[ Print Article! ]

ATI Radeon 5850 Performance Preview
September 29, 2009

Summary: Not everyone's got $400 to spend on a shiny new Radeon 5870 card, which is why it can be argued that ATI's Radeon 5850 is the more relevant GPU for a lot of gamers and hardware enthusiasts. Based on the same DX11 RV870 architecture as the 5870, the 5850 delivers next-generation performance in a smaller, more efficient package, and most importantly it's priced for less than $300. ATI pitches it as their answer to the GeForce GTX 285...See how it compares in this article!


ATI Radeon 5850 Performance PreviewPage:: ( 1 / 20 )

[image]

<% print_image("01"); %><% print_image("02"); %><% print_image("03"); %>


With approximately 2.15 billion transistors, it’s the most complex chip ever created to grace the inside of your PC. It also holds the title of the fastest GPU money can buy today. This is a distinction previously held by NVIDIA’s GeForce GTX 285.

But not everyone can afford to shell out $379 for the Radeon HD 5870. This is a premium of 26% over the launch price of the Radeon 4870 last year. Of course, given the Radeon 5870’s larger die and superior performance, we think ATI’s new asking price for the 5870 is fair and reasonable, but given the state of today’s economy where hardware prices have fallen dramatically, consumers are now out to get the most bang from their buck.

Therefore ATI has also concocted the Radeon HD 5850. Built on the same RV870 architecture as the 5870, the 5850 gives you 90% of the shading and texturing units as the 5870 for $120 less.

Now because it runs at slower clock speeds also, that doesn’t mean it delivers 90% of the 5870’s performance, but it is built to give you most of the performance in a smaller, more case-friendly package that also consumes less power and generates very little noise. ATI pitches it as the perfect card for the performance-minded gamer on a budget who may have been contemplating a GeForce GTX 285 purchase.

Here are the raw specs on ATI’s Radeon 5850:

Radeon 5850 Specifications


TeraScale 2 Unified Processing Architecture

  • 1440 Stream Processing Units
  • 72 Texture Units
  • 128 Z/Stencil ROP Units
  • 32 Color ROP Units

    GDDR5 interface with 128.0 GB/sec of memory bandwidth

    PCI Express 2.1 x16 bus interface

    DirectX 11 support
  • Shader Model 5.0
  • DirectCompute 11
  • Programmable hardware tessellation unit
  • Accelerated multi-threading
  • HDR texture compression
  • Order-independent transparency

    OpenGL 3.2 support

    Image quality enhancement technology
  • Up to 24x multi-sample and super-sample anti-aliasing modes
  • Adaptive anti-aliasing
  • 16x angle independent anisotropic texture filtering
  • 128-bit floating point HDR rendering

    ATI Eyefinity multi-display technology
  • Three independent display controllers drive three displays simultaneously with independent resolutions, refresh rates, color controls, and video overlays
  • Display grouping: Combine multiple displays to behave like a single large display

    ATI Stream acceleration technology
  • OpenCL 1.0 compliant
  • DirectCompute 11
  • Double precision floating point processing support
  • Accelerated video encoding, transcoding, and upscaling

    ATI CrossFireX™ multi-GPU technology
  • Dual, triple, and quad GPU scaling
  • Dual-channel bridge interconnect

    ATI Avivo HD Video & Display technology
  • UVD 2 dedicated video playback accelerator
  • Advanced post-processing and scaling8
  • Dynamic contrast enhancement and color correction
  • Brighter whites processing (blue stretch)
  • Independent video gamma control
  • Dynamic video range control
  • Support for H.264, VC-1, and MPEG-2
  • Dual-stream 1080p playback support
  • DXVA 1.0 & 2.0 support
  • Integrated dual-link DVI output with HDCP (Max resolution: 2560x1600)
  • Integrated DisplayPort output (Max resolution: 2560x1600)
  • Integrated HDMI 1.3 output with Deep Color, xvYCC wide gamut support, and high bit-rate audio (Max resolution: 1920x1200)
  • Integrated VGA output (Max resolution: 2048x1536)

