The World's Fastest Graphics Card, Full Review!!
One hell of an Enormous Graphics power by ATI
Introduction
Know what really sucks? Developing a piece of hardware that everyone wants, and then not being able to make enough of it to go around. Of course, what happens next is a result of supply and demand. Suddenly your $249 Radeon HD 5850s cost $300—when they’re in stock.
Wouldn’t it follow, then, that a card with two of those über-valuable ASICs would do little to address supply issues of ATI’s Radeon HD 5800-series cards? According to the company, the exact reason it held off on launching its potent Hemlock board related to availability. Now, it says, there are enough Cypress GPUs to support sales of this…monster.
Clearly, Size Matters
It goes 5970, 5870, and 5850
Meet the Radeon HD 5970. Apparently, Radeon HD 5870 X2 too-conspicuously conveyed the dual-GPU nature of this card—an attribute that has not necessarily been celebrated in the past by those wary of scaling or stuttering issues.
Nevertheless, this is ATI’s dual-GPU, single-PCB follow-up to the Radeon HD 5870 (and answer to Nvidia’s GeForce GTX 295—a card that, up until today, was the fastest discrete graphics board you could buy). Gone are all auspices of value. This thing is set to sell for $600. Almost ironically, though, the 5970 might end up being the way to go for anyone who was previously considering a pair of Radeon HD 5850s in CrossFire, which would add up to $600 today anyway.
ATI’s approach here looks a lot like what Nvidia did with its GeForce GTX 295 earlier this year.
1) Take two top-end GPUs. In Nvidia’s case, it was the GT200 with 240 stream processors. In ATI’s, it’s the 1,600-shader Cypress found in its Radeon HD 5870.
2) Drop clock rates a bit in order to keep power and thermals under control.
3) Profit
The most significant difference is that, while Nvidia hacked and slashed the back-end of its GPU to match what you’d get from a pair of GeForce GTX 275s, cutting ROPs and 64-bits of the originally-512-bit memory interface, ATI’s Radeon HD 5970 sports a pair of uncut Cypress chips. Only the clocks are dropped (to Radeon HD 5850 levels), and for a fairly sensible reason that we'll get into shortly, too.
What results is a massive, single-PCB card that stretches a foot long—an inch more than the already-lengthy Radeon HD 5870 and an inch and a half longer than ATI’s Radeon HD 4870 X2. This will unquestionably be a sticking point for enthusiasts with cramped enclosures. Enthusiasts are a resourceful bunch, though.
More Sweet Spot Action
Hemlock represents the third of four projected launches based on the Evergreen family, first discussed in early September. The first, of course, was Cypress—the Radeon HD 5870 and Radeon HD 5850. The second was Juniper—the Radeon HD 5770 and 5750. The next launch, after today’s, is expected in the first quarter of 2010 and will consist of the entry-level Cedar and Redwood components.
At least on paper, AMD’s Sweet Spot strategy is going off without a hitch.
In reality, the same availability issues that plagued the Radeon HD 4770 shortly after launch are taking their toll on Radeon HD 5870 and Radeon HD 5850 supply (the 5770 and 5750s seem to be a little more widely available). Blame, of course, is being directed at TSMC’s 40nm manufacturing node and its resulting yields.
Of course, that’s a shame since supply affects pricing. But ATI tells us that only the 5850 is being hit by a $50 price jump. The 5870 remains at its $399 price point (though you’ll see some vendors selling the card for more). Moreover, the Radeon HD 5970’s $599 MSRP is being set with already-known supply issues priced in—expect it to remain around that $600 level.
Now, let’s dig into more depth on what you’re getting for $600—$100 more than Nvidia’s GeForce GTX 295.
The Making Of A Radeon HD 5970
As with the Radeon HD 4870 X2 and Nvidia’s second-generation GeForce GTX 295, ATI’s Radeon HD 5970 populates a single PCB. It consists of two 2.15 billion transistor Cypress GPUs with 1GB of GDDR5 memory each, joined by a 48-lane PLX PCI Express bridge. The bridge is, for the most part, the same one seen on last-generation’s Radeon HD 4870 X2. However, it has been updated for PCI Express 2.1, an incremental and non-performance-related evolution.
