Gran Turismo 7 “Power Pack” Paid DLC Now Available: New Races Including Endurance Events, Qualifying, and “Sophy 3.0”

[Fezzik]

I think there was a mistake in the marketing. The PP doesn't use Sophy 3.0, it uses Sarah 3.0!

 

[jmopborba]

Has anyone tested the new patch?
Any changes on the Big Willow race?

 

[Fezzik]

IMHO, for those comfortable with the game, start at 2 pepper and work backwards.

This is a good idea and I think I might restart from the beginning, now they've dropped this update. I was beginning to have things a bit too easy just using the 1 chilli vehicles.

 

[GTRedman]

Training an AI is not running an AI. I find absolutely hilarious people debating an early 2010s cheap tablet CPU being capable of running something like this locally.

This is what is called an argument from ignorance. Put some effort your post to verify what you say is true or not. I've already posted a link of it running on PS4, and when it was trained it was running on PS4 in real-time.

 

[Nebuc72]

Also, did anyone manage to beat the 2nd civic race using the 87' Civic (2 peppers) ? I think it's virtually impossible to get a win there unless you cut that chicane like crazy.

2nd one yes, it is literally just how much time you can pull back on the chicane before they catch on the straights.

Haven't even come close at Suzuka though..

 

[05XR8]

But why are you even trying easy? You're not a newbie, and easy is meant for the newbies. I did some easy events just to blast through them, but easy is not meant for people who know what they are doing.

IMHO, for those comfortable with the game, start at 2 pepper and work backwards.


Let's hope they fixed the bug where the AI surged into the corner (Reggie AI).

I did a race no qualifying with the NSX at Fuji and was down by 18 seconds on lap 4. Again, I’m not trying to blow through the events. Not like this is a limited time offer.
I’ve got the gift cars I like and using those in Custom Races. I’ll get to other events and might rerun some at another time. I’m enjoying doing my thing.

 

[pabloprp69]

2nd one yes, it is literally just how much time you can pull back on the chicane before they catch on the straights.

Haven't even come close at Suzuka though..

Almost got it last night, but I screwed up in the last lap, at the chicane, my car went flying and I met the wall sideways 😆

 

[CTznOfTime]

Correct me if you think I am wrong but

I think this is my understanding of the Difficulty level in the Option:

This is the difficulty of the regular pre-programmed computer cars, regular AI found on all the tracks within the main solo plAyer mode of the world circuits section.

But it doesn't apply to everything, just most but not all: as follows

Yeah, the difficulty selection in the Options menu doesn't seem to make a difference here. Just tried putting it on Hard, did the GTI race on Alsace with the non-chili car, and Sophy didn't seem faster than it was before I switched it to Hard.

The difficulty level in the Option excludes and does NOT apply to:

Races with chili 🌶️ 🌶️ 🌶️
Races and tracks with ai Sophy 2.0
Races with ai Sophy 3.0 (PP DLC)

Did I forget any else?

 

[GTRedman]

This doesn't tell the whole picture.

AI is a processor intensive feature, always has been, always will be. As the processors get more powerful, coders make the AI suck up more power. Graphics also suck up a lot of power.

On the PS5 Pro, If you run a custom race with the cars packed close (like standing starts or minimum distance) and you adjust the graphics for highest quality (instead of higher framerate) you can see the slow down. I imagine that Sophy sucks up just enough power that there is an unacceptable slowdown in many cases on the PS4.

I was just making a point that it could technically run on PS4, how many instances it could support or what bottlenecks it would face is unknown.

 

[art3mis]

The Power Pack is fun. For better or worse, it has the DNA of Gran Turismo - some brilliance mixed with some peculiarities.

The nuisances:

• 
  
• The spread of the field
  (though i guess it gives a wide variety of skill levels a chance to race against somebody in the field)
  - The randomness of difficulty from event to event - I found the FF Tuner challenge a little tricky even with the fast car.
  Then the 300hp tuner challenge is a blowout with the fast car?
  - Sophy doesn't adapt to your speed or the in game difficulty - they seem to run the same lap times regardless of anything. Maybe this is the scriptedness again but why can't PD figure out an adaptive AI strength that scales to your lap times or at least a difficulty slider?

Overall, I hope this becomes a building block for GT - either leading to another expansions and/or influencing the design of GT8's career mode.

I think you're really on to something here. When I was cruising around Le Mans last night I realized that the AI cars were basically breaking up in to groups or chunks. Each group had a minimum of 2 cars in it but no more than 5 and the lap times were so similar and the gaps were so consistent that it occurred to me it was on purpose.

Each "level of ability" was represented here. From the very beginner never picked up GT before to the Alien who's capable of being top 300 in the world can jump into the race and find a group to compete against.

Maybe I'm looking at the purpose of the DLC the wrong way.

 

[Theufcveteran]

Maybe true if training was simulated, but Sophy is trained in real-time running the game just like we do when we play it.

