Planet side 2- Need to code for Multicore CPU's

[Rroff]

Theres a BIG BIG difference between coding multi thread support into something like banking and security software, where the specific order a lot of stuff is processed doesn't matter and something like a game engine where there are large amounts where you either have to process them in a specific order and each depends on the one before it or where threading the code produces so little in the way of gains its not worth the effort and potential issues threading it.

How well a game will thread will depend massively on the nature of the engine AND the game built on top of it. Even then most games only really load up 1 or 2 threads heavily and the extra threads don't always require massive processing capabilities.

Applications that are processing a lot of data but aren't frametime sensitive can generally be threaded much more easily and effectively than typical game engines where you need to get the workload done inside a few ms per frame and most of the processing has to be done inside the frame rather than left to do its own thing for a few frames.

 

[teppic]

What I'm getting as is why the results are questionable, as someone else mentioned and he was asked why they're questionable, and when I've answered, you've just moved the goal posts and answer in a silly "I'm not wrong" manner.

The results are questionable as they show a hierarchy which many other reviews show differently.

But we'll forget it, because you're simply not worth it.

Saying the benchmarks are questionable suggests that they're misrepresenting performance in the game, but these are not unusual results. Nobody expects every benchmark on every system to give the exact same hierarchy, especially when it's within a few FPS. Since it seems that even mentioning games like Crysis 3 run particularly well on AMD 8320/8350s gets Intel fanboys up in arms, I've given another example of software that uses 8 cores efficiently (Handbrake), which also makes use of all the latest CPU instructions.

Since you don't seem willing to accept anything I say regardless, it's best to forget it, yes.

 

[Martini1991]

Anyone who thinks that Crysis 3 runs poorly on FX6's/FX8's isn't worth listening to.
But whatever.

 

[Marine-RX179]

Why are they questionable? When all 8 cores are used properly and the software makes use of modern CPU instructions, the processor performs on par with Intel CPUs.



The AMD CPUs are only weak if the cores aren't being utilised, which is all to do with the software being run.

Except that we don't live in a perfect world which all games use 8 cores fully (and never will). With the launch of the next gen console it would no doubt help a lot in making multi-platform games much more likely to use the cores fully, but situation won't change much for the PC exclusive games (i.e. PlanestSide 2, and other mmos and real-time strategies in general...which hammer CPU the most), the Piledriver will still lag behind the likes of i5.

Piledriver is similar to how Richland is at the moment for gaming...they are in the right direction, but not quite there yet. Hopefully both Kaveri and Steamroller would be the answer to people's prayer in terms of performance.

Intel's performance increase over the last few gen since SandyBridge quite frankly was a bit a of joke...so I guess it is the perfect opportunity for AMD to catch up, and now that they already have 8 cores, focus on where it counts- upping the IPC.

 

[CAT-THE-FIFTH]

Good news to hear that PS2 is finally being rejigged to use more than 2 cores.

Even with a Core i5 in my desktop some of the slowdowns are frankly hilarious,and the game lists a CPU bottleneck.

 

[alex_123_fra]

Seriously though who benches games at resolutions of 1280 x 720?

Maybe 10-15 years ago.

 

[Marine-RX179]

Seriously though who benches games at resolutions of 1280 x 720?

Maybe 10-15 years ago.

That's because with lower res it removes the GPU limitation, thus compare strictly CPU vs CPU performance. Sure people can do it on 1920 res, but there's no single GPU graphic card available on the market which can do minimum frame rate of 120fps on ALL games for testing CPU performance.

 

[alex_123_fra]

That's because with lower res it removes the GPU limitation, thus compare strictly CPU vs CPU performance. Sure people can do it on 1920 res, but there's no single GPU graphic card available on the market which can do minimum frame rate of 120fps on ALL games for testing CPU performance.

Yes granted but doing this on Crysis 3 is pointless. Whether you bench at 1920 or 1280, your minimums are going to be way below 120 fps anyway.

 

[Mei]

its all great news
its going to be that console ports are better optimized for the pc than games that were made just for the pc lol

more they are forced to make stuff run good on multi threaded the better

 

[Marine-RX179]

Yes granted but doing this on Crysis 3 is pointless. Whether you bench at 1920 or 1280, your minimums are going to be way below 120 fps anyway.

It's not pointless. It's about finding what frame rate the CPU can push, or the point which having faster graphic will no longer increase frame rate due to CPU holding it back.

 

[humbug]

Except that we don't live in a perfect world which all games use 8 cores fully (and never will). With the launch of the next gen console it would no doubt help a lot in making multi-platform games much more likely to use the cores fully, but situation won't change much for the PC exclusive games (i.e. PlanestSide 2, and other mmos and real-time strategies in general...which hammer CPU the most), the Piledriver will still lag behind the likes of i5.

As Sony themselves say Planet Side 2 will not work on the PS4 as it currently is, so it will get a complete overhaul and as a result AMD Desktop users will get a huge boost in performance.

Their words

This is the point, the PS4 and xBox One are x86 with a CPU that's only strong enough if the load is distributed across most or all of its 8 weak cores.

By in large it will change the landscape, what's more once developers take that step forward then game features and technology will also start to develop as they will have more than just a small fraction of a CPU's power available to them.

MMO's that don't want to be on Game Consoles or just bring their game development up-to-date will continue to be stuck in 2003, and that's their prerogative, but it means those 'PC only Games' will be like games from 2003 in comparison to their Game Console counter parts.

