Digg! View blog authority View Philip Taylor's profile on LinkedIn Technorati Profile
Welcome to MSDN Blogs Sign in | Join | Help

PTaylor's WebLog

Flight Sim discussions; mostly on the core platform ( graphics and terrain, platform team, sim engine, geo tools and geo data, international, internal tools, technical art ) plus news on releases and updates, and the occasional tweaks and tips.
Holy FPS, Batman!

Toms posted a OC CPU-GPU review on May 15th that included FSX-SP2 and shows a whopping 83.1 FPS at 19x12 with no AA,trilinear and Ultra Quality, and a still excellent 72.7 FPS at 19x12 with AA, Anisotropic and Ultra Quality.

Wow, we are getting close to the timeframe where FSX starts to shine! Some key notes from the article:

"The results should clarify how much performance is obtained by upgrading the various components, as well as which combinations work best or are cheapest. It will also show us how high the basic performance level of the processor needs to be in order for the new G92 graphics chips from Nvidia to develop their full 3D speed potential."

"The test platform used is based on an X38 chipset with DDR3 memory and PCI Express 2.0 interface, and remains identical for all of the individual tests. Generally speaking, only the graphics and CPU performance is changed for each run. In order to ensure that the E2160 and Q6600 are able to keep up with the other processors, two additional test runs have been performed in which the Front Side Bus (FSB) was overclocked by 33-34%. The test results indicate whether simply overclocking the small cache budget CPU is able to compensate for its initial performance deficit, and how much performance the quad core is able to pull from its reserves as a result of the increased frequencies."

Key results:

With no AA and Trilinear: the 8800 GT 512 OC and the Q6600 OC won out regardless of screen resolution.

With AA and Anisotropic: at 1280x1024 the 8800 GTS 512 OC won out, at 1650x 1050 the 9800 GTX 512 won out, at 1920x1200 the 9600 1024 OC won out.

From my read, at 1920 with AA the combination of the extra memory and the OC is what gives the 9600 the edge, although elsewhere Toms' does state that the 9600 should be called the 9700 ( page 3 ) so that also explains how it wins out as it isn't a typical x600 part.

This article should really help the community understand the performance envelope and dispel the myth that good performance cannot be obtained today, as well as help people mix and match components. See it here.

Note that this comparo did not include the X48 mobo which enables 1600Mhz FSB support with no OC. See info about that here. While that may be of marginal benefit unless folk find their X38s' unstable under the OC to 1600, the X58 with Nehalem having the memory controller integrated into the CPU  and due this fall looks much more interesting, see it here.

Could this fall finally be the ultimate performance break out time for FSX? Let's hope so.

 

Posted: Wednesday, May 21, 2008 11:53 AM by Phil Taylor
Filed under:

Comments

Jera said:

Phil, you should know that these results are obtained on  Sitka approach, and there are almost none autogen objects around. Try with same settings to fly to any large airport in urban area and you will get around 7 fps !!! I also have Q6600@3.6GHz and 8800GT/512M and

I am absolutely sure that we will NEVER get playable performance from FSX (on max settings) if processors don't get close to 5GHz or more, and it looks like we will only get more and more cores that FSX cannot use. Right now during flight it cannot even fully use 2 cores. So I really hope that for next version of FS you will use new engine that will be multi-core oriented and only then we can hope for new breakthrough in flight sim.

# May 21, 2008 9:24 PM

Phil Taylor said:

Jera:

I believe, any imperfections aside, that article shows:

1)where 8800, 9600,and 9800 stand relative to each other.

2)where CPU variants stand relative to each other.

3)what an OC on both CPU and GPU can get you.

and not everyone knows 3, much les 1 or 2. It is in that spirit I posted. Sorry if that is objectionable.

# May 21, 2008 9:30 PM

NickN said:

Jera

No Worries.. the tech is coming. I don’t think its any secret FSX is CPU bound however the main problem right now with Intel is the memory bandwidth is hindered by the FSB. Intel's Northbridge is the problem and by moving it to the CPU all that data/bandwidth FSX is currently trying to get through is bottlenecked.

With X58, (actually the second iteration of Nehalem IMHO will really smoke) will not require the 5GHz you are referring to. At that point you can expect a very significant increase which when paired with the right VC will be the point where FSX will come into its time.

Regardless of the design, which is over and done with.. as I see it the hardware to do the job as people expect will start arriving at the end of this year.  It does mean a move to DDR3 which I have been on since last year, but none the less it won’t require a 5GHz CPU because based on current information were probably looking at a solid 40% increase with just the first iteration of Nehalem when it hits the market.

FS9 took almost 2 years to have the resources available to it for flat out running.. FSX seems to be on about the same schedule.

# May 21, 2008 10:40 PM

runwaypimp said:

I get 100+ FPS around Friday Island on my 3.8GHzOC & 9800GTX. However, fly over NYC with water reflections & Bloom & I get single digit FPS. Even without the eye candy, a complex ac can drag it down into the teens over modest locations.

Loads of work left to do Phil. FSXI needs a good 300% speed boost. Don't just hope Intel does your work for you. Trashing the obsolete 10 year old terrain engine will be an awesome start.

Please please please re-write the engine. Another re badged madden style upgrade will disappoint.

# May 21, 2008 11:36 PM

N562AC said:

I think the people at Orbyx / FTX have the right idea, in that to my uneducated eye, they seem to have redesigned the FSX scenery platform somehow to the point where performance in a complex jet and into crowded urban areas like Sydney FAR outshines anything I get in non-FTX areas, specifically around urban USA airports with comparable AI traffic loads.

# May 21, 2008 11:50 PM

Phil Taylor said:

N5:

OrbX simply works within the SDK and the existing landclass system, and exhibits one of the deeper sets of knowledge of how that system works including replacing LCLookup.BGL

# May 21, 2008 11:59 PM

NickN said:

Although I completely understand the desire to change default files such as lclookup.bgl and others (and why), in doing so the way that addon works the downfall is the user must stop/save/reboot the sim to continue a flight when leaving a covered area.

Regardless of a desire to make a product better in my opinion it should never interfere with a flight in such a way the person must completely exit the sim. I can understand a scenery database rebuild and that can be done in-game without rebooting however I think it very important addon dev’s work with the sim and its design to make sure their product does not interfere with a persons experience or change how another addon may work (correctly with the sim) in lieu of their product.

I think devs have a responsibility to the communality to try and stay with the program and that should override desires to make their product dominate.

Its my understanding that Orbix is finally looking into the possibility of doing away with the on/off mode issue and that in my opinion is the right way to develop. I do hope they follow through on that.

# May 22, 2008 2:10 PM
Anonymous comments are disabled
Page view tracker