Why Not Go All-Out?
Much of the board’s length can be attributed to the onboard power circuitry needed to drive the two Cypress GPUs. Here’s where many of the design decisions behind this card were actually made.
For example, ATI reduced the voltages it used and correspondingly dropped the 5970’s clock rates to Radeon HD 5850 levels. This was done to keep maximum board power to 294W—under the defined 300W PCI-SIG electromechanical specification, delivered through the physical slot (75W), one six-pin auxiliary connection (another 75W), and an eight-pin auxiliary connector (150W). Pushing Radeon HD 5870 frequencies (850 MHz core/1,200 MHz memory) would have pushed max. board power closer to 375-400W and immediately cut out a segment of enthusiasts who don’t have twin eight-pin auxiliary power connectors on their power supplies.
However, ATI says the Radeon HD 5970 was designed to run at those clock rates. It features specially-screened low-leakage ASICs that run cooler than higher-leakage parts. It employs 5 Gb/s GDDR5 memory actually rated for 1,250 MHz. And perhaps most important, its vapor chamber-based cooling solution is designed with enough capacity to dissipate as much as 400W.
The only missing piece is an official voltage tweaking utility. In an unprecedented move by a GPU vendor, ATI went so far as to provide us with such a utility—a reference app—to give us access to those elevated settings. According to the rep who briefed us, third-party board vendors will bundle their own voltage apps along with hardware so that those with capable-enough PSUs will get the chance to push the hardware a little further—at least to 5870 levels, we’re hoping.
That’s A Huge Board
Those of you who thought the Radeon HD 5870 was already “healthy”-sized, this 5970 is even larger—an inch longer, to be exact. Despite the increase in length, the board’s exterior isn’t much different from what you saw on the Radeon HD 5870—the same fully-shrouded red/black motif with faux rear-vents are still in effect. Because ATI kept power down under 300W, it gets away with one six-pin and one eight-pin auxiliary connector on the board’s top-edge. There’s a single CrossFire connector, should the holiday season treat you well and you want to spend $1,200 on graphics cards.
Perhaps most notable is the rear I/O bracket. Gone is the HDMI output connector (you’d have to be crazy to tie this card into an HTPC). Instead, ATI exposes two dual-link DVI outputs and a single mini-DisplayPort output. Thus, Eyefinity is still viable here, right up to 3 x 2560x1600.
ATI uses the space freed up by the smaller DisplayPort and missing HDMI outputs for a full-length exhaust vent. While there’s ventilation all along the top of the board, most of the card’s heated air exits the back. In contrast, the Radeon HD 5870 is far guiltier of re-circulating hot air.
Overclocking ATI’s Radeon HD 5970
Although ATI felt the pressure to keep its Radeon HD 5970 under 300W for the sake of those folks with 750 and 850W power supplies, the card was originally designed to hit 5870 frequencies and deal with the corresponding thermal load that would have created. Thus, you have hardware on-board well-suited for overclocking, yet technically overkill for the 5970’s stock specifications.
There’s the 400W-capable vapor chamber-based cooler, for example. ATI also uses a programmable PWM fan controller able to monitor 12 different points on the board. Screened ASICs, higher-binned memory chips, and beefier digital VRMs are all part of the effort to infuse extra headroom “similar to what you get on a Black Edition Phenom II,” AMD says. That last part is a bit ironic; anyone willing to spend $600 on graphics should be looking to an overclocked Core i7 to help balance it out.
Complementing the purportedly more-scalable hardware is a bit of special software. The voltage tweaking utility comes first. ATI’s reference example took our GPU from 1.05V to 1.1625V and our memory from 1.1V to 1.15V. We’re curious to see if third-party board vendors choose any voltage levels above or below those levels. Second, ATI caps the core and memory clocks much higher, letting you choose up to 1,000/1,500 MHz frequencies. We were able to get our sample stable at 925/1,300MHz.
Extracted from http://www.tomshardware.co.uk
No comments:
Post a Comment