fsx-ms Come on FSX, it's 2015 !

[shivers9 1,174]

 

 

but what FSX really needs is a full overhaul and made into a 64-bit program. Until then

 

That process has been on going for some time now. It's called Prepar3D.

[magnetite 38]

Most games and such usually only get around 1 to 1 1/2 years of support for patches. FSX got two years and had planned for four years. Even if they had finished the four years of support, I don't recall them ever mentioning 64-bit. The plan was FSX with DX10 and 32-bit.

[Luke 782]

Because, for FSX, clock cycles make the biggest difference out of all possible variables

 

No. It's single-threaded performance. And while it's roughly related to clock speed for similar processors, there's a difference.

 

Go take that 3.4Ghz Pentium 4 and match it up against a modern Haswell. For fun, underclock the Haswell to 1.2Ghz. Who's going to win?

 

Even if clock speeds haven't dramatically changed over the past 10 years, single core performance has continually gone up.

 

Cheers!

 

Luke

[JimmiG 244]

No. It's single-threaded performance. And while it's roughly related to clock speed for similar processors, there's a difference.

 

Go take that 3.4Ghz Pentium 4 and match it up against a modern Haswell. For fun, underclock the Haswell to 1.2Ghz. Who's going to win?

 

Even if clock speeds haven't dramatically changed over the past 10 years, single core performance has continually gone up.

 

Cheers!

 

Luke

 

True, single-threaded performance has gone up, but at nowhere near the same rate as GPU performance has improved. If the sim took advantage of this, it would run at hundreds or even thousands of FPS.

 

The 7800 GTX was high-end when FSX came out in 2006. It has 165 Gflops (0.165 Tflops) of processing power. The Titan X has got 7 Tflops (7,000 gigaflops) of processing power.

[Luke 782]

 

 

True, single-threaded performance has gone up, but at nowhere near the same rate as GPU performance has improved.

 

GPU performance has gone up in the same way as CPU performance - it's wider. Modern GPUs barely run at 1Ghz - they just keep throwing more and more execution units onto the silicon because graphics is an "embarrassingly parallel" compute problem.

 

Single-threaded performance is yesterday. Going forward, one absolutely needs to rewrite code to handle multiple parallel execution streams. It's hard, but necessary.

 

Cheers!

 

Luke

[Griphos 138]

No. It's single-threaded performance. And while it's roughly related to clock speed for similar processors, there's a difference.

 

Go take that 3.4Ghz Pentium 4 and match it up against a modern Haswell. For fun, underclock the Haswell to 1.2Ghz. Who's going to win?

 

Even if clock speeds haven't dramatically changed over the past 10 years, single core performance has continually gone up.

 

Cheers!

 

Luke

 

Again, I said for FSX, not some general notion of performance.  And, for FSX, the biggest difference in frames per second comes not from improved single-thread performance, but from clock speed.  

[Luke 782]

Again, I said for FSX, not some general notion of performance.  And, for FSX, the biggest difference in frames per second comes not from improved single-thread performance, but from clock speed.  

 

I'm sorry, but you're 100% wrong. Do you really believe that a 3.4Ghz Pentium 4 would run FSX faster than a 1.2Ghz Haswell?

 

Cheers!

 

Luke

[Griphos 138]

Have you tried it?  I'm sorry, but your example is rather silly, isn't it?  No one is going to be running FSX on a Haswell chip at 1.2.  We are almost all running the most recent architectures, and switching from one of those to another will make less difference than increasing clock speed.  

 

I've run FSX on LOTS of different CPU architectures, and without exception, the faster the clock cycle, the faster the FPS.  There is, of course, improvement with improved architecture, but FSX is simply clock-cycle hungry.  And I said the biggest factor in performance, not the only factor.  

 

We are, though, beating a very literal dead horse.  There is, as I have pointed out, no point to this conversation really.  There is no development team for FSX any more.  There will be no changes to the coding.  It is what it is, and current expectations are moot, and new architectures won't change the coding.  .  

[neucoas 82]

Not only the GHz count in a processor, the IPC (intructions per cicle) has to performance. So thats why a P4 3.0GHz barely could move FSX at full settings, and now a Core i3 at the same clock speed can manage it without issues.

[Bjoern 1,157]

I'm sorry, but you're 100% wrong. Do you really believe that a 3.4Ghz Pentium 4 would run FSX faster than a 1.2Ghz Haswell?

 

Definitely.

 

As much as I like my i5, had I only got it for improved FSX framerates at the same clock speed as my overclocked Q9450, it would have been a waste of money because the difference would be marginal. You'll get a bit more difference the farther you go back in CPU generations, but anything that can do 3.5 GHz is going to run FSX well.

 

There's much more to the equation than CPU clock speed. Memory bandwidth, latency and clockspeed has improved quite a bit over time and SSDs were a game changer and probably the single biggest hardware-side improvement one could have for FSX.

[Luke 782]

 

 

Have you tried it?  I'm sorry, but your example is rather silly, isn't it?  No one is going to be running FSX on a Haswell chip at 1.2.  We are almost all running the most recent architectures, and switching from one of those to another will make less difference than increasing clock speed.

 

Of course it's a silly example that no one would do in real life. But it's also a really useful example, because it demonstrates just how far Intel (especially) have divorced clock speed from single-threaded performance. The former has stayed more or less the same. If you look here at the single-thread benchmarks, Haswells have three times the single threaded performance at roughly the same clock speed as the Pentium EE.

