Lost Planet 2 DX11 - noticeable tessellation

[evolucion8]

If its true that Lost Planet 2 can't see the 4 remaining logical units of the i7, why then the i7 860 is faster than the i5 760 in the Lost Planet 2 benchmark? Because after all, the i7 860 and the i5 760 runs at the same 2.80GHz core clock, besides of Hyperthreading, basically both chips are the same.

Something is fishy there, for software, Hyperthreading is pretty transparent in terms of cores, which means that software doesn't need optimizations to take advantage of it, standard multithreading optimizations should be enough because software can't find the difference between logical and physical cores. And how come that the Phenom II X6 1055T have a performance advantage over the Phenom II X4 965 if the core scaling isn't there?

 

[AtenRa]

Arggg, typo

Core i7 920 has 4 Physical cores and 8 logical cores.

thx

 

[AtenRa]

If it only runs 4 threads, then it has SPECIFICALLY checked that it is:
1) A quadcore CPU
2) With HT

And concluded that it wants to run only 4 threads.

Meaning that it doesn’t utilize HT maybe????

 

[Skurge]

It doesn't matter how many cores, threads, cache, Vram, SPs the chip has, if the price is the same, than the comparison is fair to me. If AMD price a 6core the same as an intel 4core then its fair to compare them. Just like you don't compare a 5970 to a 480

 

[Skurge]

Consider also that according to a HardoCP review, the GTX460 in SLI beats the Radeon HD 5870 in CrossfireX.

This review says otherwise, it struggles to match 5850 crossfire. http://techreport.com/articles.x/19404

Sorry, don't mean to derail the thread.

Carry on.

 

[Scali]

Meaning that it doesn’t utilize HT maybe????

Technically it can't make that choice.
You select whether or not you want to use HT at the POST stage.
After that, the CPU is either in HT mode or not, and it cannot be changed while the OS is running (which also means that the resources are segmented, whether you use two threads on a physical core or not).

It might just be purposely avoiding to run more than one thread per physical core (which is probably not a good idea, but various developers I've spoken had this notion that more than one thread per physical core cannot work... which is wrong obviously, else HT wouldn't exist in the first place).

 

[AtenRa]

Consider also that according to a HardoCP review, the GTX460 in SLI beats the Radeon HD 5870 in CrossfireX.

This review says otherwise, it struggles to match 5850 crossfire. http://techreport.com/articles.x/19404

Sorry, don't mean to derail the thread.

Carry on.

HardOCP review was with a factory OC GTX460 1G at 800MHz, techreport's review is with a stock clock GTX460 1G

Sorry, don't mean to derail the thread.

 

[bunnyfubbles]

future proof is such a bullshit term, I feel sorry for anyone who buys introductory hardware specifically for its ability to support brand new technologies thinking they're going to have a viable card for years and years to come.

It was bullshit when the Radeon 8500 supported DX8.1 and the faster GF4 did not. It was bullshit when the GF6 supported SM3 and the Radeon X did not. Hell, any first gen DX part is going to be too slow for the games that eventually utilize the new version to appreciable extent.

 

[Scali]

Hell, any first gen DX part is going to be too slow for the games that eventually utilize the new version to appreciable extent.

Not necessarily.
The GeForce 8800GTX/Ultra and the Radeon 9700Pro were excellent cards with good longevity. If you have an 8800Ultra today, you'll still be doing okay in the latest games. Yea, perhaps you can't turn ALL the eyecandy up, and run super-high resolutions... but 'too slow'? Not quite.
On an 8800Ultra you can play something like Crysis pretty well.

 

[toyota]

future proof is such a bullshit term, I feel sorry for anyone who buys introductory hardware specifically for its ability to support brand new technologies thinking they're going to have a viable card for years and years to come.

It was bullshit when the Radeon 8500 supported DX8.1 and the faster GF4 did not. It was bullshit when the GF6 supported SM3 and the Radeon X did not. Hell, any first gen DX part is going to be too slow for the games that eventually utilize the new version to appreciable extent.

not always true so no need for a crazy tangent. lol. plenty of people got bit in the ass on the SM 3.0 alone.

