1911 replies to this topic

[SnagaDance]

Karl Berg, on 26 December 2014 - 12:21 PM, said:

Exceedingly cool! Thank you very much for the links sir! I will definitely be reading through them in detail.

I've followed GPGPU work at a high level for some time, as it really seems to be the closest one can achieve to Cell / SPU style programming on a PC. I'm really excited to one day see that first GPU vendor really nail down an SPU style interface, with full application level control over DMA channels, a working C/C++ compiler for program binaries, and full intrinsics for assembler support, as well as 128-bit or better SIMD vector support. I initially thought AMD might provide such an interface when I first read about their new APU tech, but they seem to have taken a different direction. Regardless, once someone opens raw metal access with a decent API, I think the sheer number of FLOPs those GPU chips are capable of will simply explode in terms of high profile applications.

For us on the server side specifically, since our code is so integrated with CryEngine and CryPhysics, it would be a significant challenge to re-implement our HSR systems to use GPGPU algorithms. Being able to write external helper programs with explicit DMA control, as described above would mitigate that pain a great deal, since there would be potential for sharing code and potentially even reusing some of the PS3 job scheduling logic already built into the engine.

Client side, there are some super easy wins with GPU based particle simulation. Our version of CryEngine doesn't support GPU based particle simulation; and we haven't had the resources to attempt this rewrite ourselves yet.

This is almost complete gibberish to me, but I love the enthousiasm I'm detecting here!

[Karl Berg]

Cimarb, on 26 December 2014 - 03:00 PM, said:

Not 100% sure, but I think he meant "unit" as in new mechs. So, when the Timber Wolf was released, how did that affect the Elo distribution? (correct me if I am wrong, Goose)

Ahh, I see. I'd have to go back and look at the historical data for those release dates when I get back in the office.

[Karl Berg]

Rebas Kradd, on 26 December 2014 - 03:06 PM, said:

Karl,

How is morale in the PGI offices? You guys have had a hell of a comeback this fall and I hope you know how much we appreciate everything you've accomplished.

Everyone was pretty tired out before breaking for Christmas Morale seemed pretty good though. I think most or all of us really like the direction CW ended up taking, and were excited to put in those extra hours to help make the launch as smooth as it could be.

Also a big shout out to those players who were invited to private testing, and took their own time, every day for at least a couple of weeks, to pre-test the whole feature for us. We ended up several issues due to that extra testing, and I know Russ and Paul used the playtesting results to make extensive balance changes prior to public beta release.

[Karl Berg]

Modo44, on 26 December 2014 - 10:38 PM, said:

Thing is, I get randomly bad hit registration despite <1 ping jitter and a high-speed connection (90 Mbps both ways). One match goes great, another has half my shots disappear into thin air. Be straight now. Is the system hurting on communication latency issues? Are there too many CryServer instances per physical server? Or maybe buggy CryServers crapping out? Something else?

That is weird. If you manage to catch any video footage of the game acting up Modo, please be sure to send it our way. If you could, for all videos you send, please remember to open the scoreboard for at least a few seconds so we can observe what the game thinks your ping is. Could be due to any number of issues, but good video footage helps immensely when trying to track specific issues down.

[Karl Berg]

Sereglach, on 26 December 2014 - 10:47 PM, said:

Dear Mr. Berg,

For one, I have to agree with the sentiments that PGI has been doing amazing things since the separation from IGP. I greatly look forward to seeing what the new year has to bring. With that said, it ties into my question:

First, a little backstory. Long ago (back in January of 2014) we were told that the Flamer would be receiving a rework that was going to tie in with the release of the Firestarter. In February we received word that it was in engineering's hands and that they were making progress on it, here: Paul Inouye's Feb 24th Quick Update. Then, in September, we were told here, by Russ, that nothing would be happening until they've finished frying the big fish (CW Phase 2 release).

Now that CW Phase 2 has released, do you know if Flamers are back on the table yet? or when they will be addressed again? Any ideas what they're planning on doing to them?

Trust me, as my signature and stats show, I've got plenty of experience with them (currently sitting on over 74k flamer damage between current and archived stats), and plenty of ideas to help. I'll be happy to do whatever I can to assist in it. They're my babies and I hope you forgive me making the direct prod to see if there's anything being done to help them out.

Thank you for your time and consideration.

Flamers, eh? I forced Mr. Windover to put together a test map for me, consisting of dropships carrying only flamer turrets when he was building out the new dropship turret system. It was every bit as spectacular as I thought it would be.

I'll pester Paul when we're both back to see if he has any pending flamer plans. Best bet for a prompt answer on this topic is to bug him directly though. He's posted on the forums a couple times during his break, so he might see a nice prominent thread with his name on it.

