Testing the ATI Radeon HD 5870

Posted by Jon Peddie and Robert Dow on September 30th 2009 | Comments Closed
Categories: Hardware Review
Tags: gpu ati amd 3d graphics radeon gpgpu 2d teraflops

Figure 1: ATI’s Radeon HD5870 AIB

By now you’ve probably read our review of the RV870 and a half dozen others so you should already know it’s suppose to be a 2.7 TFLOPS chip with 1,600 processors and ultra fast 2GB of GDDR5 memory.

The board is totally enclosed, with air vents at the back, and oddly the chip’s connector and heat-sink retention bracket is exposed, which adds a strangely aesthetic appeal.

It could be we got an early test unit and the production version will have a cover plate over the chip. You can see the two six-pin power connectors at the top of the rear of the AIB, and in the front the two Crossfire connectors, so the AIB is clearly designed to operate with up to four boards.

Notice the four connectors, one HDMI, one DisplayPort, and two dual-link DVI. The card supports up to three displays so you can combine the two DVI connections with either the DisplayPort or HDMI output. You can run up to three displays off the board use the DVI connectors and either the HDMI or the DisplayPort for the third.
However, the DisplayPort connection requires an active adaptor dongle to convert it to a DVI connection for the third monitor. Active adaptors are pretty hard to find and if you can, they cost about $100. ATI may end up having to offer their own adaptor, instead of just dumping the problem on the unsuspecting customer.

Test results

Figure 2: ATI Radeon HD5870 test results

We ran the new AIB on Nehalem Bloomfield Core i7 CPU 870 @ 3.20GHz with Windows 7, and tested it against a Radeon HD4870, and an Nvidia GTX285.

The actual scores are shown in the following table

Table 1: Test results for ATI Radeon HD5870
Vantage Pre-Sets GPU Test 1 Test 2
HD5870
  Performance 15,966 48.53 44.55
  High 10,998 33.69 30.69
  Extreme 8,023 25.53 25.4
GTX 285
  Performance 12,575 37.71 35.93
  High 8,651 27.47 23.13
  Extreme 6,063 20.42 15.01
HD4870
  Performance 9,318 23.56 28.06
  High 5,816 17.73 16.32
  Extreme 4,150 13.64 10.64

The average GPU score difference between the HD5870 and the GTX285 was 28.8%, and between the HD 5870 and the HD4870 84.6%. So the 5870 is a significant gain over the last generation ATI chip and a pretty good gain over Nvidia’s latest generation chip.

When we ran the tests on the 5870 and tested it against a 4870 and an Nvidia 285 (see the table above) the results of the 1,600 core 5870 compared to the 800 core 4870 were spectacular, but against the 240 core 285 although they were good, were not as spectacular. This leads us to two questions:

  • What the heck are all those processors doing besides consuming power? Why does the 4870 which is close in performance to the 285 look overwhelmed by the 5870, and the 285 doesn’t?
  • As we were pondering these questions, Robert wondered if we were running a significant enough test? Well, perhaps the drivers are tweaked for the 285 and perhaps the 5870’s are still a little green, but Robert’s probably right -- but tuning them up isn’t going to double the scores.

Upping the ante

We added two more tests, the Unigene and in-game Resident Evil, and we turned on all the cinematic features we could adjust.

Unigine

Figure 3: High res, all filters test of AIBs in Unigene

Unigine (Topics), running on DirectX 10.1, and turned AA up to 8x, AF up to 16x, shaders high, reflections on, ambient occlusion (AO) on, and off, and ran it at the highest resolution the test supports, 1920 x 1200. We also ran a series with 4x AA and 8x AF. Unigene gives you two measurements: FPS and an overall score.

Here you can see the benefit of lots of processors when you turn on all the filters you can get on.

Table 2: Performance relative to the HD5870
  Score FPS
  5870 4870 285 5870 4870 285
"8xAA
16xAF
OA on"
100.0% 59.1% 46.1% 100.0% 59.2% 46.1%
"8xAA
16xAF
OA off"
100.0% 58.2% 56.4% 100.0% 58.5% 56.3%
"4xAA
8xAF
OA on"
100.0% 59.9% 55.2% 100.0% 60.3% 55.2%
"4xAA
8xAF
OA off"
100.0% 59.0% 67.3% 100.0% 58.9% 67.3%

Resident Evil 5

Figure 4: AA test of AIBs in Resident Evil 5
Figure 5: Effects of motion blur on frame rate

Using DirectX 10 we could see significant improvement when max AA was turned on at high resolution.

We ran out of time before we could run all the tests, but turning on motion blur really impacted the score. Look at the difference between the 5870 and 285 with motion blur on and off.

Interestingly turning on motion blur had no effect on the Radeon HD5870.

DirectX 11

Similar to the launch of the ATI Radeon 9000 series which, came with a coupon for Half-Life 2, the Radeon HD 5800 and HD 5700 series come with a coupon for DiRT 2, a DirectX 11 game coming in December.

ATI said over 20 DirectX 11 games are expected and at the press and analyst’s meeting mid month showed clips from EA’s Battlefield, Rebbion’s Alien vs. Predator, Code Master’s Dirt.

Less power

Figure 6: DirectX11 games (Source AMD)

One of the most notable features of the new chip/AIB was a major drop in idle power consumption to just over 27 watts compared to the previous generation which is about 90 watts. Maximum board power has increased on the HD 5870 by about 17% or so, but with 2x improvements in most parameters that seems justified.

The AIB definitely runs cooler than its predecessor. The HD 4870 and 4890 cards got really hot along the tops and the back during prolonged use.

What do we think?

Figure 7: Increase in FLOPS of GPUs (log plot)

ATI has been leading the market with product introductions and in terms of FLOPS for the past few years, as shown in the next chart

With competition from Intel’s forthcoming Larrabee, Nvidia’s relentless pursuit of graphics dominance, and even from the future Fusion product, there’s no breathing room or safe harbor for ATI, so we expect to see the division continue to churn out breath taking new parts

Discuss this entry

Commenting is not available in this section entry.

Comments