Haswell o'clock to 7.1Ghz...

[Noel 4,488]

foibles, on 06 May 2013 - 4:12 PM, said:

I've been saying that for 5 years. Any day now ...

I rest my case. It sure seems odd because from what I gather XPlane 10.10 is very decent in many ways, sans those two biggish areas. I say biggish because I'm not sure why either a 3rd party vendor or LR doesn't just address airports & their buildings the way FSX did from the beginning. This doesn't sound like a huge design nightmare, more of a mundane & tedious task done airport by airport. I think this likely could jettison XPlane to a considerably larger user base, and in turn would likely spawn more buy in from other 3rd party developers as the user base grows. Heck, it sounds like many airports have already been done by various hobby or commercial developers, so the task then is to compile them all into one installation, or better yet sell it to LR as a joint venture.

[haztronic 1]

What kind of fps could u get at 7.1ghz in fsx say in the ngx?

[gizmosellsbunnys 106]

See, better single thread performance not so much cpu speed increases so much anymore. Aiports in X-plane will start to fill slowly not like Xplane has the team to build airports all day. Just hope they fix the autogen misplacement with better data information. X-plane not does overreach like fsx with software tech so much does. Xplane like open frontier

[Manny 1,293]

FSX without DX10 would still give us OOM error even with 10Ghz system.

 

Tsk Tsk! Sorry..couldn't resisit!

 

 

I managed to get OOM even with DX10. I flew from KLAS to KLAX and right when I was over the LA City...Bam..OOM! Sigh!

[Simmer2050 28]

FSX without DX10 would still give us OOM error even with 10Ghz system.

 

Tsk Tsk! Sorry..couldn't resisit!

 

 

I managed to get OOM even with DX10. I flew from KLAS to KLAX and right when I was over the LA City...Bam..OOM! Sigh!

I have found up'ing MOBO north bridge IO  voltages helps decrease OOM, but north bridge gets hotter too, so got to keep an eye on temps as you tune

 

I got good air cooling so works ok.

[Howellerman 104]

I tend to agree w/ that statement with the caveat that you're talking mainly about clock speed, which is why I'm intrigued by a new simulator (or improved XPlane 64) that can utilized more parallelism than FSX does.  I'd rather have a 16 core, duel CPU box that doesn't overclock and run software designed to exploit it maximally, than have a 4 core machine that has to have exotic cooling and overvolting to run well using FSX.

Parallelism is indeed the direction things have been going.

 

Intel is to be commended for providing 5+ generations (Pentium, Netburst, Core2, Nehalem, and most recently Sandy Bridge) of new architectures while maintaining the same thermal design point (TDP). On the server side, where I live and breath, this is 60W, 90W,and 130W of thermal loads. This consistency, despite going from single core (Pentium) to 8-core (Sandy Bridge) and now 12 core (Ivy Bridge), is remarkable and allows server vendors to maintain air cooling up to 95F ambient environments.

 

Remember, these are SERVER numbers, not PC numbers, where multiple cores have long been acknowledged as good, and have allowed Moore's Law to continue to progress with transistor density, which is basically the combination of cores and cache. The upcoming (again on the server side) jump from Sandy Bridge (the architecture, the 'toc' in Intel design cadence) to Ivy Bridge (the feature shrink, or the 'tic') is taking an architecture implemented on 28 nm technology and shrinking it to 22 nm technology, increasing the core counts (from 8 to 12) and the "last level cache" size (from 24MB to 30, perhaps larger).

 

I have a number of friends at Intel who have told me "yeah, we could have clocked NetBurst to 5+ GHz, but it would have required liquid cooling." In a server environment, this is not a good alternative, where hot-aisle/cold-aisle airflow is the dominant cooling model. 

 

Moving forward, I think that Intel will maintain their existing TDP, continue to increase the core density, and make multiprocessor design disciplines a moot point. I come from the age of absolutely EXQUISITE engineering required  to create multiprocessor systems (mainframes on down), and now? All on a single piece of silicon. 

 

So, parallelism is the new goal. That being said, there are very few applications that scale well with increased parallelism (HPC being an notable exception, but that is a very different discipline). Not to get too technical, there is a multi-processor curve that most applications hit and when it hits, gains in performance can turn to losses with increased core counts. That is why (again, on the server side) virtual machine technology is so pervasive: chop up all those big symmetrical multi-processing systems into discrete 1-2 core virtual machines where things run more efficiently.

 

When I record what FSX is doing on my system I see the primary CPU (1, in my case, as I have affinity mask set to 14) totally maxed, while the other two cores are dipping up and down in cadence. The "cadence" is what I find worrisome about FSX on multiprocessor systems. Cadence typically indicates that if core 2 is "bored" it will steal work from the process queues on core 3. Core 3, being bored, will steal them back from core 2, and the lockstep goes back and forth. All the while, core 1 is doing all the heavy lifting. This lack of intrinsic parallelism in FSX is the limiting factor going forward on new Intel designs, including Haswell (the new architecture, or "toc") and beyond.

 

Anyway, I hope I haven't stripped too many syncros here, and I welcome input from the "PC" side of the house. 

 

JKH

[Noel 4,488]

 

 

Anyway, I hope I haven't stripped too many syncros here, and I welcome input from the "PC" side of the house. 
 
JKH

 

Thank you for the detailed overview of the various issues surrounding multicore/multiprocessor.  I am not much of an engineer but I sense making software to optimize server functions to take better advantage of parallelism might be easier than optimizing for something like flight simulation, but I don't know enough to be sure.  Have you tried looking to see how your non-primary cores behave using XPlane 64 bit?

[Howellerman 104]

Thank you for the detailed overview of the various issues surrounding multicore/multiprocessor.  I am not much of an engineer but I sense making software to optimize server functions to take better advantage of parallelism might be easier than optimizing for something like flight simulation, but I don't know enough to be sure.  Have you tried looking to see how your non-primary cores behave using XPlane 64 bit?

Heya Noel, 

 

   Um, no.

 

   I am afeared of trying XPlane - I finally got FSX to run right, and got the right airplanes, scenery, and utilities. I originally flew freeware aircraft, but the payware stuff is soooo nice, and made me aware of the concept of "flying an aircraft" - flight planning, starting cold and dark, and basically running a flight from takeoff to landing in real time. [Note: Mister Cattaneo, Craig Richardson, and other freeware aircraft are in a class by themselves]. And now I have UTX, and GEX, and REX, and a serious (I mean SERIOUS) addiction for Orbx stuff...  

 

   I can say that the crash analysis guys that created and perfected LSDyna (used by pretty much every auto manufacturer) and the HPC community have gotten multicore/multisystem parallel programming down cold, so maybe the folks at Prepar3D or Xplane can learn to exploit these new Haswell multicore systems (FSX, alas, is at a dead-end). The main challenge is realizing what can run in parallel (I would say weather and terrain, since that is well established in the scientific community), what is strictly scalar (the aircraft), and creating a good maitre'd to synchronize them all. 

 

   I am optimistic, and as people like Orbx and other sim vendors create Prepar3D offerings in parallel (pun totally intended), things will only get better.

 

JKH

[martin-w 3,234]

haztronic, on 07 May 2013 - 03:26 AM, said:

What kind of fps could u get at 7.1ghz in fsx say in the ngx?

If, you are at 4.5 now, and get 30 frames per second.

 

7.1 would give you 48 frames per second.