WTF ... IS WTF!?
We are a collective of people who believe in freedom of speech, the rights of individuals, and free pancakes! We share our lives, struggles, frustrations, successes, joys, and prescribe to our own special brand of humor and insanity. If you are looking for a great place to hang out, make new friends, find new nemeses, and just be yourself, WTF.com is your new home.

Kick ass video card.

Jung

???
Premium
13,980
1,398
487
#2
I was just about to post that. :mad: It's jsut crappy SLI though.
 

jamesp

In Memory...
1,714
1
0
#3
:lol:

Yeah, but its also cheaper than SLI.
 

Jung

???
Premium
13,980
1,398
487
#4
I've yet to find a game that I can't play at 1680x1050 on the highest settings with a single 7800GTX, I see no point in having a second one.
 

jamesp

In Memory...
1,714
1
0
#5
That is a rediculous resolution. I only go to 1200 x whatever in any game, often much less. But whatever floats your boat. I cant even imagine playing HL2 at that.
 

jamesp

In Memory...
1,714
1
0
#7
Wow, those HL screenshots are pretty sweet. I think Im gonna go play it in a minute. Then maybe some Counter-Strike Source.
 

tw33k

Theoretical Realist
146
0
0
#8
Correct me if I'm wrong, but even if you have all that processing power, it goes into 1 slot and isn't that going to be a huge bottleneck? It's like those 512 MB cards that have to force their way through the 256 MB pipeline. I thought that was the purpose of SLI, doubling up the pipeline. Or am I missing something?
 

Jung

???
Premium
13,980
1,398
487
#9
tw33k said:
Correct me if I'm wrong, but even if you have all that processing power, it goes into 1 slot and isn't that going to be a huge bottleneck?
PCI-Express is capable of 4GB/s throughput each way, so I seriously doubt that card, or any other for a long time to come, will come anywhere near saturation of the bus. In fact, most SLI motherboards split a single 16x bus into two 8x buses. Having both chips on the same card would actually increase performance by reducing latency.
It's like those 512 MB cards that have to force their way through the 256 MB pipeline.
What?
I thought that was the purpose of SLI, doubling up the pipeline. Or am I missing something?
You seem to be missing a lot. SLI alternates each line between cards, so it's sort of load balancing the rendering between two cards. For example, card 1 would render lines 1 and 3 and so on, card 2 would render lines 2 and 4 and so on. The point wasn't to overcome any PCI-E limitations though; it was theoretically supposed to double performance, but never really came close.
 

tw33k

Theoretical Realist
146
0
0
#10
I didn't mean the throughput of PCIe bus I meant the ouptput of the card to the bus. If a VC has 512MB of RAM and a 256MB memory interface, it won't outperform a card with 256MB RAM and the same 256MB interface. So basically 2 VCs on one board still have to push the output through a 256MB interface, right? I realize I didn't state my question properly saying "pipeline". You seem to know your shit so I appreciate you enlightening me.
 

BRiT

CRaZY
Founder
11,662
2,402
487
#11
tw33k said:
I didn't mean the throughput of PCIe bus I meant the ouptput of the card to the bus. If a VC has 512MB of RAM and a 256MB memory interface, it won't outperform a card with 256MB RAM and the same 256MB interface. So basically 2 VCs on one board still have to push the output through a 256MB interface, right? I realize I didn't state my question properly saying "pipeline". You seem to know your shit so I appreciate you enlightening me.
Sorry, You are greatly confused and you fail to understand how video cards work.

There is no such thing as "256MB interface". There is such a thing as a "256bit memory interface". That describes how wide the bus is. The 512MB describes how large the storage is.

If it were described as a highway between two points, a source and destination ... The PC would be the source location, the video memory would be the destination, the 256bit would be how many lanes, the PCI-Express 4Gigs/sec bus would be the speed limit, and the 512MB would be how many cars could fit at the destination.

And for your statement of a 512M card with a 256bit bus not being able to outperform a 256M card with 256bit bus is wrong. More video memory helps out because you can keep more useful information, such as textures and geometry data, in the video card. This eliminates the PC having to keep resend the same data to the card. Also the speed of the local video memory (32G/s - 48G/s) is far in excess the speed of the transfer connection (4G/s PCI-Express).

Another parallel would be in general retrieving info from PC main memory versus having to retrieve it from hard-drive. The hard drive is significantly slower than PC main memory. The local Video memory is significantly faster than the PC main memory.

I hope this has helped some...
 