[Cimarb]

Karl Berg, on 27 December 2014 - 05:25 PM, said:

Flamers, eh? I forced Mr. Windover to put together a test map for me, consisting of dropships carrying only flamer turrets when he was building out the new dropship turret system. It was every bit as spectacular as I thought it would be.

I'll pester Paul when we're both back to see if he has any pending flamer plans. Best bet for a prompt answer on this topic is to bug him directly though. He's posted on the forums a couple times during his break, so he might see a nice prominent thread with his name on it.

Screenshot or it did not happen! (seriously, I want to see a screenshot, or even better, a video!)

Oh, and I thought it was against the forum rules to call someone out in the thread name?...(with the notable exception of when you make a joke about it, such as in this thread)

[Karl Berg]

Goose, on 27 December 2014 - 11:34 AM, said:

The later was where I was going ("intro" was to have been the phrase,) but Karl's version sounds like great fun: Do Teams learn/ adapt meta faster then PuGs?

It's part of my quest to get Elo-per-'Mech, in which I didn't hear a "no" in that one Podcast, the way the reporting on the Podcast said it had …

While I'm here: We've gotten a couple of statements along the lines of "this loads the GPU, while that loads the CPU," and as time has gone one, I find this less and less adequate. I just had one small test where I'd turned up everything 'cept Object Detail and Damage Glow, and gotten good results. In retrospect, it seems obvious a hexacore like me could'a done something like that, and even then I'm assuming Teh Object Slider effects the ca_thread0Affinity the most above the others.

I've got this half-formed list of things like "Shaders moves GPU Load 10 percentage points at a time" that I still need to test out, but it'd be nice if we could get something from Y'all Devs on the subject; Something concise …

Elo per mech model makes a good amount of sense. It's one of those things we'd probably have done already, had the simulations backed it up. I suspect the reason it appears to perform worse is due to large amounts of players bouncing between various trial mechs, and not playing enough games on a particular mech to properly converge their Elo.

More investigation and simulation is needed on this topic.

As for settings and associated resource costs, funnily enough I suspect our QA/Dev support department would have the best insight into that, if anyone. I don't know of any engineers that have spent time examining the different load points for each setting across a variety of hardware. Sorry, I don't have a concise answer for you

[Karl Berg]

Cimarb, on 27 December 2014 - 05:30 PM, said:

Screenshot or it did not happen! (seriously, I want to see a screenshot, or even better, a video!)

Oh, and I thought it was against the forum rules to call someone out in the thread name?...(with the notable exception of when you make a joke about it, such as in this thread)

I'll see what I can do on the screenshot front. I didn't take any pics at the time, mostly was just mesmerized staring at the screen.

Is putting a dev name in the thread title really against forum TOS? You could be right (*cough* I haven't read the TOS). Maybe instead of his real name, use 'Glorious Lead Designer', or 'He Who Shall Not Be Named', or even just 'Pink Thunder'; although I can't imagine attaching his name to a polite and well written question would cause issues, considering the title of this thread.

I could be entirely mistaken though, so please don't do anything against TOS just because I told you to do so.

edit: Looks like Paul may have removed his old sig. 'Pink Thunder' may no longer work, which is a shame since that sig was pretty fantastically done.

[Kell Morgan]

Karl Berg, on 27 December 2014 - 05:33 PM, said:

As for settings and associated resource costs, funnily enough I suspect our QA/Dev support department would have the best insight into that, if anyone. I don't know of any engineers that have spent time examining the different load points for each setting across a variety of hardware. Sorry, I don't have a concise answer for you

I have also noticed an abnormal CPU usage. I have been playing with my system and OC settings.

If I OC my video card clock speeds I notice almost no difference in FPS. If I OC my CPU from 4.0Ghz to 4.2Ghz then all of a sudden I have an average of 10-20 FPS more.

I was wondering with the new 64 bit client how it uses multiple cores available on modern machines. (Like my AMD 8 core) Vs the old 32 bit client. From what I understand for most programs the 64 bit coding is more efficient than 32 bit. (Talking about CPU usage not the 32 bit RAM limits.)

Also I was wondering with the new 64 bit client can we see more optimizations of the client coding making it run liter and not as heavily reliable on CPU strength?

or would this be something for the QA department also?

[Wintersdark]

Kell Morgan, on 27 December 2014 - 05:54 PM, said:

I have also noticed an abnormal CPU usage. I have been playing with my system and OC settings.

If I OC my video card clock speeds I notice almost no difference in FPS. If I OC my CPU from 4.0Ghz to 4.2Ghz then all of a sudden I have an average of 10-20 FPS more.