 

[toyota]

Not necessarily.
The GeForce 8800GTX/Ultra and the Radeon 9700Pro were excellent cards with good longevity. If you have an 8800Ultra today, you'll still be doing okay in the latest games. Yea, perhaps you can't turn ALL the eyecandy up, and run super-high resolutions... but 'too slow'? Not quite.
On an 8800Ultra you can play something like Crysis pretty well.

yeah but the 8800gtx and especially ultra were extremely expensive so they better have some longevity. I skipped all the 8800gtx/ultra nonsense and got a gtx260 for much much cheaper when Nvidia had to lower the price because of the 4000 series. an $800 8800ultra looks pretty darn silly compared to a gtx260 that could be had for less than $200 just a few months later.

 

[AtenRa]

Technically it can't make that choice.
You select whether or not you want to use HT at the POST stage.
After that, the CPU is either in HT mode or not, and it cannot be changed while the OS is running (which also means that the resources are segmented, whether you use two threads on a physical core or not).

It might just be purposely avoiding to run more than one thread per physical core (which is probably not a good idea, but various developers I've spoken had this notion that more than one thread per physical core cannot work... which is wrong obviously, else HT wouldn't exist in the first place).

Let me rephrase then, perhaps it’s not HT optimized?

Technically, the HT threads, only get in the pipe-line in order to keep it from stalling thought only one thread is being executed in one cycle in a physical core, am I correct ??

 

[Attic]

It sure is playable, on both ATI and NV.

snip...

Here are my scores from the other thread:

1680x1050, Full Screen, Refresh: 59.88, VSync OFF, MSAA4x, Motion Blur: ON, Shadow Detail: High, Texture Detail: High, Rendering Level: High, DX11 Features: High.

Test A:
Lowest fps observed: 45fps
Scene1: 58.5, Scene2: 54.1, Scene3: 67.4, Overall Average: 58.1fps RANK "B"

Test B:
Lowest fps observed: 31fps, Scene1: 49.2, Overall Average: 47.2fps RANK "B"

Single GTX480 @ stock. 258.96 drivers
i7 860 @ 3.4GHz
Win7 64.
8GB DDR3

I question the argument here you make here for "playable" and the GTX480. You are considering the resolution of 1680x1050 to make your point that the GTX480 is playable,... that is a fail when the GTX480 is priced for enthusiast gamers who are looking at 1920x1080+ performance from the card.

What's the story at 1920x1080 for this game? That's where the GTX480 is going to be evaluated, not 1680x1050.

 

[Keysplayr]

I question the argument here you make here for "playable" and the GTX480. You are considering the resolution of 1680x1050 to make your point that the GTX480 is playable,... that is a fail when the GTX480 is priced for enthusiast gamers who are looking at 1920x1080+ performance from the card.

What's the story at 1920x1080 for this game? That's where the GTX480 is going to be evaluated, not 1680x1050.

So I take it you want lower numbers then? Very well, I'll hook up the 24" screen.
And just for a refresher, Edplayer posted some 1920x1080 with a 460.



Test A: Sc1 - 36.8, Sc2 - 31.5, Sc3 - 40
Rank B, 36fps


Test B: Rank B, 31.4fps, Sc1 - 32.7


Q9550 at 3400Mhz, 4GB, GTX 460/768MB at stock. DX11, no AA, everything else at high or max

That's with a 460 768MB.

And I'd like to see the rule that applies to high performance cards only running a game at 19x10 or higher.

"the GTX480 is priced for enthusiast gamers who are looking at 1920x1080+ performance from the card."

^ This is bull.

 

[Grooveriding]

I question the argument here you make here for "playable" and the GTX480. You are considering the resolution of 1680x1050 to make your point that the GTX480 is playable,... that is a fail when the GTX480 is priced for enthusiast gamers who are looking at 1920x1080+ performance from the card.

What's the story at 1920x1080 for this game? That's where the GTX480 is going to be evaluated, not 1680x1050.

