Sli

rolf

I was just looking around and found some things about SLI, and thought that I should share it with you (how nice of me).
For those who dont know SLI is a technology that lets you run 2 Nvidia graphical adapters simultaneously, the 2 adapters working together to improve the performance of video rendering.
The system has got some interesting aspects. But first, if you want to go SLI, you'll need an SLI compatible board. Of course you'll need a good CPU because it doesnt make sense to have a powerful graphical adapter with a weak CPU! AMD is always better for games - so do yourself a favour ;) Then you'll need some SLI graphic adapters. I'm not sure, about the 5 series, but the geforce 6xxx seem to be SLI compatible. You can buy one then buy the second later.
Here's a great deal I stumbled upon:
http://www.tigerdirect.com/applications/SearchTools/item-details.asp?EdpNo=1726995&Sku=P450-7814%202
So it seems there are "SLI compatible" power supply out there. I dont know if it's required, but you'll need a fat power supply to run 2 video adapter, consider 500w and more.

Here's for the performance part:
http://hardware.gamershell.com/articles/xfx_7800_gtx_sli_performance/3.html

As you can see, the SLI performance gain is the most noticeable at higher resolutions such as 1600x1200 , where the gain can reach 80%, making it interesting. So you dont want to build and SLI system without having a big wide (and expensive :D ) display.

All in all this technology is quite interesting, but only if you can afford the rest of the computer ;).
But wouldn't you like such a thing in your PC?
http://www.tomshardware.com/2005/12/14/sneak_preview_of_the_nvidia_quad_gpu_setup/

In the above article, they have used SLI to combine 2 ASUS 7800GT Dual GPU boards, wich makes a total of FOUR 7800GT GPUs working together... :/

mahdoum

crazy :P i saw something similar on pimp my pc

karim-soubra

how about running 4 graphics card on sli using 2 asus 7800gt top
look also for the new pci express 32x motherboard such as the asusa8n32-sli deluxe

LebaneseChiphead

Hi Karim , Does it matter how many graphic cards you are using?, would the graphic bus width on the motherboard cause a bottle neck? I mean if you are transfering data via 32 bit graphic bus, you will always transfer the same amount of Data whether you 1 or 2 card etc. Don't you think it better to a load the card with a lot of Graphic memory say like 1 GB of video memory, Appreciate your feedback
Chips

karim-soubra

Sure there will be bottlenecking on the CPU level even with the dual core and 64 bit
This will be corrected as new cpus appear mainly after 2 years with new Architecture and chip wafers
Meanwhile your best bet would be to overclock the fsb and decrease the multiplier of the CPU and overvolting it
and if your ram are compliant play around the timing
As for the 1 GB graphic card I think it will be useful in the coming 2 years as it appears 512mb graphic is a bit of an overkill it is only useful for higher resolutions around 1600*1200 or hdr
So only if you have a monitor or a high def TV surpassing the 1280*1024 resolution don't bother getting more than 256mb ram on your card (bottlenecking aside)
Also ram is not the only Characteristic for a graphic card you have to look at the pixel pipes vertex pipes and memory and core frequency not to forget directx support (sm3)
as for sli configuration well it depends on the cards for example if you put 2 6600gt in sli it wouldn't an overkill even with monitors using sub 1280*1024 you will benefit from the af and the aa resulting in a better looking graphics and higher fps
As for using 2 7800 gtx 512 MB in sli you will only see a difference when using 1600*1200 and higher resolutions
if you have any questions don't hesitate to ask

LebaneseChiphead

Okay Karim, I agree that RAM is not the only player in your garphic card performance, I agree with overclocking the Front Side bus FSB it always help. but Please allow me to differ about "512mb graphic is a bit of an overkill" . I thought SLI technology was presented few years ago by DFX then after Nivindia took over they are bring it back to the market, I did not mean to diverte the discussion here, but my intent was , Was it worth it to go through all of this: Buying a mother board that can handle SLI, then buying Two Video card, because it is usless if you do not have two cards, your SCAN LINE INTERLEAVE would not work " one card Reading Line 0 meanwhile you need another card to read line 1 that is why it is very fast . For gamers it is wonderfull, but you will only see a difference in the 3D application , others will be minimum, So that lead me to the question, is it worth it for that little gain :)
Chips

karim-soubra

it depends if you are a heavy gamers that demands that everything is maxed out or you wont play the game
the gain is not so little it ranges from 50-70% depending on the componants and the configuration you are using
again you wont notice the difference unless you have high resolution monitor or hd tv
so it is worth a sli configuration for 20'' and higher monitors or high def tv
but if you have a 19'' monitor (or less) even with a sli configuration you wont benefit from it even if you max out everything
sli=Scalable Link Interface
interleaving is one of four way that nvidia uses sli to process the image
ati crossfire uses same principals but in a different way

darkstar

A 7800GTX 512Mb at very high res 2048x1536 with 4xAA performs identically to a 256Mb version. So adding the extra memory will not boost gaming performance as it is "today" but for future games the difference will show. It all comes down to the core and memory clock. Comparing the 7800GTX which runs at a core clock of 550Mhz and 7800GT which runs at 430Mhz we can see that the 7800GTX clearly outperforms(over 40%) the 7800GT in any game and at very high resolutions.