    Integrated HD audio controller
  • Output protected high bit rate 7.1 channel surround sound over HDMI with no additional cables required
  • Supports AC-3, AAC, Dolby TrueHD and DTS Master Audio formats

    Speeds and feeds
  • Engine clock speed: 725 MHz
  • Processing power (single precision): 2.09 TeraFLOPS
  • Processing power (double precision): 418 GigaFLOPS
  • Polygon throughput: 725M polygons/sec
  • Data fetch rate (32-bit): 209 billion fetches/sec
  • Texel fill rate (bilinear filtered): 52.2 Gigatexels/sec
  • Pixel fill rate: 23.2 Gigapixels/sec
  • Anti-aliased pixel fill rate: 92.8 Gigasamples/sec
  • Memory clock speed: 1 GHz
  • Memory data rate: 4.0 Gbps
  • Memory bandwidth: 128 GB/sec
  • Maximum board power: 151 Watts
  • Idle board power: 27 Watts

    Notes

    For the Radeon 5850 ATI takes the same RV870 chip used in the Radeon 5870 and disables two SIMD units, dropping the total number of active SIMD units to 18. If you recall, each SIMD unit contains 80 stream processors and 1 texture unit (4 effective), so ultimately ATI disables 160 stream processors and 2 texture units (8 effective) for the Radeon 5850.

    Besides turning off some of the chip’s functionality, ATI also drops the clocks down to 725MHz for the graphics core and 1.0GHz for the memory. With fewer shaders and slower clocks, ATI is able to source cheaper GDDR5 memory modules and a smaller PCB with less power circuitry and smaller cooling is needed. This helps to keep the Radeon 5850’s price and power consumption down in comparison to the 5870. We’ll be taking a closer look at the 5850 board itself on the next page.

    [image]

    <% print_image("04"); %><% print_image("05"); %>

    The following chart sums up how the Radeon 5850 stacks up against the Radeon 4870, 5870 and NVIDIA’s GeForce GTX 285:

    High-end GPU Comparison
    GPURadeon HD 4870Radeon HD 5850Radeon HD 5870GeForce GTX 285
    Manufacturing Process55-nm40-nm40-nm55-nm
    Graphics Core Clock Speed750MHz725MHz850MHz648MHz
    Stream Processor Clock Speed750MHz725MHz850MHz1,476MHz
    # of Stream Processors80014401600240
    Memory Clock Speed900MHz (3.6 Gbps data rate)1000MHz (4.0 Gbps data rate)1200MHz (4.8 Gbps data rate)1,242MHz (2,484MHz effective)
    Memory Interface256-bit256-bit256-bit512-bit
    Memory Bandwidth115.2 GB/sec128 GB/sec153.6 GB/sec159GB/sec
    Memory Size512MB/1GB1GB1GB1GB/2GB
    ROPs16323232
    Texture Filtering Units40728080
    Texture Filtering Rate30.0 GigaTexels/sec52.2 GigaTexels/sec68 GigaTexels/sec51.8 GigaTexels/sec
    Pixel Fill Rate12 GigaPixels/sec23.2 Gigapixels/sec27.2 Gigapixels/sec21.4 GigaPixels/sec
    Power Connectors2x6-pin2x6-pin2x6-pin2x6-pin
    Max Board Power160W151W188W183W
    Price$144.99-$229.99$259$379$325.99-$397.99





    Board analysisPage:: ( 2 / 20 )

    [image]

    <% print_image("06"); %>

    At first glance the Radeon 5850 looks very similar to the Radeon 5870, but ATI’s made some changes to the board that should make it more accommodating to end users.