"Training to drive in Gran Turismo takes place in real time, as the game speed cannot be modified. Under this condition, we would need data from multiple instantiations of the game, running in parallel, to be able to provide GT Sophy the experiences relevant to high-speed driving in a reasonable amount of time.

To facilitate training GT Sophy on a large scale, we partnered with Sony Interactive Entertainment and gained access to more than a thousand PlayStation consoles (PS4s) in the cloud. To train a time trial agent, we used 10 PS4s, each running the game with 20 cars in a race evenly spaced around the track."

From what I remember, Sophy is pre-trained and runs from a “static model” once on your console. A couple of things point towards this being the case - Sophy can only drive certain cars, and their lines don’t change based on your driving as far as I know.

 

[sonnyboychris]

I'd imagine PD could have made Sophy work on the PS4 if it was part of GT7's initial development. But there would have been tradeoffs in other areas, such as smaller grids, decreased LOD, inferior audio, poorer physics calculations, etc that are typically CPU intensive.

Being a first party developer, PD has the in-house knowledge to extract every ounce of performance out of the hardware. I doubt there was much untapped potential in the PS4/Pro versions that would allow them to shoehorn Sophy in without it feeling/looking like a diff game.

 

[JC_Dude]

The GT500 race at Fuji is a lot easier now. I was unable to win it yesterday despite like 6 or 7 tries, but did it straight away today with the NSX.

 

[CTznOfTime]

Friend of mine (French) has completed all the events in the pack, barring the endurances.
We noticed that he is still missing a reward.

View attachment 1498351

I think that might be game-on for a reward for completing them all. Would make up for the payouts!

Since the beginning, every time I look at this map, it reminded me of something very familiar, core memory
But I keep on forgetting to ask if anyone also see this:

The GT7 PP Map reminds me of this, don't you think?

I'd imagine PD could have made Sophy work on the PS4 if it was part of GT7's initial development. But there would have been tradeoffs in other areas, such as smaller grids, decreased LOD, inferior audio, poorer physics calculations, etc that are typically CPU intensive.

Being a first party developer, PD has the in-house knowledge to extract every ounce of performance out of the hardware. I doubt there was much untapped potential in the PS4/Pro versions that would allow them to shoehorn Sophy in without it feeling/looking like a diff game.

I agree with you, but Clearly what you describing is a heck lot more work for PD....

Worth it ?

It would be similar to me driving hours on end, just so that I could shave 1 sec of my OTT ....

Doable, but ultimately due to limited available time and resources, so this may not be the best effective use of time...

 

[pabloprp69]

The GT500 race at Fuji is a lot easier now. I was unable to win it yesterday despite like 6 or 7 tries, but did it straight away today with the NSX.

Does the grid still have a big difference in times or the race is more closely together?

 

[art3mis]

The AI At Group C Le Mans Endurance has increased pace by between 5 and 30 seconds per lap.
Last night the fastest they could run was 3:12:xxx. Post update the fastest lap Gallo ran was 3:07:xxx or a full second faster than my fastest lap.

I first noticed this on the restart after the update. I selected start from current position. First lap from the restart the AI threw down a 3:07 and then kept that pace for the next 5 laps.

I decided to restart from the beginning.
Turned a 3:08 on Mediums with the Nissan to open qualifying.

The first car to set a time went 3:07 with the Mazda. Last night that exact driver was lapping in the 3:20-30 range. The rest of the times started coming in and the spread was 10ths instead of full seconds and drivers that were previously clocking nearly 3:40 laps were down in the :08s

I am now topping out at 232 at the end of the Mulsanne up from fluttering between 226/227. Braking distance has drastically increased though. I used to be able to hammer the brakes just before the marshal stand into Mulsanne corner. I now have to hit them about 20 yards earlier.

I have to brake for the kink before Indianapolis. I didn't have to last night and was able to start the braking process during the kink.
Braking before the Ford Chicane, Dunlop Curve, and Tertre Rouge is also lengthened.

I haven't tried any other races yet b/c it's still a work day I'll do more testing tonight though but it definitely appears that in this specific race the AI has been boosted by an incredible amount.

 

[appu]

Do I remember it correctly that this was 3 laps before the hotfix?

I did this race last night and again this morning to check out the hotfix (🌶️🌶️). Observations:
1. I got pole by almost two seconds. Bonelli's lap was 1:37.x, 1 second slower than its practice time (maybe it had slipstream in practice).
2. No change at the race start. Two Sophy's still leapfrogged me, just like last night. A third would also have jumped me but had to slow down for turn 2.
3. The race felt the same as last night. Despite squeezing Bonelli's Sophy at corner entries to compromise its exit speed, it simply rocketed past me on straights. Rinse and repeat for multiple corners every lap. Bonelli won by finishing 0.3 seconds ahead of me, again same as last night.
4. I don't remember very well but the field might have been tighter today compared to last night, despite the race being 2 laps longer.

 

[Bojador]

Plenty to criticize about the PP, but overall I absolutely love it.

To my surprise the cars are my favorite thing about it. Two of them have made it into my favorites tab.