I was wondering with the new 64 bit client how it uses multiple cores available on modern machines. (Like my AMD 8 core) Vs the old 32 bit client. From what I understand for most programs the 64 bit coding is more efficient than 32 bit. (Talking about CPU usage not the 32 bit RAM limits.)

It is more efficient, but not by a huge margin.

As for using more cores, that's entirely down to how the game itself is written. Most games (though it is improving) don't really utilize large numbers of cores efficiently. That has nothing really to do with 64vs.32bit, however. You'll find most 64 bit games will use more cores better, but that's more just forward thinking than anything directly attributable to being 64bit.

Quote

Also I was wondering with the new 64 bit client can we see more optimizations of the client coding making it run liter and not as heavily reliable on CPU strength?

I'm sure we'll keep seeing client side optimizations, but I suspect that MWO is going to remain largely CPU bound for the near future.

[Sereglach]

Karl Berg, on 27 December 2014 - 05:25 PM, said:

Flamers, eh? I forced Mr. Windover to put together a test map for me, consisting of dropships carrying only flamer turrets when he was building out the new dropship turret system. It was every bit as spectacular as I thought it would be.

I'll pester Paul when we're both back to see if he has any pending flamer plans. Best bet for a prompt answer on this topic is to bug him directly though. He's posted on the forums a couple times during his break, so he might see a nice prominent thread with his name on it.

Thank you for the prompt response, sir. I checked the TOS and saw nothing about using someone's name specifically, unless it is falls under the "Name and Shame" policy. I do not believe asking him a question about flamers will qualify as a name and shame offense.

FYI, it doesn't seem as though is signature has changed . . . it seems as though it now goes to a dead image link. Looking at his posts directly still has the "Thanks Nekomimi for the new sig!", but the image appears to be a dead link, as it doesn't load regardless of what I did.

Anyway, I have started a thread, hoping he will give a reply, HERE. If you have any ability to draw his attention to the thread (or any of my other flamer threads, as well) that would be greatly appreciated.

Thank you again for the response, and for your time.

[pbiggz]

Hey karl, whats the likelihood of us seeing something like player made maps in the future?

In my opinion it is perhaps the best thing pgi could do to combat the limited map selection. Have players submit maps and then have the community vote on their favourites. The winners get a quick quality pass from pgi staff (just to make sure everything is up to snuff, though you'd be surprised at what alot of people in this community can do) and then the maps are added to the rotation.

[Void Angel]

Cimarb, on 27 December 2014 - 05:30 PM, said:

Screenshot or it did not happen! (seriously, I want to see a screenshot, or even better, a video!)

Oh, and I thought it was against the forum rules to call someone out in the thread name?...(with the notable exception of when you make a joke about it, such as in this thread)

Karl Berg, on 27 December 2014 - 05:46 PM, said:

I'll see what I can do on the screenshot front. I didn't take any pics at the time, mostly was just mesmerized staring at the screen.

Is putting a dev name in the thread title really against forum TOS? You could be right (*cough* I haven't read the TOS). Maybe instead of his real name, use 'Glorious Lead Designer', or 'He Who Shall Not Be Named', or even just 'Pink Thunder'; although I can't imagine attaching his name to a polite and well written question would cause issues, considering the title of this thread.

I could be entirely mistaken though, so please don't do anything against TOS just because I told you to do so.

Sereglach, on 27 December 2014 - 07:44 PM, said:

Thank you for the prompt response, sir. I checked the TOS and saw nothing about using someone's name specifically, unless it is falls under the "Name and Shame" policy. I do not believe asking him a question about flamers will qualify as a name and shame offense.

That's correct - The name and shame policy only prohibits "calling people out" in the sense of "discussing" in-game misbehavior. This is so we don't have long flame wars about who shot who in the back during some random match; a very good thing. A person can still be named, in the title or body of a thread, and can be held accountable/criticized for their actions on the forum - so long as basic civility and other forum rules are followed. Thus, this thread does not violate the forum rules, but I am very much glad to have it hidden away here where interested parties can use it without ... the masses:

[TheCaptainJZ]

Regarding names in titles: I think it was more of a mod staff discouragement than a rule. Possibly an IGP mod rule? As many things the devs said often had to be double checked against the publisher

[Vassago Rain]

TheCaptainJZ, on 28 December 2014 - 12:15 AM, said:

Regarding names in titles: I think it was more of a mod staff discouragement than a rule. Possibly an IGP mod rule? As many things the devs said often had to be double checked against the publisher

It's ancient Niko rules from 2012. Someone needs to revise them.