    [image]
    <% print_image("07"); %><% print_image("08"); %><% print_image("09"); %>

    Most notably is the size of the board itself. Like previous high-end graphics cards from ATI, the Radeon 5850 measures 9.5” in length. This makes it nearly 1.5” shorter than the Radeon 5870, and about 1” shorter than the GeForce GTX 285 from NVIDIA. This should make life easier for those of you with smaller cases: the Radeon 5870 is so long it could interfere with the power/data cables of your hard drive(s) in some system cases.

    [image]
    <% print_image("10"); %><% print_image("11"); %>

    With its shorter PCB, ATI places the two power connectors on the end of the board, similar to the Radeon 4870 and Radeon 4890.

    The cooling system itself is pretty similar to the Radeon 5870. Like the 5870, the Radeon 5850 employs a quad-heatpipe cooling design. The heatpipes are made from copper to increase their effectiveness, just like the Radeon 5870. Three of the heatpipes are used to draw heat off the GPU, while the fourth rests over the voltage regulators. Speaking of the VRMs, ATI employs newer regulators that send temperature info to the GPU, so it can monitor temps of the regulators and adjust clocks to prevent overheating.

    A dual-slot aluminum heatsink (with copper base plate) is used to draw heat off the GPU. Obviously the heatsink used isn’t as large as the unit found on the 5870, but with fewer shaders cranking along at slower clocks, it doesn’t have to be as large either.

    The fan itself is the same part used on the Radeon 5870. The new fan ATI employs on their 5800 series cards is outfitted with quieter bearings; audibly the fan’s tone is lower, making it easier on the ears than previous ATI designs. It also spins at just 1200 RPMs at idle versus the 1600 RPMs used on the Radeon 4890 fan.

    [image]
    <% print_image("12"); %><% print_image("13"); %><% print_image("14"); %>

    The end result is a card that runs eerily quiet. Even when two cards are combined for CrossFire, noise output remains respectable; right around 54 decibels in our testing.

    Despite its increase in transistors and faster clocks, the 5800 series are able to run cooler and quieter than their predecessors thanks in part to the smaller 40-nm manufacturing process, but also thanks to new low power modes that reduce clock speeds and voltages more aggressively than ever. At idle the 5850 runs at just 157MHz core/300MHz memory and consumes just 27W of power.

    [image]
    <% print_image("15"); %>



    5850+5870 CrossFire

    One neat trick ATI GPUs have supported for the past two generations is the ability to mix and match graphics cards from the same family together. This tradition continues with RV870, as we were successfully able to combine our Radeon 5850 board with the 5870 for 2-Way CrossFire.

    When the two cards are combined for CrossFire, each card runs at its own independent clock speeds and SIMD configuration. In other words, the 5850 doesn’t OC itself to 5870 speeds nor does the 5870 underclock itself to 5850 speeds or disable two SIMDs. Performance was only up only marginally over the 5850 CrossFire configuration however, suggesting a potential quirk in the driver. We’ll have to investigate this more thoroughly when time warrants.

    DiRT 2 and availability

    Like the 5870, Radeon 5850 boards will be bundled with vouchers to download DiRT 2 once it arrives in November. We’ve also been told that cards will be hitting retailers shelves beginning today.

    Like the 5870, availability is going to be very tight. Supply won’t be sufficient to meet demand. So if you want one of these boards, you better have a fast internet connection constantly refreshing your favorite retailers and good eye/mouse coordination.





    System SetupPage:: ( 3 / 20 )

    3D Performance Testbed

    Intel Core i7-975 Extreme Edition @3.33GHz

    Gigabyte GA-EX58 Extreme
    6GB (2x3GB) OCZ Reaper HPC 1600 @ DDR-1066 Speeds

    NVIDIA GeForce GTX 295
    NVIDIA GeForce GTX 285
    NVIDIA GeForce GTX 275
    NVIDIA GeForce GTX 260-216
    NVIDIA GeForce 8800 GTX
    NVIDIA GeForce 9800 GX2
    NVIDIA GeForce 8800 GT 512MB
    ForceWare 190.62

    ATI Radeon HD 5870 1GB
    ATI Radeon HD 5850 1GB
    Catalyst 9.10 beta (8.66)

