I've decided to try to design my next machine to be optimized for XPlane 64bit while still being able to run FSX reasonably well (as in significantly better than my current 5 y/o Q9650/280GTX).  Thoughts?  I'm ok waiting up to a year if need be.   I keep thinking 6+ cores is best, but it never hurts to be able to do a decent o'clock as well.

You don't need 6 cores nor hyperthreading, neither XP nor FSX will benefit from it in praxis. Go for a fast 4 core CPU with a high clock (i5/i7). GPU needs to be 2GB VRam minimum and should perform at the level of todays GTX670 (...or better...). XP needs both: Fast CPU cores and a powerful GPU.

 

Cheers

Flo

 

 

You don't need 6 cores nor hyperthreading, neither XP nor FSX will benefit from it in praxis.

 

Thanks FloB,

 

Can you tell me, is this recommendation from experience?  I ask because of this comment on their site:

 

X-Plane will take advantage of as many cores or distinct processors as you can afford. Having 16 cores split among 4 CPUs is not required by any means, but Version 10 would be able to use every one. No more than 4 GB of RAM is necessary, but the more VRAM you have, the better–X-Plane 10 can easily use 1.5 GB of VRAM at the maximum settings.

 

Because of this comment re 'as many cores' I've wondered whether it would be smartest longer term to go for an 8+ core Haswell but then I understand overclocking either isn't possible or is more limited w/ these.  I would actually prefer, if XP would run as well, to be able to run stock clockspeed w/ more cores, and skip overclocking outright, in the interests of maximum stability and longevity.  I keep my builds as long as possible these days which is why I'm still on a 5 y/o machine which still runs well. 

It is from experience, observation and also by theory. Read this for more insight:

 

http://forums.x-plane.org/index.php?showtopic=55346

 

Cheers

Flo

Can you tell me, is this recommendation from experience?  I ask because of this comment on their site:

 

X-Plane will take advantage of as many cores or distinct processors as you can afford. Having 16 cores split among 4 CPUs is not required by any means, but Version 10 would be able to use every one. No more than 4 GB of RAM is necessary, but the more VRAM you have, the better–X-Plane 10 can easily use 1.5 GB of VRAM at the maximum settings.

 

Because of this comment re 'as many cores' I've wondered whether it would be smartest longer term to go for an 8+ core Haswell but then I understand overclocking either isn't possible or is more limited w/ these.  I would actually prefer, if XP would run as well, to be able to run stock clockspeed w/ more cores, and skip overclocking outright, in the interests of maximum stability and longevity.  I keep my builds as long as possible these days which is why I'm still on a 5 y/o machine which still runs well. 

 

That line is misleading and should've been removed or clarified ages ago, as it severely skews expectations. Marketing is what it is, though.

 

As FloB's link mentioned, "take advantage of as many cores... Version 10 would be able to use every one" comes with a huge qualifier. I have an i7 3770K overclocked to 4.2 GHz, bought for the same reasons you're thinking about. The only time the cores and hyperthreads are maxed out is during part of the scenery loading phase. The rest of the time, the bulk of the work is done by a single core running around 50-75% (average of all cores is about 30-40%) at settings I think balance well (defined as max fps while video card isn't always at 100%.

 

If I knew then what I know now... well, I'd probably still get the 3770K. I'd look at the 3930K a bit harder, but for almost 2x the price and only 25% performance increase according to cpubenchmarks.com, I can't really justify the extra cost.

 

Also, the "No more than 4GB of RAM is necessary" is also outdated and unclear. Even back then, it should have meant 4GB for X-Plane, not 4GB for the entire system. Now that X-Plane is 64-bit, I'd drop at least 8GB into your system, preferably 16 GB.

The other main area where all cores would be used is for the AI aircraft which all use real flight models.

 

    This IMO is the wrong way to do things, it doesnt perform well and I think it will be superseded by 3rd party AI traffic that will not be such a burden on the CPU requiring as many threads as you have.

Rendering of all the final pieces that are loaded into memory can only be done on one thread. Other threads can be used to load data into memory and for AI cars, boats, planes and clouds, but drawing all of that has to be done in the main thread. That is the limitation of any graphics engine due to program achitecture. XP can use other cores and threads for the peripheral experience and run your gauges or ATC or anything else on any other core for the system, but it is similar to FSX limits in that the performance of the system is still bogged down the the maximum limits of a single core. The best single-core performance with limited additional cores for peripheral threads is a win/win. Overclocking isn't necessary for performance completely, I notice almost no improvement in FPS when I go from 4.5 to 5.0gHz with my clocks, but there is more stability in going from 3.8 to 4.5. There are fewer things that happen with the processing bandwidth afforded by my system at 4.5gHz.

 

GPUs are relatively the same on the outside, but the inside is vastly different from one company to another. Check your clocks on the card. A 2GB card isn't just a 2GB card. It has a RAM clock and a GPU clock. When one clock has 900mHz clock and another has a 1200mHz clock, there will be significantly more capability in processing power of the 1200mHz clock as it is working 130% of the 900mHz card at full load. The RAM clock tells of how big the pipeline to the memory in and out (kinda like a superhighway, is it 4 lanes both directions or 8 lanes - obviously at rushour, the 8 lane highway can move more traffic in the same amount of time). The GPU clock tells you how fast the actual GPU can calculate the mesh and textures and create the world you see. You may also see a Fill Rate or Gigaflops per second if you research the card. These are good indicators of how the card can perform, but not a great indicator for how the card will do with gaming, just benchmarking. Some cards do great at the benchmarks, but miserably at the gaming table, kinda like some people can pass any written test in the world, but applying that knowledge to a real-world scenario is lost on them.