UberSkippy

a.k.a. FuckTheBullShit
7,529
28
142
#12
junglizm said:
What? You seem to be missing a lot. SLI alternates each line between cards, so it's sort of load balancing the rendering between two cards. For example, card 1 would render lines 1 and 3 and so on, card 2 would render lines 2 and 4 and so on. The point wasn't to overcome any PCI-E limitations though; it was theoretically supposed to double performance, but never really came close.
Actually that's not technically accurate. SLI used to be Scan Line Interleave in which card 1 was responsible for lines 1 3 5... and card two did the rest. This was the case with the voodoo2 cards that you could run two at a time.

SLI (Scalable Link Interface) today alternates frames instead of lines.

Clickity

theregister.co.uk said:
The third and final method is alternate-frame rendering, which just like SLi allows each card to create alternate frames ' for this method, just like SLi, CrossFire needs a game profile.
NOTE: This is actually a review of ATI's crossfire technology but they do mention how SLI works in the last paragraph on the first page.
 

BRiT

CRaZY
Founder
11,662
2,402
487
#13
Even back in the old days of Voodoo5, it's SLI mode worked in bands of scan-lines, where one chip rendered lines 1-127 and the second chip rendered lines 128-255. That improved texture cache performance somewhat.

Actually, today's SLI has different modes.

One SLI mode is Scissor mode, where one card does the top portion of a screen and the second card does the lower portion of the screen. This accelerates texture and fill-rate, but does not accelerate geometry processing. This is the typical mode SLI runs in.

Another SLI mode is Alternate Frame Rendering, where each card takes turns rendering a full and complete frame. This accelerates geometry processing, but adds latency to the display.
 

UberSkippy

a.k.a. FuckTheBullShit
7,529
28
142
#14
And Crossfire adds a third mode to the mix. Tiling. Cut the screan up into tiles and alternate which tile is rendered on which card. Kinda like mixing the two and probably adding more latency.
 

BRiT

CRaZY
Founder
11,662
2,402
487
#15
UberSkippy said:
And Crossfire adds a third mode to the mix. Tiling. Cut the screan up into tiles and alternate which tile is rendered on which card. Kinda like mixing the two and probably adding more latency.
Tiling on Crossfire does NOT add any latency (the time between when the screen is drawn to when it's actually displayed).
 

Jung

???
Premium
13,980
1,398
487
#16
Actually that's not technically accurate.
My bad, sorry for the confusion. I knew SLI did something similar to splitting up the scan lines, I guess I read it wrong way-back-when.
 

tw33k

Theoretical Realist
146
0
0
#17
OK, I dug out the article that I read and was (fuzzily) recalling. Maximum PC August 2005 reviewed the 512MB Radeon X800 XL (Sapphire Hybrid X800 XL) and the 512MB GeForce 6800 Ultra (XFX GeForce 6800 Ultra 512MB). Once agian I was wrong about my nomenclature, as they discuss frame buffers. Quote: More is always better, right? If 256MB of graphics memory is fabulous, 512MB must be sublime. Well, when it comes to the waning days of the current generation of videocards, our testing shows a bigger frame buffer is overkill. Nearly all games today expect 256MB of RAM at most, so there's little performance benefit to be had from doubling the memory. Both of the 512MB cards reviewed here performed on par with their respective 256MB counterparts. End Quote. So I gather from that is most games now can't take advantage of the larger RAM, but that is a programming limitation, not a result of the interface size. Thanx your input guys.
 

BRiT

CRaZY
Founder
11,662
2,402
487
#18
That's review is dated... Actually, even reviews a week or two ago are dated. That's the wonders of technology. Have a look at some of the online reviews of the X1800XT 512M card. In some situations with extreme settings, higher resolution with AA and AF, the games will easily exceed 384M of texture usage. I believe F.E.A.R might be one of them. I'll dig up the specific reviews I'm thinking of in a bit.

EDIT: Found the review spot I was thinking of. It's in Hexus.net's review of the ATI X1800XT video card with F.E.A.R.

Hexus.net said:
At 1600x1200, F.E.A.R. is nearly twice as fast on X1800 XT than GeForce 7800 GTX is, without any of the usual IQ enhancing options enabled in the game.

I repeatedly reran the runthrough multiple times to check its validity, on both sets of hardware. The reasons for F.E.A.R. doing so well on the new ATI hardware are likely shader efficiency and framebuffer space related, the 512MiB memory size on ATI's hardware allowing it to do very well.


 

Descent

Hella Constipated
7,686
109
157
#19
My God. BRiT, that's fucking amazing.

It's like the 9700 Pro in that respect.

I read a review of it on Tom's Retardware this morning. I didn't read their bullshit. Just looked at the block layouts and shit.

Impressive shit.

Looking at these performance benchmarks, when is Tom's going to stop fanboying like they are in the devout cult of nVidia/AMD?

And I'm with Junglizm on this - I'm more of a fan of ATi/AMD, but seriously, they need to stop their bullshit, they use their reputation to saunter this crap to newbies.

That content has no place on a review site.