I'd love to be able to get all of them. So many cool cars in this pack - and we only get to keep 6 of those. (Two of them being rally cars...)

Tuned cars could become a new obsession of mine.

Really hope the format does well and we can get new PPs like every 3-4 months or so.

Much better than a freaking Panda, that's for sure.

 

[Obizzz]

As of 1.66 you can now disable the names over the cars... in the replay but not when racing

 

[Lomic]

As of 1.66 you can now disable the names over the cars... in the replay but not when racing

ah ! short joy before reading the end of your sentence 😑

 

[Talon16]

As of 1.66 you can now disable the names over the cars... in the replay but not when racing

 

[mynameis66]

I did a race no qualifying with the NSX at Fuji and was down by 18 seconds on lap 4. Again, I’m not trying to blow through the events. Not like this is a limited time offer.
I’ve got the gift cars I like and using those in Custom Races. I’ll get to other events and might rerun some at another time. I’m enjoying doing my thing.

As do I.Im no newbie but just not very good.lol I started on 2 peppers but found it too hard so played on zero.Im not complaining just enjoying the game how i want to play it.👍

 

[Mike Hamilton]

Looks like Sophy 3.0 likes to rejoin track safely.

I noticed that as well. Only thing missing is the turn signal… 😉

 

[JC_Dude]

Does the grid still have a big difference in times or the race is more closely together?

I didn’t pay too much attention to the rest of the pack, but I lapped 1 car and finished around 20s ahead of P8. P3-8 were all pretty close.

 

[Skinny McLean]

I was just making a point that it could technically run on PS4, how many instances it could support or what bottlenecks it would face is unknown.

But Sophy never ran on those PS4s - all the "learning" (adjusting weights and biases) happened on PCs with expensive GPUs. The PS4s were merely used to let the models interact with the game.

 

[wowbaggerBR]

I was just making a point that it could technically run on PS4, how many instances it could support or what bottlenecks it would face is unknown.

and I am making the point that, no, it can't and that the video you are linking here does not show Sophy running on a PS4.

From the paper, which is linked in that YouTube video, seens pretty clear the game was an unmodified version and that the AI was running on a much, much more powerful PC.

We conduct our experiment using Gran Turismo
Sport (GTS). We train our algorithm on a desktop with an
i7-8700 CPU and a GTX 1080Ti GPU. We use a custom
implementation [6] of the Soft Actor-Critic algorithm that
is based on the open-source baselines [21]. GTS runs on a
PlayStation 4, and we interact with GTS using an Ethernet
connection. We treat GTS as a black-box simulator since we
do not have direct access to the vehicle dynamics and the
environment in GTS. We choose “Audi TT Cup 16” as the
simulated car model and “Tokyo Expressway - Central Outer
Loop” as the race track. The hyperparameters of SAC for the
training are listed in TABLE II.

 

[GTvsForza]

Hopefully now I'll win the Willow FF and Deep Forest Tuned races without having Miyazono racing in an X2019 test mule...

 

[art3mis]

I've got new screens!

ok, so the first screen is the previous GT ONE Le Mans grid. Shows large gaps between the AI and a blisteringly fast time for the 1 Chilli car

This next shot is the exact same race same situation. I did not leave the pits to qualify. Check the gaps now AND check the qualifying

order is different.

Finally here's my latest qualifying grid for the Group C Le Mans 24 Endurance race:

I dropped my QT from a 3:08:xxx to a 3:05

Every single AI qualified a minimum of 4 seconds faster than the fastest AI time I saw in qualifying last night. Best time last night was a 3:12, slowest time I saw this afternoon was a 3:08.
The group is tighter as well and the order is different as well. Last night the Jag was the clear AI car to beat with V. Gallo followed by K Drumont in the Porsche and Lopez in the Jag rounding out the top 3.

Today there's far less "disorder" in the ranks. Porsches clumped to the top after me. A lone Mercedes in 7th breaking up the Porsche domination. Then comes the Jags, then the big group of Mercedes, finally below the line are the two Mazdas and singular Nissan.

Obviously the patch shook things up a lot. I think they recognized that gaps were super wide in many qualifying sessions. I've clearly adjusted to the brake distance increasing and found speed in other places due to the increase in acceleration and overall top speed.

For this race I think the AI has definitely been boosted and that's probably a good thing. The Endurances are the End Game Content of the DLC if only because it takes the most stars to unlock them. The Le Mans Endurance did not feel like a true End Game race last night.

It does today.

 

[eran0004]

But Sophy never ran on those PS4s - all the "learning" (adjusting weights and biases) happened on PCs with expensive GPUs. The PS4s were merely used to let the models interact with the game.

Yeah, it seems like the PS4 was used as a "multiplayer" platform where the AI agents were controlled by an external computer. All the PS4 had to do was to send sensor data to the external computers and collect steering inputs from them.

 

[WJHMH]

I expected with the recent patch, the penalty system with the Power Pack would be fixed. I just raced the Porsche event at Monza & you can still cut the track without any fear of penalties.