You are digging yourself a hole here with this argument. Look at post #8 in this thread http://forums.anandtech.com/showthread.php?t=2098321&highlight=

Those are about the highest settings and resolution you could ask for and the game is totally playable. I'm sure if I turned on some crazy AA on it would make it unplayable.

Also to note, these are the maximum settings, not everyone needs to play their games on the highest of settings to be happy. I would expect it would be even more playable of course if a few settings were dialed down.

Expect this game to shine on nvidia hardware, there is a direct link to download the benchmark on nvidia's site.

 

[Keysplayr]

Groover, aren't you using 480 SLI in post #8 benches?

 

[Grooveriding]

Groover, aren't you using 480 SLI in post #8 benches?

Yes, those were SLI. It is playable though. With those numbers, I'm sure a single at 1920x1200 would do just fine.

I also wanted to point out that taking into account turning down one setting or another would also make the game playable.

 

[jvroig]

Let me rephrase then, perhaps it’s not HT optimized?

Technically, the HT threads, only get in the pipe-line in order to keep it from stalling thought only one thread is being executed in one cycle in a physical core, am I correct ??

An application doesn't have to be "HT optimized". Applications just spawn threads and that's it. The OS handles it, and what the OS sees are 8 cores (when HT is on) or 4 cores (when HT is off).

Additionally, modern operating systems (OS, not apps) are "HT aware", so if number of threads is less than or equal to 4 (if you haven't noticed yet, I am basing all figures on a quad core HT enabled processor like an i7 920), then the scheduler is smart enough to assign the threads in such a way that they will end up in separate "physical cores", so that no resource contention scenario of any sort will happen which may affect performance (which can be the case if 2 threads were assigned to logical cores randomly, and they end up being the 2 logical cores belonging to what amounts to one physical core).

If Lost Planet 2 is limited to 4 threads in an i7 920 with HT on, then that means the developers check the CPU and if HT is enabled, and purposely limit the threads so that it won't exceed 4 ("4" being half of the total number of logical cores available, which naturally amount to what would be the core count if HT were off), probably thinking (or maybe they tested it, who knows) that spawning more threads than that will affect performance negatively instead of help. So it is not that Lost Planet 2 is "not optimized for HT", because apps don't need to do anything special to make use of HT that they wouldn't already be doing to make use of quad, hex, or octo cores.

If Lost Planet 2 does not make use of more than 4 threads even when HT is on for an i7 920 or whatever model, then the devs did it on purpose. I cannot imagine why they would go to the trouble of doing that, except if they noticed performance suffers from HT instead of improves.

 

[thilanliyan]

"the GTX480 is priced for enthusiast gamers who are looking at 1920x1080+ performance from the card."

^ This is bull.

This is just my opinion but I agree with him. I'd say its pretty likely that the only reason you're using a 22" monitor is because it's 120Hz for 3D. If somebody is gaming in 2D at 1680 I sincerely doubt they would be looking for a GTX480.

 

[Keysplayr]

This is just my opinion but I agree with him. I'd say its pretty likely that the only reason you're using a 22" monitor is because it's 120Hz for 3D. If somebody is gaming in 2D at 1680 I sincerely doubt they would be looking for a GTX480.

I'd say it's pretty likely that the majority res out there at this point is 16x10 anyway. This game is playable at 19x10 on a stock GTX460 768 with AA off. I think a GTX465/470/480 can handle it also. But you want numbers, so I'll hook up my 24" and get those coveted numbers in the morning.

 

[thilanliyan]

I'd say it's pretty likely that the majority res out there at this point is 16x10 anyway. This game is playable at 19x10 on a stock GTX460 768 with AA off. I think a GTX465/470/480 can handle it also. But you want numbers, so I'll hook up my 24" and get those coveted numbers in the morning.

Maybe 16x10 is the majority res, but again people playing at that res are not likely looking for a 480. When people start threads in this forum asking for video card recommendations, do you really recommend a 480 when they say they play at 1680x1050? I doubt it.

If you can give us the numbers that would be great. It doesn't really help me but the more information the better.

