x9100 & gaming; P7805

Dirt 2 is a fun game though


Are you specifically referring to the 9800m or turbocache in general?

 

The 9800m, I am by no means an expert on TurboCache, HyperMemory or D3D. I do have a good basic grasp of them though.

Older cards would software emulate the functions required to read the system memory to be D3D compliant. The 9800m, amongst others, is fully compliant and has the hardware acceleration required to read this memory over the PCIe bus.

Now the full envelope of TurboCache, and I think HyperMemory, is supposedly read/write access to the assigned system ram memory blocks. I am not convinced by anything I've seen to date that the 9800m uses the full envelope. I could be wrong here or it could be that side of the envelope is just software emulated within DirectX/OpenGL.

The reason I feel that is sofware emulated is if it were not the OS could assign memory to the card and then it could be accessable by Cuda. From all I've seen Cuda does not have this access built within it. Again I could be wrong. Now this too then could be totally different for ATI.

All emails specifically stated Vista or Windows 7 as apparently XP is not capable. How the OS handles the PCIe bus is probably the reason it has no "texture memory". This may be driver resolvable, I am not sure of that though as it may be something that can't be built into the PnP layers etc. M$ I am sure is in no hurry to resolve the issues.

It leaves alot up in the air and I'll appologize I am not the one who can bring them down to the ground with an answer. I only know that I can see the 9800m does have access too and can read OS assigned system memory for texture.

 

I'm fairly certain that CUDA can use shared memory for calculations. However, Cuda is unable to render textures.

hmm, it would make sense if it was software emulated by the OS. It would explain why the 9800m itself doesn't support shared memory but can read it.

If that is the case, by the time the Shared memory was used it would be much much slower than turbocache or hypermemory. To the point where it really wouldn't be worth using.

 

It isn't specifically for using as in manipuate the textures from system ram. It primarilly caches them. If say the game has 2GB total textures it then can load them to this specific ram texture storage. Then the GPU calls them from ram as needed and loads them to dedicated texture buffer say between levels etc.

This dedicated cache is more funcitonal for a few reasons.

1.) unlike disk cache it will not flush as the OS has reserved it.
2.) The GPU over PCIe can read it harware accelerated, off loading the CPU cycles and the related memory latencies.
3.) lower disk activity lowering PCI bus issues such as latencies or restarting a sleeping drive etc.
4.) like Disk ram cache this should done with the FREE memory so applications are not affected and it is invisible to the user.
5.) in free memory it is readily available to garbage collection. Making the resources readily available when the game is off loaded.

The list goes on, it just makes sence for a high end card to use this to keep things as fluid as possible. It doesn't make any sence, at least to me, to have the GPU write the textures. They are what they are and to make the GPU load them from the CPU and then GPU then to ram is silly when you could through software emulation just put them there to begin with freeing the GPU and unneeded PCIe traffic etc.

Edit; the developers of the game though have to impliment and/or optimize for this....................

 

we have literally the EXACT same thing. but your ratings are higher.. did i do something wrong or something. .i dont get it.

 

Here Is my response from Nvidia. All 9800m don't use shared memory gtx gt gts.
According to Nvidia the 9800m can't use Windows 7's shared memory.


ReaperWolf. Your Ram latencies are slower than Tanware's. The speed of ram effects the CPU's speed.

 

oh, since he has 1333, and i have 1066 it makes it a bit faster.. ok then. thank you

 

clock speeds latencies and timings all have effect over the ram which has an effect over the CPU.

 

My ratings were with the CPU at 3.7 GHz, the CPU and ram rate higher. I could run it at 4.0 but that is too close to the edge and I do not normally use it at that speed. I do however use 3.7 so I leave it at that.

The 1333 MHz memory runs at 1066 MHz 7.7.7.24, exactly what my original 4GB ran at. This has no effect on my numbers for WEI. In fact with the 4GB I set the RAM for 1066 to 6.6.6.20 timings and it had no effect either on the numbers.

8GB increases free ram, this can have its benefits but they are minimal. Gaming CAN be smoother as it reduces hard faults in memory. I also run with the Page file disabled to avoid writes to the SSD and keep the XT out of it.

I am playing Dirt2 at 3.4 GHz, I run Ultra (DX10 does not allow all ultra like post proccesing) at 1920x1200 with 4xAA and get a minimum of 31 FPS and average 36 with vsynch on, without AA I get 35 FPS minimum and 41 Average. I get slightly higher with vsynch off but with it on there is no tear in cut sceens.

Now from what I've seen around this is well above what most get with the 9800m GTS. I can tell you it is buttery smooth and absolutely no stutter when racing...................

Edit; since some here have the proper 1440x900 screen I created a resolution profile and tested Dirt 2 on those settings, 3.4 GHz

1440x1990 vsynch on
Setting min.. - Avg
4 x AA - 38.8 - 47.2
0 x AA - 42.9 - 50.8

 

Hmm, when I upgraded the ram on my FX I remember my scores changing i think it went up 1.2 . I don't remember what the timings or anything were,it has been a long time since then.

Wei is a very inaccurate benching tool, it provides a very rough estimate.

On the machine I have now I never get the same rating twice, it ranges form 7.0- 7.3, my guess is it has something to do with twinturbo. ( Its is with a fresh install of windows with nothing but drivers installed)

 

The i7 is an entirely new world, way more flexible in the clock speeds. Without even EIST disabled etc with C2D systems they will score much more consistently. This system the ram, since going to 6MB L2 CPU's, has always mirrored the CPU WEI rating. Agreed WEI is about useless for determining a system as a benchmark.

As far as SSD you need to be sure you run at full power profiles. The ICH9 is affected with read and writes speeds by power management. Run ATTO on both settings and you will see. While I do not have ICH10 mobile my understanding is they are affected even more and there are tweaks that can be implimented to reduce the power management affect on I/O to the SSD. from what has been shown the desktop ICH10 does not suffer the issues, or at least not to the same degree.............

 

Well when you get a more reliable benchmark tool the i7's don't fluctuate as much.

Good advice on the SSD.

 

Right, I meant this as far as WEI is concerned.

 

Hi,
Correct they do not use memory as in full envelope Turbocache. With full envelope it is read/write and the card can reserve memory for all functions. This could include any functions a GPU can proccess. The 9800m just uses it as texture cache.

My question apparently got kicked to another tech as my last responce was more D3D, OpenGL and Cuda API releated.

Question;

Answer;

 

Hmm well it seems we are both right and wrong in a since. Do you have a screenshot of the convo?

 

I am not sure if that would violate ToS, so I coppied and pasted to a PM to you.

Edit, if you need pictures I'd have to do so in at least two or more, way to long for just one...........