a big peak early on, then a dip, then another peak... it was a little crazy, but I figured it would be good enough to get it onto the dyno so we could sort it from there.

The tuner had a different approach than I think we might here. He wanted to control the wastegate through PID. Sort of force it to do what we wanted. We only had limited time, and we went from that peaky, overboosty tune that I had installed on the car to a much smoother, safer tune in one session. Interestingly, I think I was making 285 ft-lbs of torque at the wheels on the tune I created, but it was chaos, so we were happier when the car was dropped to the low 200s in torque and about 230 horsepower. The final numbers are in that dyno chart up in post #1. We didn't have a great deal of time on our hands that day, so we didn't get it dialed in where I want it. That will happen this winter as we go back in for a full day and get it where I want it.

OK, so the car was tuned well enough. It looked safe. We had ample fuel, timing was very good, and it looked like it was ready. Time for New Jersey!

Have you guys ever used a race car fire suppression system? I have! It's pretty cool to see it work. More on that later.

The car was ready to go to NJMP. One last thing to do--I wanted to control the heat output of the turbo a little better, so I had ordered a turbo blanket. It showed up in the mail after the dyno session, so I installed it and took the car for a little drive (by this time, I had actually re-registered and insured the car so I could do some street driving. I forgot to mention that in the last post.) The blanket seemed like it was doing its job, so off we went!

We showed up on Friday morning. There was a test day that day, so I wanted to see how the car behaved and get to relearn the course since it's been like 10 years since I've been on it. I did a session and I realized that the car was losing power in the last 300 RPM before the redline. Geeze, we only ran the dyno to like 6,200 RPM for some reason, so we never tested this regime! It was a silly mistake that I know about now, and it was easily reconciled by just shifting a little early. No big deal.

The car was fast. Not fast enough to keep up with the front-runner in my class, but fast enough to put on some serious speed. I was right at 120 mph nearing the braking zone, and as I became more confident in my run onto the straight--as well as in the braking zone--that number would eventually get to 125 I think. The front-runner was running about 10 mph faster than me, but there's another 50 horsepower that I want to access that I think will help. But hey, it was fun!

When I came into the pits, though, we saw smoke. I shut the car off and we popped the hood. Apparently, when I installed the turbo blanket, the spring that wraps around the bottom took the oil drain line and pressed it against the exhaust pipe. It burned through and leaked oil. Dang it. So we went to the auto parts store and asked for some oil-safe line. They gave me heater hose which, in the first qualifying session the next day, melted. Probably from the inside-out. And then caught on fire. I pulled the fire extinguisher and put it out (there goes $145). It was pretty neat to see it work, honestly, and cleanup wasn't so bad. I have a Lifeline Zero 2020 in there. We were able to scramble like crazy and get the car fixed, but I didn't end up running it anymore. It was a pretty disappointing way to end the season.

That's kind of where I am on tuning the thing for now. I ordered about $200 worth of fasteners, including Inconel studs for the turbo-to-exhaust flange. This is because those bolts fell out (again) at NJMP during the test day session despite being torqued down right before loading the car a few days prior. I also decided to do an AN-style oil drain with absolutely-the-right-kind-of-high-temperature-oil-rated line. I meant to start on that today, but cleaning the garage took over my attention for the day. The next thing to do, after the hardware is resecured, is to bring the car back to the dyno and stick it on there for a day. First, I want to get the tune dialed in. 285 horsepower for as long as we can make it. I'm going to make sure we run the darn thing all the way to the limiter. Then, for good measure, I want to lock the dyno to hold the car at, say, 5,000 RPM and just hold it WOT until the damn thing starts glowing. I realize that airflow will be a problem with just some fans on it, so we won't break anything, but I want to make sure this thing isn't going to just fall over on me after a single lap.

I'll report back when I have more. Sorry it took so long to kind of close this loop. I expect to try to get the car onto the dyno by January 2026. I don't know where the USTCC season will open up, but the first motorsport events where I live are usually early in the Spring--sometime around March or April.

Oh, something else occurred to me. I would like to speed up the changes that we make on the dyno. The binary takes 5:15 to upload, and I'd like to cut that down as much as possible. I always do the Full Write option when I upload, but I see that button right there that suggests something like "Write Differences" (I can't remember the button's text, and Nefmoto isn't installed on this laptop, but it's close to that). I never tried it because I fear change. But, unless I'm mistaken, as long as I compute my checksums and have a valid binary to write, using that method should speed things up significantly, right? It may end up only writing one sector instead of 19? I'm going to give that a try.  :)

Fall is the season for kids of things. Pumpkin spice (a Starbucks barrista once called me "basic" for ordering a pumpkin spice latte, so that's out for me forever now), leaves changing colors (allegedly... I don't know, I'm colorblind), and turbo upgrades!

What upgrades? Great question, I'm glad you asked. Why don't you pull up a chair, take a seat, and relax for a bit?

Oh and, yes, I can get you a glass of milk. Sure thing! Here you go.

Now where were we? Ahh, the upgrades. Well, I was getting tired of my turbo falling off, the manifold nuts loosening up constantly, and the car catching fire. None of this is really helpful in racing, where reliability is just as important as speed. So, I loaded the nuclear shotgun, pulled the trigger, and ended up with Inconel turbo exhaust flange studs, copper clad locking nuts, $4-a-piece Nord Lock washers, a new exhaust manifold gasket, and new exhaust manifold nuts. I also replaced the turbo drain line with -10 AN fittings and a custom-length steel braided hose that's specifically designed to not melt and catch fire when hot oil goes through it. I've never made an AN hose before so that was s cool experience.

Also, relevant to this thread, I had an opportunity to check on the waste gate actuator again, now that it was out of the car. I didn't know how to adjust it when I put everything together over the Summer. Now, I know there's a pretty simple way to preload the actuator. Hey, would you look at that... the actuator rod is too long thanks to the change in geometry that was required when I had to move the waste gate actuator. So, I hacked that thing down a bit so I can screw the adjuster down more and now the waste gate seems to close more affirmatively. 2-3 mm of preload. It still opens at 11 or so psi. I will get a mityvac so I can more precisely measure the cracking pressure. I'm thinking this will help turbo response immensely.

I'll have the car on the dyno sometime this winter. Once I do that, we'll have some changes, and I'll post about them here. I might go out and try the 8-run WGDC tuning protocol thing if I have time to see if I can get some of that dialed in beforehand. I'm not sure it'll be necessary though since we can do all of that on the dyno.