Why no amd multicore support.

why do beamng devs hate amd. like i understand, it's a budget solution, but why are beamng devs treating amd users like shit. beamng isn't even using half of the available technology on most amd processors. and on 8 core processors the game only accounts for 4 cores. so why.

 

AMD has inherent limitations. Such as the fact you don't really have eight cores.

 

lol.. I use an AMD 4770k, performs as it should, AMDs are not the greatest processors. That being said, you do realize no games will use all 8 cores, hell barely any will use over 1 or 2. AMD processors aren't as supported because they have a smaller market share. They are still supported obviously, but if there is an optimization that makes intel cpus run the game better but make AMDs run it a tiny bit worse, those are gonna be used.

 

(imported from here)

 

Yeah. Sure.

I'm happy with my 8320 performance in BeamNG. It can handle a lot of cars and still have a good FPS rate. (Except for the T75.)

 

Lolwat, thats not even a real thing

Sent from the 3rd galaxy from the talks of tapping


At op, beamng devs didn't do anything to cause these problems, amd cups just dont have good single threaded performance, and beamng simulates vehiclea by 1 car per thread, all you can really do is overclock, provided you have decent cooling

 

whut

..

 

AMD "cores" are really two core modules which share components, namely the floating point math unit (presumably decimal math is useful to physics and cannot be shared.) Strange bit is it reports my FX 6300 with the same "issue" as a hexcore.

As for multithreading, spawn more cars and you'll close the gap between AMD and Intel somewhat, especially against an i5 if you have an AMD 8-core. Each car is simulated on one core, and it'd be incredibly complicated to do otherwise.

And inb4 "hyperthreading makes no difference in games as shown by these 7th gen console ports! i5 and i7 are the same for games!"

 

We just explained why you have low performance, it has nothing to do with optimization, and everything to do with amds low single threaded performance

Sent from the 3rd galaxy from the talks of tapping

 

Speak for yourself.

AMD cores are not fully standalone like Intel's with huge amounts of shared resources including the most notable - the floating point unit. An AMD quad core has 2 fpus, 1 for every 2 cores, Intel has 4 on its quad cores, as far as I can tell, BeamNG hammers fpu usage so its bloody obvious that Intel has the edge over AMD there. Plus you'll find per core per GHz, Intel has a far higher MIPS rating allowing for higher performance in general. BeamNG being a highly demanding application with high floating point usage will always run better on Intel.

Application developers don't optimise for AMD vs Intel. There is incredibly limited room to do so.

BeamNG devs aren't treating AMD users like shit. Windows task management is just clever enough to not stick 2 floating point heavy threads on cores sharing the same fpu and hence halves the cores (as again, application developers don't choose which cores a program runs on, windows does)

BeamNG can do fuck all to optimise for AMD. You should be the one optimising your rig for BeamNG.

Oh, obviously you also need 5 vehicles in game for BeamNG to use 5 cores

 

There really isn't much you can do to help with AMD CPU performance without hours and hours of work. Graphics are kind of an issue with AMD cards that really should not happen however that is down to torque 3d.

 

AMD does not have hyperthreading

 

Well they do have a type of multi-threading but I really don't know why I typed that... it was early

 

"Multi threading"

You mean having multiple cores? (I think you're talking about how AMD CPUs like my 8320 have one floating point unit for every two cores instead of every single core.)

 

Nah. An AMD "octacore" chip uses 4 AMD modules. Each module being 2 cores, each core being almost fully fledged, full ALU and instruction pipeline, its own registers etc. But although there are 2 cores in that module, there is only 1 cache (which can be a benefit as direct core to core transfers no longer need to go via main memory which speeds things up a bit, but that is for specialist applications only, custom OS designed for 1 purpose, windows and even linux have no real way of taking advantage of it besides what the hardware almost automatically stores in cache - still enough for a slight edge over cross thread data sharing on intel though). There is only 1 interface to main memory, there is only 1 instruction prefetcher (thankfully one designed to accomodate for the fact it is feeding 2 cores not 1 like it would on intel chips), there is only 1 peripheral interface and there is only 1 floating point unit.
These are all things that the AMD APU's, the pre FX series AMD chips and all Intel chips have per core, its a trait unique to the FX series. Exception being the next gen APU's which arent out for consumers yet, they are being switched to the module architecture soon, the xbox one and PS4 processors are module based APUs.
Plus the different pipeline and ALU on an AMD core is slower than intels at the same clock speed. The FPU is slower at the same clock speed.
For pure integer maths, FX series octacores can be treated as pure octacore designs just with bottlenecked memory and impeded performance at a given clock speed. Floating point heavy stuff, they are crippled by having 1 FPU per module instead of per core.
More significantly. The APU modules are 4 cores per module, but those do have 2 FPUs per module. It is thought the xbox/ps4 apus may have been modded further (as it is advertised as a modified APU architecture) to be 4 fpus per module.