    ATI Radeon HD 4870 X2 2GB
    ATI Radeon HD 4890 1GB
    ATI Radeon HD 4870 1GB
    ATI Radeon HD 4850 512MB
    Catalyst 9.9

    300GB Western Digital Caviar SE

    Microsoft Windows 7 Ultimate 64-bit


    Power Consumption Testbed

    Intel Core i5-750
    Gigabyte GA-P55-UD6
    4GB (2x2GB) Kingston KHX1600C8D3K2/4GX @ DDR3-1333 Speeds
    Western Digital Raptor 150GB
    Crysis GPU bench used to test load power consumption

    Benchmarks

    Call of Duty: World at War
    Crysis
    Left 4 Dead
    Far Cry 2
    STALKER: Clear Sky
    Batman: Arkham Asylum
    Resident Evil 5
    Fallout 3
    Wolfenstein
    Tom Clancy’s HAWX



    Synthetic 3D TestingPage:: ( 4 / 20 )











    Fallout 3Page:: ( 5 / 20 )

    Fallout 3 DirectX 9






    Notes

    For the most part, the GeForce GTX 285 is able to hang with the Radeon 5850.
    Our Fallout 3 test sequence is CPU-bound at 16x12 and 19x12 with the SLI/CrossFire setups, while the Radeon 5870 is largely CPU-bound at 16x12, but at 2560x1600 the 5850 is only running 5% faster than the GTX 285.

    Of course, Fallout 3 is based on the same Gamebryo engine used by Oblivion, so it’s a pretty old graphics engine, and all the high-end cards are delivering playable frame rates. This is probably a case where we should’ve enabled 8xAA but we didn’t so that older cards like the 8800 GT wouldn’t turn into a complete slideshow.



    Call of Duty: World At WarPage:: ( 6 / 20 )

    CoD: WaW – DirectX 9






    Notes

    Like Fallout 3, the GeForce GTX 285 is able to keep up with the Radeon 5850 in CoD: World at War. The CoD 4 game engine that World at War is based on has traditionally favored GeForce cards, so it honestly doesn’t come as to much of a surprise to see the GTX 285 performing so evenly with the Radeon 5850. In fact, the GTX 285 outruns the Radeon 5850 by 5% at 2560x1600.

    So far, it doesn’t look like the Radeon 5850 is getting off to a good start, but again, we’re getting very good frame rates out of the 5850, and once again this is an older graphics engine: Modern Warfare 2 sports an entirely new game engine.



    Crysis DX10 High SettingsPage:: ( 7 / 20 )

    Crysis – DirectX 10





    Notes

    Unlike the previous 2 games we tested, Crysis is still one of the most graphically-intensive games out there. We’ve cranked the settings up to the game’s high settings with 4xAA, and as you can see, the Radeon 5850’s 1600-shader architecture really shines in this game. At 1600x1200, the 5850 ran 15% faster than the GeForce GTX 285, and that lead opens up to 20% by 2560x1600.



    Crysis DX10 Very HighPage:: ( 8 / 20 )

    Crysis – DirectX 10






    Notes

    Under Crysis’ Very High graphics settings, the GPU is pushed even further, and the Radeon 5850 responds by expanding its lead over the GeForce GTX 285. At 16x12 the Radeon 5850 is running 18% faster than the GeForce GTX 285, and that edge increases to 21% by 2560x1600 (although admittedly neither card is nowhere near playable levels here). The CrossFire setups scale nicely as well.



    Far Cry 2Page:: ( 9 / 20 )

    Far Cry 2 – DirectX 10






    Notes

    While it isn’t quite as graphically demanding as Crysis or STALKER: Clear Sky, Far Cry 2 does push the graphics cards better than most games out there, especially with the graphics settings cranked up to Ultra High. The Radeon 5850 starts out with a 9% lead over the GTX 285 at 1600x1200, but as the screen resolution increases, the GTX 285 manages to close within 3% of the 5850 by 2560x1600. Obviously the 5850’s at a memory bandwidth disadvantage compared to the GTX 285, and that could be playing a role in the GTX 285’s strong showing in this game at 2560x1600 even though we’re just running 4xAA.



