JUST TWO WEEKS prior to the expected launch of the Evergreen ATI RV870 series - otherwise known as the HD 5870 - it's clear that the new cards' performance may leave the current crop in the dust.
So, for those who bought the HD 4890 and don't exactly want to replace them yet with the new stuff, is adding more HD 4890 cards worthwhile now they're at fire-sale prices?

Here we test the double and triple-CrossFire - called TriFire - HD 4890 setup on an Asus Z7S-WS mainboard. It's a dual Xeon mainboard in ATX format which is overclockable somewhat.
The dual X5482 quad core Xeons (same speed and cores as QX9770, but two of them in total for eight cores altogether) are overclocked to FSB 1800 and 3.6GHz speed, with 8GB of quad-channel FBD-900 CL5 memory. Too bad that Asus killed off its workstation division - sorry, merged it into servers - and never made a successor with two Nehalem Xeons and dual GPUs in ATX format.
There are also two independent PCIe x16 v2 slots, as well as one PCIe x16 physically (but x4 only electrical signaling) slot for the third card. That means, the third card would be somewhat impaired in terms of system throughput.
So, we put together three Asus HD 4890 cards, all running at 900MHz GPU speed, and risked a house fire to check the 3DMark Vantage scaling between dual-GPU and tri-GPU setups. All tests were done using 64-bit Windows Vista Ultimate and the new Catalyst 9.8 drivers, running off the A-Data XPG 64GB SSD drive and on a 26-inch 1920x1200 display.
Since multi-GPU users would surely use top notch displays, I ran the Extreme benchmark mode only at the native 1920x1200 resolution.
Here are the dual GPU results...

... and the triple GPU results - notice, the CPU results are very close to the dual W5590 Nehalem Xeon we tested earlier.

Yes, it is far less than the 50-80-plus per cent hike you get out of going from one to two GPUs, but even getting 15 per cent more on a bus-constrained third card was beyond my expectation.
If instead you use a board like the Asus P6T7WS with four full x16 width slots, even a Quadfire 4890 setup may scale well, at least in benchmarks.
In summary, yes, it's worth it for those with a suitable mobo and who insist on sticking with the current GPU. µ
The Asus specs page for P6T7 WS lists only SLI, not Crossfire.
Sure is one heck of a board, but Pricewatch has it at $456!
What in the world are you smoking, dear author? Even at "fire-sale" prices, buying a 3rd graphics card to get a measly extra 15% of performance is utter stupidity! How can you even recommend it?
Honestly, at least try to pretend like you don't swim in money and high-end components unlike us the average gamers :)
"If I Buy..."
that is all
So the author proves its a waste of money to add a third card but is pushing it as a good idea. Huh?
So you get a 15% boost going tri-fire versus buying a next gen card. The reasonble conclusion should have said to hold off and see what the next gen brings.
As the next gen could for all we know be 2x faster. It has happend a few times in graphics card history.
Just wait 2-4 freaking weeks adn then decide. Drive through come again.
I think your all gettin the wrong end of the author.
he was merely stating facts and even i am surprised from a third card making 15 %.
the author is not saying it is a good idea as he also mentions the risk of fire so the author is stating the findings. u f@ckin tits.
the prices of the 4890 went from about 230 to about 140 in a couple of months.
get real people the author is talking sense.
my only objection is the 4 cARDS AT A FULL X16 WHICH HARDWIRED YES this is correct but the links are adjusted to 4 x 8 links or 2 x 16 links.
Tyan S7025
http://www.tyan.com/product_board_detail.aspx?pid=641
Supermicro have one too.
umm, how about benchmark on stuff, you know, you can actually play?
like, umm, a...game!
Count your beans: If it's a 15% improvement "over the dual cards" it is a decent improvement.
If it was only 15% improvement of a "single" card it would be lame.
For those of you that are bad at math and play the lottery - Simple math example to show 15% of a larger number is much bigger
15% of 10,000 = 1,500
15% of 20,000 = 3,000
now go eat another doughnut~~