 

The last bit of advice is that 6-12 months a new line of cards comes out. Usually, the beginning of the year is the most exciting. Don't rush and buy anything til the gaming market has put up some results. They can speculate all they want, but once you drop cold cash down on something that should perform great, but doesn't you'll feel cheated. I tracked and debated for 4 months before choosing my current setup and am very happy with the cost to performance ratio. My wife didn't kill me, I have more money to spend on addons, bills and golf and yet still have an amazing system that doesn't seem to be held back by any game. Simulators, on the other hand, will never be 100% as they're trying to render the whole world instead of a small sandbox that is artfully crafted for looks AND performance at every angle. Simulators don't have that luxury of everything being fine-tuned.

Thanks all for your insights and advice.  I'm sittin' here have just TO from KBOS to CYVR (Orbx PNW scenery) in the lovely albeit simple QW757 (love this plane's incredible climb power), REX Essentials Plus, FSPax, etc, on 5y/o hardware and it's beautiful!  No, can't do too much w/ the NGX around dense terminals so I default to the 757 for now.  Because I still get so much out of this sim w/o a high end box I'm OK waiting to see what Haswell will bring.   I'm also waiting to see if P3D V2 will be the place to be over XP64bit so I am imagining at this point I probably don't need 6 cores, though I'd sure love to have it.  Sounds like in some areas w/ XPlane 64 it does matter, though not as much as I thought.  If Haswell turns out to be a great CPU I will have to decide if it matters enough to wait til a 6 core variant happens but geez that's not til late 2014 I think.  Right now I'm planning on GTX Titan 6gb of GDDR-5 over a 384-bit bus:  that's some bandwidth!

Just make sure the motherboard you choose can handle that bandwidth from the PCIx16. There are a few cheepos out there that have a PCIx16 slot, but the chip (usually a via or other separable chip) that controls the video side of the board is underwhelming. To be sure, a board with no on-board video capabilities should do it. Sometimes, because the Intel chips come with video capabilities built-in and the board utilizes it, you have to share memory bandwidth between the two, even if you disable the onboard video in BIOS or windows, it still has an effect on what bandwidth is available to your card. ASUS, Gigabyte and any other high-end board manufacturers should be safe. The key is DO YOUR RESEARCH on all the parts. A kick-&@($* CPU/Video Card can be hamstrung easily by a mobo that can't cut the mustard all around. It sounds like price is not the overall deciding factor for your purchases, so don't go overboard, but don't go inexpensive either.

 

 

There are a few cheepos out there that have a PCIx16 slot, but the chip (usually a via or other separable chip) that controls the video side of the board is underwhelming. To be sure, a board with no on-board video capabilities should do it

 

Thanks again.  I'm thinking any of the 'enthusiast' boards would suffice? But most importantly, how can one accurately assess the very issue you describe?  Is there a motherboard spec that cites this?  

Check out this item at NewEgg:

http://www.newegg.com/Product/Product.aspx?Item=N82E16813157288

 

Basically, look at the specs where it says 16/8/16/0 or 16/8/8/8..

This means that all in all, the memory bandwidth for all cards is 40x (16 + 8 + 8 + 8 = 16 + 16 + 8 = 32 + 8 = 40). This means that if you have 3 PCIx16 cards (for anything mind you, not just video, but other devices may one day use the 16x bus for processing data and offloading from the CPU for task-specific functions), you can run two (first and third slots) at 16x, the second at 8x and the fourth will have nothing available. Putting in a card to the 4th slot will cause the 3rd slot to downscale from 16x to 8x (splitting the bandwidth).

 

This other board is for the 1155 series intel CPU, but the same idea applies:

http://www.newegg.com/Product/Product.aspx?Item=N82E16813157330

 

This one has 2 PCIx16 slots, but only one can be used if you're going to run either at full capability. Putting anything in the second slot will yield a limit to performance. If you were wanting to run an SLI system (which is pretty much obsolete except for multi-monitor outputs beyond what one card can pump out) and allow both cards the same full 16x bandwidth, the second board wouldn't allow it. This was the case with a system I built before this one and had selected a board just because it had a nVidia NF200 NorthBridge instead of using the i3/i5/i7 integrated video NorthBridge and all the intel NorthBridges were 16/8/1 or something like that and the NF200 was 16x/16x/1x or 16x/8x/8x depending on what you needed. Note that for an SLI config, all 3 cards will balance with the bandwidth supplied to the lowest card, so if one is getting 1x, all 3 will get 1x. 8x effectively cuts the bandwidth from CPU buffer to the GPU memory buffer in half and 16x is full steam ahead, but getting enough 16x slots on your board may be a bit tough to find the right combination depending on how many monitors and cards you want in the tower.

Thanks Aaron, much appreciated.

Indeed--thanks for the detailed overview.   Do you know what chipset will be used for Haswell?  I know the onboard graphics are a big part of Haswell, but that doesn't mean it won't be a good CPU performer, though it would be better for Intel to offer Haswell w/o those units dedicated to GPU function.  Maybe that's Haswell-E.