I think modules may be how AMD gets cost down these days. Much simpler design having half the FPUs alone (they are very complicated devices, they are almost their own processor) and for the majority of applications it makes no difference. FX chips used to have the real edge in file/web/etc servers where their ability to have 8 threads running (with almost no floating point math) trumped over intel being able to have 6 at once.


Hyperthreading, you only have 1 core. But the pipeline is more intelligent and will have the instructions for 2 threads and the ability to bump them up and down the pipeline a bit. Hyperthreading technically being 1 core exposed as 2 tends to have some issues under high loads, a stall in 1 thread can cause the 2nd thread to stall itself. If not carefully managed it can cause severe problems. Alot of professional intel servers have it disabled as 1 core without hyperthreading can be relied upon to be there 24/7, however with hyperthreading, it has the potential to stall. Alot also leave it enabled and just rely on the ability of windows or linux to recover the stalled core (which they can do quite effectively, but do you want to rely on that when business depends on it?). IBM's next generation POWER chips also get a hyperthreading-alike feature, but with 8 threads per core and implemented differently (I think slightly more effectively, I think their approach is more akin to a prioritised barrel processor than intels very bizarre intelligent pipeline approach), plus with 8 core models being available, 64 threads per chip, to add to that, most POWER servers have multiple processors, hell, again, 8 isnt uncommon, 8 cpus * 8 cores * 8 threads per core = 512 logical cores per motherboard....

 

Just to add to the insanity that the PowerPC chips are, intel is coming out with a new proc (Itianium or Xeon, I can't remember.) that is 32core/quad-threaded/8 to a board. That's 32*4*8 or 1024 "cores"
What will they use it for?

 

Running BeamNG MP servers?

Depending on how fast each thread is.

 

LOL shit, I meant FX-4100.

 

I think it is itanium (which is not an x86 chip like all their other products). Itanium really isnt very popular though, typically it wasnt any better than either x86 or PowerPC, cost more and of course being newer to the scene there was no infrastructure to support it so in order to gain popularity it is going to have to do something that the others don't really really well.

Oh, turns out my previous maths was wrong as POWER8 will be available in upto 12 core models not 8.
4ghz base clock although claimed to run anywhere between 3 and 5ghz (clock unit wont go below 3 when running within spec). 64kb L1 data and 32kb L1 instruction caches (does cache them seperately but is not itself a harvard or modified harvard achitecture) per core. 512kb L2 cache per core and 8mb of L3 per core. L4 cache resides on the motherboard, upto 128mb per core but shared.
Each core has 4 floating point units, screw that AMD FX series 0.5 FPUs per core. 2 instruction fetch units, again, AMD has 1 per module. 1 true decimal floating point unit (now those are hideously complicated devices compared to a traditional floating point unit). 2 fixed point arithmetic units too, these are less complex than a traditional floating point unit but not as commonly used. Currently no x86 device on the market has either a fixed point unit or a decimal unit.
Oh and get this. It has another processor within the chip specifically for monitoring voltages and performance, it in turn operates all the 1700+ voltage regulators to ensure the core voltage is stable. So much for tinkering with that from the bios, it does it itself.
A company called TYAN have unveiled a POWER8 motherboard in standard PC ITX formfactor. NVidia are to release GPUs compatible with the POWER8 databus (should be easy as it is just another protocol layered atop of and electrically compatible with PCIe 3.0).
Very powerful chips. Probably quite expensive and of course they are rather special purpose. Been a long time since POWER has been used in home computing. POWER4 with the low performance model PowerPC 970 in the old apple G5's. Yeah, the Powermac G5 used what was considered a low end PowerPC chip, so much for a high performance workstation.

 

Microsoft was also working on IA-128 with a release in 2016-2020 IIRC so that just adds more layers. Supercomputerao
are just getting crazier. (I honestly think it is more interesting than the desktop area)