Radeon HD 5830 Performance Preview

Alex Castle

AMD exploits a price point with the Radeon HD 5830, but the implementation is so weird, we’re scratching our heads.

If you’ve got $250 to burn for a graphics card, you’ll find a dearth of cards at that price point. Hit any of the major web retailers for PC gear looking for $250 cards, and you’ll find a couple of models of Nvidia’s GeForce GTX 275 and… that’s it. The much faster Radeon HD 5850s are around $300, and you’ll find older GTX 260s and not much more.

AMD decided to fill the gap with the Radeon HD 5830. But the HD 5830 is a really odd duck. It’s slower than a Radeon HD 5850, but the reference implementation is huge – the same size as the Radeon HD 5870. The 5830 also consumes more power at full throttle than the HD 5850 – hence the larger cooler on the reference design.

Four Radeons HD 5xxx cards lined up, in order of performance. The HD 5870 is at the top, the HD 5770 at the bottom. Note how much longer the HD 5830 reference card is, compared to the Diamond HD 5850, which is based on the reference design.

AMD is apparently aware of these inconsistencies, so you won’t find many reference designs for sale. In a departure from the norm, most of the cards you’ll be able to buy at retail ship with OEM customized cooling solutions. So we’ll be seeing a lot of cards that at least look different at launch date, such as this board for XFX:

This board from Sapphire:

And this board from Gigabyte:

The HD 5830 is still based on the same Cypress GPU as the higher end Radeon HD 5870 and HD 5850 boards. So what exactly is different? And how does it differ from the sub-$200 Radeon HD 5770? Let’s take a look.
Feature Comparison
Radeon HD 5770 Radeon HD 5830 Radeon HD 5850 Radeon HD 5870
GPU Core Juniper
Cypress Cypress Cypress
Transistor Count 1.04 Billion 2.15 Billion 2.15 Billion 2.15 Billion
Core Clock 850MHz 800MHz
725MHz 850MHz
Stream Processors 800 1120 1440 1600
Compute Performance 1.36 TFLOPS 1.79 TFLOPS 2.09 TFLOPS
Texture Units 40 56 72 80
ROPs 16 16 32 32
Z/Stencil (Gsamples / sec) 54.4 51.2 92.8 108.8
Memory Clock 1200MHz 1000MHz 1000MHz
Memory Bus 128-bit 256-bit 256-bit 256-bit
Memory Bandwidth 76.8 GB/Sec 128 GB/Sec 128 GB/Sec 153.6 GB/Sec
Idle Power 18W 25W 27W 27W
Full Throttle Power 108W 175W 151W 188W

The oddities just jump out at you.

Find out what we make of it, and see the all-important benchmarks on page two!

The number of ROPs are half that of the 5850, which will likely affect AA performance. Stream processors are right in line, but memory bandwidth is out of whack – on the high side – for a “middle” product. Then there’s the core clock: 800MHz. Since the 5830 uses the same Cypress GPU as the 5850 and 5870, with more functional units disabled, the clock speed had to be juiced to hit performance targets.

All those decisions come together to explain why the maximum board power is higher than the HD 5850 while performance is lower. But how much lower is the performance? Let’s find out. We ran benchmarks against a Radeon HD 5770 and HD 5850 to compare.

Performance Results

We ran the reference HD 5830 in our standard graphics testbed:
•    Intel Core i7 975@ 3.3GHz
•    Asus Rampage II Extreme X58 motherboard
•    6GB Corsair Dominator DDR3 @ 1333MHz
•    Corsair TX850W 850W PSU
•    Seagate 7200.12 1TB hard drive
•    Lite-On 20x DVD+/-RW optical drive
•    Gateway FPD2485 24-inch display (1920x1200)

We’ve updated our standard benchmark suite to include a couple of DirectX 11 benchmarks. We’re now running:
•    3DMark Vantage
•    Unigine Heaven DX11 Benchmark
•    Crysis (DX10 mode)
•    Far Cry 2 (DX10 mode)
•    STALKER: Call of Pripiyat (DX11)
•    DiRT2 (DX11)
•    Tom Clancy’s HAWX (DX10)
•    Battle Forge (DX11)

We ran our tests at two settings: 1920x1200, max detail, AA off and 1920x1200, max detail, 4x AA. 3DMark Vantage includes the performance, high and extreme settings.

First, let’s take a look at the two synthetic benchmarks.

Synthetic Benchmarks
Radeon HD 5770 Radeon HD 5830 Radeon HD 5850
Unigine Heaven, AA Off 22.5 fps
25.6 fps
Unigine Heaven, AA On 16.8 fps 19.4 fps
3DMark Vantage Perf 10,963 (score) 14,042 (score)
14,416 (score)
3DMark Vantage High 6,807 (score) 8,968 (score) 9,605
3DMarkVantage Extreme 4,661 (score) 6,043 (score) 6,579

3DMark Vantage tells us that the HD 5830 really is close to an HD 5850. The more recent Unigine Heaven benchmark – which is based on an actual game engine – shows the HD 5830 behaves more like the lesser HD 5770. So which synthetic benchmark more accurately resembles reality? Let’s see what stories the game benchmarks have to tell.
Game Benchmarks
Radeon HD 5770 Radeon HD 5830 Radeon HD 5850
Battle Forge (AA Off) 33
41 49
Battle Forge (AA On) 24 29 39
FC2 (Action, AA Off) 48 53 63
FC2 (Action, AA On) 38 42 53
FC2 (Ranch Long, AA Off) 62 67 83
FC2 (Ranch Long, AA On) 42 44 61
HAWX (AA On) 52 59 67
HAWX (AA Off) 42 46 58
STALKER CoP (AA Off, Tess On) 47 49 68
STALKER CoP (AA, Tess On) 23 26 33
Crysis (AA Off) 20 21 25
Crysis (AA On) 17 21 25

Well, this doesn’t look good. The Radeon HD 5850 essentially crushes the 5830, more so than the $49 price differential might warrant. That’s not to say the HD 5830 is a bad card – it certainly outpaces the HD 5770. But it’s closer in performance to a 5770 than a 5850 in real game tests.

We’ll be reviewing actual shipping cards in the near future; it’s possible that benchmarks will change a bit with officially shipping drivers. Given that Radeon HD 5770s are priced around $170, the HD 5830 looks a little overpriced. This card would be a better deal at around $215-$220, and I’m guessing the price will eventually settle closer to $200. At $250, it’s much better to squeeze out a few more dollars and pick up a Radeon HD 5850 for $299.

Around the web