    STALKER: Clear SkyPage:: ( 10 / 20 )

    STALKER Clear Sky – DirectX 10






    Notes

    The Radeon 5850 managed to outgun the GeForce GTX 285 in STALKER Clear Sky by 8% at 1600x1200, and that lead opens up to 16% at 2560x1600.



    Left 4 DeadPage:: ( 11 / 20 )

    Left 4 Dead – DirectX9





    Notes

    Considering that Left 4 Dead is based on the absolutely ancient Source engine, we were a bit surprised to see the Radeon 5850 perform so well in comparison to the GTX 285 in this test. This sequence comes from the game’s very last level though, which is a madhouse with zombies coming from all directions. The GeForce GTX 285 trails the Radeon 5850 by 12% at 1600x1200, and this lead opens up to 14% by 2560x1600.



    Resident Evil 5Page:: ( 12 / 20 )

    Resident Evil – DirectX 10





    Notes

    The GeForce GTX 285 is largely able to hang close with the Radeon 5850 in Resident Evil 5, which is based on Capcom’s MT Framework game engine (the same engine used for Lost Planet, Devil May Cry 4, Street Fighter 4). At 1600x1200, the 285 trails the Radeon 5850 by just over 1%, close enough to call it a draw. As we increase the screen resolution the Radeon 5850 is able to pull away, but even at 2560x1600 the margin separating both cards is just 6%.



    Batman: Arkham AsylumPage:: ( 13 / 20 )

    Batman:AA – DirectX 9





    Notes

    We test Batman: Arkham Asylum with and without AA as GeForce boards currently have the advantage of an in-game AA option with Batman, whereas ATI cards are forced to rely on the control panel to enable AA. This gives the GeForce boards an advantage. To quote ATI:


    In this game, Nvidia has an in-game option for AA, whereas, gamers using ATI Graphics Cards are required to force AA on in the Catalyst Control Center.

    The advantage of in-game AA is that the engine can run AA selectively on scenes whereas Forced AA in CCC is required to use brute force to apply AA on every scene and object, requiring much more work.

    Additionally, the in-game AA option was removed when ATI cards are detected. We were able to confirm this by changing the ids of ATI graphics cards in the Batman demo. By tricking the application, we were able to get in-game AA option where our performance was significantly enhanced. This option is not available for the retail game as there is a secure rom.


    Even without AA though the GeForce GTX 285 manages to run even with the Radeon 5850. We rely on FRAPS runs for our testing with this title, so considering the margin of error we’re calling this one a definite draw.

    That obviously doesn’t bode well for the 5850 once AA is turned on, as you’ll see on the next page.




    Batman: AA w/4xAAPage:: ( 14 / 20 )

    Batman:AA – DirectX 9






    Notes

    The results aren’t pretty for the Radeon 5850 once AA is forced via control panel. Not only does the GeForce GTX 285 outrun the 5850, so does the GTX 275 and GTX 280 at lower resolutions. By 2560x1600 though the 5850 does manage to close within 12% of the GTX 285’s performance.

    Hopefully ATI will be able to work with Square-Eidos on getting in-game AA integrated into Batman: Arkham Asylum. While the Radeon cards are clearly delivering playable frame rates with AA forced via control panel, we’d like to see both ATI and NVIDIA hardware running on as level a playing field as possible, especially considering this is such a good game.



    HAWXPage:: ( 15 / 20 )

    HAWX – DirectX 10





    Notes

    Thanks to DirectX 10.1, ATI’s hardware is able to deliver excellent performance in HAWX. Just for a frame of reference, forcing DX10 lowered the Radeon 4890’s frame rate to just 48fps at 1920x1200.