[Rogue Jedi]

Thank you Karl for taking the time to answer these questions, especially during your Christmas break.

this sort of thing earn you a lot of goodwill from myself and the other players, much like Russ's town halls and Paul's command chair posts where they actually take the time to go over player questions and concerns,

[Modo44]

Karl Berg, on 27 December 2014 - 05:33 PM, said:

Elo per mech model makes a good amount of sense. It's one of those things we'd probably have done already, had the simulations backed it up. I suspect the reason it appears to perform worse is due to large amounts of players bouncing between various trial mechs, and not playing enough games on a particular mech to properly converge their Elo.

More investigation and simulation is needed on this topic.

Perhaps smarter Elo seeding heuristics would help. You have average player performance data for any chassis. You also have specific performance data for each player. That is a lot of solid, up to date information. The idea would be to start a player close to their expected Elo for any given chassis. Calculate player Elo based on their current Elo for similar chassis, adjust depending on current global chassis Elo delta (how much better/worse the new chassis is globally than the ones you track for that player). That should clear a few of the first, big Elo moves, right?

With that working, you could also add re-seeding for chassis that were not used for longer periods of time. This would hopefully kill or at least lessen the constant bias source -- mech switching without Elo adjustments. You will never really adjust for build changes, nor unique skill (e.g. ****** TW pilots, Lolcust maniacs), but you can do a lot to prevent big Elo moves that seem to be happening all the time, for everyone.

Separately, consider giving noobies even lower Elo. MWO is really hard to learn, and treating each new pilot as just below average (1300 last I heard?) is another source of constant bias.

Edited by Modo44, 28 December 2014 - 10:44 AM.

[Goose]

Karl Berg, on 27 December 2014 - 05:33 PM, said:

I suspect the reason it appears to perform worse is due to large amounts of players bouncing between various trial mechs, and not playing enough games on a particular mech to properly converge their Elo.

Given that Elo is a trailing indicator of it, and distend to fail depicting the five minutes where someone has it all down-pat before getting slow with age, I don't know there is a "fast enough" to be had. Certainly, Elo not reflecting someone being much better at Conquest then Skirmish is chucked up as "it's a trailing indicator," as it should …

But lets play with this some: I say Clan Slow Lights, and the current Raven, play real different from Jenners and Spiders; How "converged" (to coin a phrase) is the Light-class Elo of someone whom's good at one set, and trying to learn the other sets' play-style? Will the K-factor ever move things along at the right pace for this guy, regardless of how often he's switching back and forth?

I can see how "all Catapults don't play the same" becomes the next limit to Elo accuracy, but "can't be caught dead without my cheese" is a thing we could use less of …

And about that K-factor Thing: How does one pick it? How does one pick it, when the "rule" about Elo not moving on an expected result gets bent? And how come it's not indexed to anything? If the K-factor drops way down for the stratospherically good Chess players, but not the stratospherically bad ones, then what am I looking at?

Karl Berg, on 27 December 2014 - 05:33 PM, said:

Sorry, I don't have a concise answer for you

That's what I get for using three dollar words; Do you know of any mere generalizations on the subject? The sweeping generalizations aren't cutting the mustard …



Thank you for all your time and hard work, Sir: It's very gratifying to have this chance to speak with you …

[Kageru Ikazuchi]

Karl,

Thanks for taking the time during your vacation to provide these great insights into the inner workings of MW:O!

One item that the HSR diagrams don't quite address (that I think I understand, but would kind of like a bit more detail) are the differences between what Player A sees and what Player B sees, and how HSR compensates ... and how this is different for two low ping opponents, two high ping opponents, and one of each.

A common gripe is that occasionally, when there is at least one high ping player involved, damage application occurs after a 'mech has moved behind cover. It seems that more often than not, it is the low ping player who complains about it, because it happens to them less often and each incident is more pronounced. My belief is that high ping players are affected by it much more frequently, and have just gotten used to it.

[IraqiWalker]

Karl Berg, on 27 December 2014 - 05:20 PM, said:

That is weird. If you manage to catch any video footage of the game acting up Modo, please be sure to send it our way. If you could, for all videos you send, please remember to open the scoreboard for at least a few seconds so we can observe what the game thinks your ping is. Could be due to any number of issues, but good video footage helps immensely when trying to track specific issues down.

One thing I should bring up is that terrain rendering sometimes causes part of this. I remember taking pot shots at a stalker that I had a clear line of sight to, and not getting a single hit to register. Moved two steps forward, one to the right, and suddenly there was a rock there that was blocking not only my shots, but also my line of sight. I should not have been able to see the Stalker, but because the terrain was there, despite not rendering, I did see it.