Also we do realise that most games today barely support SLI so we will see a SLI Boost in upcoming games.

LebaneseChiphead

Thank U for the answers Karim and Darkstar, I am getting really old Karim Forgive me I should have checked on the "SLI" acrom before I typed so fast.
Darkstar: you hit the nail on the head, my whole discussion here was to get to this statment :
""It all comes down to the core and memory clock"" That is why I ask about the bus Limitation or bottleneck in the first Email. But you are 100% correct.
Things will change you will see a lot faster clocks soon, and by the way clocks are a nightmare in my Job, if you do not have a clock nothing will fire up, neither the chipset otr the mother boards " you will never come out of Reset mode " better have a good clock generating moduler and a good clock tree, and watch out when your clock cross different frequency domains. anyway, Thank you Karim and Darkstar, it is a pleasure to continue discussions with you.
Chips

karim-soubra

you're most welcomed

LebaneseChiphead

Darkstar, Please correct me if I am wrong, The more Memory you have the better and faster graphic you will have, as you know the video card will use the VRAM on it as a buffer where he will keep the most used file in this buffer, so it is really acting as a cache for the video card, so the bigger the amount of the cache makes a bigger buffer, and that will save the video card from running all the way to the system memory to get the files that he needs to process. as you know games that do a lot of texturing and details requires a lot bigger VRAM cache on the card.
Sabaho Karim, I see you R on line so I thought I drop a hello to you too and I appreciate your feedback on this issue too.
Chips

darkstar

Video RAM (VRAM) is a buffer between the computer processor and the display and is often called the frame buffer.VRAM is dual ported which means that the cpu can write to the vram and the gpu can read from it at the same time with no delay(Which greatly differs from how RAM works). When images are to be sent to the display, they are first read by the processor as data from RAM and then written to video RAM. From video RAM (the frame buffer), the data is converted by a RAM digital-to-analog converter (RAMDAC) into analog signals that are sent to the display presentation mechanism such as a cathode ray tube (CRT).Some cards have multiple RAMDACs, which can improve performance and support more than one monitor.

VRAM is expensive (VRAM is more complex and requires more silicon per bit than standard DRAM, which makes it cost more)so u cannot find huge quantities of it but instead now manifacurers use SGRAM (synchronous graphics RAM) instead.Its performance is nearly the same, but SGRAM is cheaper. Concerning performance well yes the more vram it holds the more completed images it can carry at a time to display them. Too much VRAM (which is unlikely) can definetly cause a bottleneck at the GPU.To avoid confusion, VRAM makes up only a 'portion' of the RAM (ie:ram present on the vga), the rest of the memory is used to store the neccessary pixels and images.

Hope this helps and you are most welcomed.

ulias

in my opinion nVidia and ATI should creat craphic card same as todays mainboards u can upgrade the processor and the memry on it, but this will only be acheived in the futur, but as for now i have worked on the MSI 7800GTX 256 GDDR3 and well, in
SLI mode if ur not using your computer for rendering and graphical aplications beter than games then u dont need it, overpriced @ 680usd (vat not included) its a waste of money,a 6800gt will be more than enaugh
, plus one problem is known and thats the cooling system is not enaug! running @ 450mhz the msi 7800 gtx cant be overclocked unless in a freezer!!!!!! ....... the next step to get the more out of your hardware is too get rid of the slow softwares,and by installing the windows xp 64bit edition ull be doing this, upgrade ur rams to 8gb and this will be a hell of a pc , i tryed it ;) : pentium D3.2ghz 2x1mb MSI 7800gtx 256 Gddr3 MSi P4N-sli 4gb Kingstone
the only limitation was the operating system .

LebaneseChiphead

Hi Ulais, I am just curious, You mentioned that you worked on the video card with GDDR3, what brand name are these GDDR3 Dimms that were used ? what did you do on that card ? Any idea about the transfer rate and what kind speed that bus have?
Chips

karim-soubra

theses are not dimms but the type of rams on the graphic card
the speeds vary from card to card
but mainly memory clock is around 1200 MHz (for a 7800gtx 256MB)
and an interface of 256 bit ramdac is 400MHz

LebaneseChiphead

Thank you Karim , I called them Dimms, they are a group of Chips that populate that card, If all these chips are the same speed "let us say 7 nano seconds speed" and the bus interface is the same and main memory clock is the same on all cards from that manufacturere, why would be a difference between cards as you stated? Is there a design difference any where between the cards . if not please elborate on why the speeds on the cards are different?
Thank U
Chips IMR

karim-soubra

Some have better cooling leading to a better overclocking
These days the overclocking is done on the manufacturer level rather than the user's which in it turn will reduce the number of voided guaranty leading to less trouble for the manufacturer
As for the design I don't think there is much difference between the cards as the core is provided by nvidia and the ram mainly by samsung
So it all comes to how much power you can squeeze out of the card resulting in different core and memory clock speeds