AMD absolutely need to do better on per core per clock performance, I have said the same so many times here my self, but that is not what this thread is about.

I find it absolutely ironic and somewhat disturbing that it is Game Consoles that are or will drive forward gaming technologies, not PC's!
Some people need to ask themselves how that has happened.

 

[alex_123_fra]

It's not pointless. It's about finding what frame rate the CPU can push, or the point which having faster graphic will no longer increase frame rate due to CPU holding it back.

No but it is as it defeats the objective. In order to be effective at eliminating GPU bias in a game like Crysis 3 you really need to go lower than 1280 such as 640 x 480 or similar to achieve far higher fps. I am not disagreeing with the principle, just the method is not ideal in this particular case.

 

[teppic]

The movement towards games expecting 4+ cores will be rapid. That's not to say they'll all run effortlessly across 8 cores evenly, but we're already seeing the change with Crysis 3 and Far Cry 3, and it's definitely going to be the same with BF4 (which is being optimised for the 8350).

It's not going to help old games, absolutely. We can hope that Steamroller does sufficiently increase single process/thread performance to the degree that the gap at least shortens in games like WoW. If AMD can manage that, with the new games taking advantage of 8 cores too, they could turn things around on the CPU front.

 

[humbug]

I added to my post #31

 

[ubersonic]

I'm still unconvinced that many weak cores is a better way to go than a few strong ones for future mainstream CPU's and software. Crisis 3's coding does allow the FX8 to beat the i7 but its still taking AMD twice as many cores as Intel to get the job done, is this due to the poor per core performance of the FX or is it actually because optimizing real time software (I.E not encoding apps) for 8 cores is very lossy in comparison to a few main threads?

 

[JonJ678]

First point is that developing games is expensive and risky. Smaller companies have pretty much their entire company invested in each title, and go bust if the game fails. Hours tend to be long and brutal. I'm not sure "lazy" is a fair description - if you've got four hours to implement and test something, single threading may be your only hope.

... You can also easily get to the point where creation and management of threads is costing more than the gain from the multi-threading itself. As a result you have to write for a target number of threads, perhaps with the ability to take advantage to some extent of further threads if they're available. You'll realistically never (never as in right now!) optimise for more than about 4 threads

Edit: My experience is mostly with high level languages so game engine writers may not agree!

I thought I'd add a couple of thoughts from threading in a relatively low level language (C). Thread/creation and destruction is expensive but it's also avoidable. The design pattern is a "threadpool", but the idea is simply to reuse threads to call other functions instead of create/destroy. It's harder to code than parallel-for loops but efficiency is better.

I've found some odd results from varying the number of threads. The standard advice is threads=cores, probably optimal if the processing is well behaved. I've had some preliminary success with thread count > cores, tested up to 24 threads on 4 cores without dramatic slowdown. I think it's possible to improve on one thread : one core with clever scheduling. As a side effect you end up with code which expands to meet a relatively large number of cores.

Sadly it is very possible to spend more time sending data between threads than doing useful work. That's where the real difficulty seems to be.

Theres a BIG BIG difference between coding multi thread support into something like banking and security software, where the specific order a lot of stuff is processed doesn't matter and something like a game engine where there are large amounts where you either have to process them in a specific order and each depends on the one before it or where threading the code produces so little in the way of gains its not worth the effort and potential issues threading it.

Yeah - this would be true. Different threads doing different things is awkward, when the order of execution is critical it's very hard.

I'm still unconvinced that many weak cores is a better way to go than a few strong ones for future mainstream CPU's and software. Crisis 3's coding does allow the FX8 to beat the i7 but its still taking AMD twice as many cores as Intel to get the job done, is this due to the poor per core performance of the FX or is it actually because optimizing real time software (I.E not encoding apps) for 8 cores is very lossy in comparison to a few main threads?

The old C2Q systems scaled really badly past two cores because the FSB was limiting. In the case of AMD, it may not be weak cores that are the problem, but weak connections between cores. Plus AMD count cores differently - an Intel quad core has four floating point units, an AMD eight core has four floating point units. The ones in an Intel chip will be idle some of the time, but two cores will have to fight each other for access in an AMD chip.

It might be educational to find a benchmark between an Intel eight core and an AMD eight core - the Intel oct would be expected to absolutely trounce the AMD when maths is involved.

 

[ubersonic]

The old C2Q systems scaled really badly past two cores because the FSB was limiting. In the case of AMD, it may not be weak cores that are the problem, but weak connections between cores.

As a Piledriver core is not much faster clock for clock than a Core2 core I would guess its weak cores.

 

[JonJ678]

Let's not rule out weak cores and weak connections between them...

 

[=XDC=FluphyBunny]

So in summary buy intel 4930K - 6 cores, fast clock speed jobs-a-good-'un. (money no object and all)


For us poor people.
Its going to be very interesting to see the BF4 reviews/benchmarks. If 8 cores are used AMD's £150 chip could potentially be near Intel's £260 i7.

 

[JonJ678]

I *think* the 3820 is probably the best Intel chip currently on the market, but I'm having little success verifying this. The reasoning in essentially quad channel ddr3 > dual channel ddr3 with a suspicion that SB-E will scale poorly above four cores. Documentation and solid comparative testing are difficult to find.

Your q9550 stands a good chance of being quicker than AMD's £150 chip, so I wouldn't get too excited.

edit: your q9550 isn't at 3.5V though, whatever your sig says!