 

 

 

I've run FSX on LOTS of different CPU architectures, and without exception, the faster the clock cycle, the faster the FPS.

 

And did you control for other variables?

 

Look, the two biggest jumps in my FSX performance were when I went from my 3.2Ghz P4 to a 2.4Ghz Core2 Duo, and from my 3.2Ghz Q9500 to a 3.0Ghz Sandy Bridge. In both cases the CPU frequency went down, but single-core performance went up. In FSX, and everything else.

 

 

It's very, very hard to write code that is purely frequency dependent on the CPU, and performs linearly with CPU speed across different architectures. The only thing that I can think of is swapping registers around a lot, but moving the value of EAX into EBX and back isn't particularly interesting or useful. Useful code has two hard things to deal with - memory access and branches, and both are complicated.

 

(For those following at home, if you think that CPU cores take instructions and process them one at a time, in order, the 1980s want their processors back. Ever since the P6 (Pentium II) core, x86 and x64 processors have been "out of order execution". They have long pipelines that depend on decoding instructions and fetching data in advance of what is actually being executed.)

 

Let's say you want to do something simple, like loop through an array of numbers. How many do you need to add together for the sum to be or exceed some arbitrary amount? Well, there's some basic addition, but more importantly you need to load values from memory. That depends a great deal on the memory controller, the caches and the latency of the memory. And until the memory loads, the CPU waits. Then once do the addition and the comparison, we have a decision to make. Are we equal or greater than the arbitrary amount? The CPU can't wait for the answer - it needs to decode instructions in advance so it guesses as to what branch is going to be taken. And if it guesses wrong, all that work it's done decoding and fetching data needs to be discarded and it waits (AGAIN) for the new data and instructions to be decoded and executed. Then once we hit the magic sum we probably need to do it again because the branch predictor will guess wrong.

 

In that simple example, the actual execution speed of the CPU doesn't matter all that much - what's far more important is how long it sits around waiting for data and instructions to be available. The Core architecture was so much better than the P4 because it gave back a few hundred megahertz in raw clock speed in return for VASTLY better branch prediction and other optimizations - that's what IPC is all about; not just about how long an instruction takes to execute, but how many clocks are spent waiting around doing nothing.

 

This is also why early Athlons wiped the P4 despite being clocked almost a whole Gigahertz lower.

 

This is why I'm so certain that FSX has nothing to do with raw clock speed. I'm pretty amazed at what Intel and AMD have done to try and tweak single-threaded performance. But the reality is that CPU performance has grown rapidly and amazingly. Programmers, on the other hand, are the same dumb sacks of meat we always were.

 

Cheers!

 

Luke

[Bjoern 1,157]

Seems I may stand corrected.

 

Still, I'd love to see some side-by-side comparisons. Does anyone have the time and available hardware?

 

 

- Edit:

 

Scratch that. The memory (DDR2 vs DDR3) and memory controller (North Bridge vs integrated) would skew the result, even if you used the same HDD and GPU in both systems.

[goates 576]

 

 

What I do find odd is that MS (ACES) didn't see it coming in 2005-2006. I imagine that what they were thinking was that in 2-3 years CPU speeds would be about 5Ghz or more... They completely missed it there.

 

FSX development started in 2004, if not earlier for the initial planning. At that point Intel was talking about 10GHz single core CPUs by 2009 or 2010. This obviously changed, and FSX was too far along to go back and change course. They made some improvements with SP1 and were moving towards bigger changes for FS11. However, before that could happen ACES dropped.

 

 

In that simple example, the actual execution speed of the CPU doesn't matter all that much - what's far more important is how long it sits around waiting for data and instructions to be available. The Core architecture was so much better than the P4 because it gave back a few hundred megahertz in raw clock speed in return for VASTLY better branch prediction and other optimizations - that's what IPC is all about; not just about how long an instruction takes to execute, but how many clocks are spent waiting around doing nothing.

 

The issue with FSX is that it was designed to expect similar increase in both clock speed as well as IPC to what had come before. P4s went from 1 to 2 to 3+ GHz as well as having improvements to IPC, which made for fairly significant performance increase. With the Core architecture, however, we've really only been getting the improvements on the IPC side which typically amount to 10-15%, maybe 20-30% in some cases, increases in performance over the prior generation (excluding the initial jump from the Netburst to Core designs). And of course after the switch to multi-core designs, other developers also had time to re-write or adapt their code accordingly, where FSX is stuck the way it is. So raw clock speed is not the entire story, it is a very large part of it where FSX is concerned.

[Boeing or not going 578]

That process has been on going for some time now. It's called Prepar3D.

No P3D process is upgrading the graphics engine and eventually moving to a higher version of DirectX, right now they are fixing DX10 preview and making it full DX10 compatible. P3D has not been about moving to 64bit if they do support it it will be a long way away and it will break every current addon.

[tf51d 284]

No P3D process is upgrading the graphics engine and eventually moving to a higher version of DirectX, right now they are fixing DX10 preview and making it full DX10 compatible. P3D has not been about moving to 64bit if they do support it it will be a long way away and it will break every current addon.

P3D is DX11 not DX10 which along with DX9 isn't even an option for P3D. I think you are thinking of FSX, with Steve's DX10 Fixer.