 

[evolucion8]

An application doesn't have to be "HT optimized". Applications just spawn threads and that's it. The OS handles it, and what the OS sees are 8 cores (when HT is on) or 4 cores (when HT is off).

Additionally, modern operating systems (OS, not apps) are "HT aware", so if number of threads is less than or equal to 4 (if you haven't noticed yet, I am basing all figures on a quad core HT enabled processor like an i7 920), then the scheduler is smart enough to assign the threads in such a way that they will end up in separate "physical cores", so that no resource contention scenario of any sort will happen which may affect performance (which can be the case if 2 threads were assigned to logical cores randomly, and they end up being the 2 logical cores belonging to what amounts to one physical core).

If Lost Planet 2 is limited to 4 threads in an i7 920 with HT on, then that means the developers check the CPU and if HT is enabled, and purposely limit the threads so that it won't exceed 4 ("4" being half of the total number of logical cores available, which naturally amount to what would be the core count if HT were off), probably thinking (or maybe they tested it, who knows) that spawning more threads than that will affect performance negatively instead of help. So it is not that Lost Planet 2 is "not optimized for HT", because apps don't need to do anything special to make use of HT that they wouldn't already be doing to make use of quad, hex, or octo cores.

If Lost Planet 2 does not make use of more than 4 threads even when HT is on for an i7 920 or whatever model, then the devs did it on purpose. I cannot imagine why they would go to the trouble of doing that, except if they noticed performance suffers from HT instead of improves.

Makes a lot of sense, there's scenarios where Hyper Threading results in a performance hit, also HT is known to cause cache trashing or aka dirty cache.

 

[3DVagabond]

never mind.

 

[Scali]

Technically, the HT threads, only get in the pipe-line in order to keep it from stalling thought only one thread is being executed in one cycle in a physical core, am I correct ??

Not correct.
Don't forget that a modern x86 is EXTREMELY superscalar and out-of-order.
You can have tons of instructions 'in flight' at the same time.
It doesn't matter from which thread these instructions come.

So both threads just run through eachother pretty 'randomly', much like how the instructions in a single thread are running pretty 'randomly' through eachother because of out-of-order execution.

There is no concept of 'physical thread' and 'HT/logical' thread.
Both threads on a physical core are first-class citizens. The next instruction being run is only determined by which instructions have all their inputs ready, and which execution ports are available.
Because the decode buffer is segmented, you will always have an equal amount of instructions from both threads waiting (assuming the decoding step is complete and the buffer is full). So it's NOT AT ALL as if only the first thread runs and the second thread is only 'inserted' when the first one stalls. The dispatcher will just dispatch whichever instructions are ready for execution. It pays no attention to threads at all, so it could dispatch instructions from both threads in the same cycle.

It's amazing how after so many years of HT, people STILL can't get their heads around it. It doesn't help that a lot of FUD is being spread about it as well.
We covered it in some detail in this thread: http://forums.anandtech.com/showthread.php?t=2093274
Also contains some links to Intel's documentation on their implementation on the various architectures (HT on an Atom is quite different from a P4/Core i7).

 

[AtenRa]

Programming with Hyper-Threading Technology

Elsewhere in this book, we have presented other issues that can harm performance; for example, 64KB alias problems and false sharing. To the extent that these problems force one thread to waste execution resources, they detrimentally affect the other processor. So, in a direct sense, it’s true that you should optimize threads individually before considering their interaction on processors that use Hyper-Threading Technology. Then, once they’re optimized, try to determine if there is a pressing need for using processor affinity to schedule threads on specific physical processors. Once this is done, look for places in which the threads can hurt each other’s performance, especially in the areas of cache and bus management. If this is done correctly, you will find that Hyper-Threading Technology imposes few constraints and requirements beyond those inherent in parallel programming, and yet it delivers visibly faster performance while using a single physical processor.

Programs do need to be HT optimized, and from the table below, its clear that HT is working but it only gives a mare 10% more frames than with out it (Core i7 860 vs Core i5 760).