    With this advantage, the Radeon 5850 is able to outperform the GTX 285 by 7% at 1600x1200, and this lead opens up to 23% by 2560x1600.



    WolfensteinPage:: ( 16 / 20 )

    Wolfenstein – OpenGL





    Notes

    ATI’s lackluster OpenGL performance rears its head in Wolfenstein. Clearly this is one issue that ATI still hasn’t managed to resolve.



    8xAA vs 4xAA ScalingPage:: ( 17 / 20 )

    Batman




    Left 4 Dead



    HAWX





    8xAA vs 4xAA Scaling (cont’d)Page:: ( 18 / 20 )

    Resident Evil 5



    Call of Duty: World at War




    Fallout 3





    Temps, Power and OverclockingPage:: ( 19 / 20 )

    Power Consumption





    Temps




    Overclocked Performance

    While Overdrive officially tops out at speeds of 775MHz core/1125MHz memory for the Radeon 5850 (speeds we were easily able to hit with both 5850 cards), we used AMD’s GPU Clock Tool to crank the speeds up even further. We managed to hit speeds of 815MHz core/1350MHz memory with card #1, while card #2 maxed out even higher, clocking in at 878MHz core/1400MHz memory.












    ConclusionPage:: ( 20 / 20 )


    If you’re a fan of NVIDIA’s hardware, you’re probably feeling better than you'd expected right now. After all, on paper ATI's Radeon 5850 outclasses the GTX 285 in shading horsepower and fill rate, yet our bone stock GeForce GTX 285 was able to keep up with the Radeon 5850 in a lot of titles, and even outran the next-generation Radeon board in games like Batman and Call of Duty. In Fallout 3 and Resident Evil 5 the two boards are pretty close, with the Radeon 5850 ultimately delivering better performance. The GeForce GTX 285 is able to close within 3% of the 5850’s performance at 2560x1600 in Far Cry 2 also.

    These are all cases where a factory OC’ed GTX 285 board from ASUS, Gigabyte, EVGA, or XFX could close the gap even further, if not take the lead in performance outright.

    Clearly ATI’s got the lead in technology though, and we’re not just talking about DirectX 11 either. You can see this reflected in DirectX 10 games that really push the hardware like Crysis and STALKER: Clear Sky. Here the Radeon 5850 is able to gain leads of 15-20% over the GeForce GTX 285 under some cases.

    The cynics will probably maintain that RV870’s 256-bit memory interface is holding it back, but again, we really don’t think that’s the case. They argue that with double the shaders, double the texture units, double the hardware-based resolve for anti-aliasing, and both pixel and texture fill-rate are over 2X greater than RV770, RV870’s memory subsystem is the only performance metric that doesn’t improve by a factor of 2X, creating an obvious bottleneck on paper. Memory bandwidth is only improved by 33%.

    We don’t agree though based on a few different factors. Number one is RV870’s 8xAA performance scaling. As we showed you last week and again today, the card generally doesn’t take an enormous performance hit when running 8xAA. In games like Resident Evil 5, Fallout 3, Batman: Arkham Asylum, and Call of Duty: World at War, the performance hit was less than 5% at 1920x1200 for 5870. And in all the games we tested the performance hit of running 8xAA was less than 10%.

    With such a slim performance hit at a fairly high resolution like 1920x1200, clearly this isn’t a card that’s starving for more memory bandwidth.

    Earlier this week, we also dabbled with 5870 overclocking, OC’ing the 5870 GPU and memory by a fixed amount of 9% respectively. In practically every case the 5870 card scaled best when the GPU/shaders were OC’ed rather than memory: performance typically improved by 4-5% in most apps when the GPU was running at 930MHz, while OC’ing the memory to 1320MHz only improved performance by 2-3% in the same games.

    If the card was truly being bottlenecked by its memory interface, it should’ve shown more significant gains when we OC’ed the memory.

    The other reason why we’re pretty certain the driver is holding the 5870’s performance potential back is history: when was the last time you saw a next-generation architecture deliver 100% of its performance on launch day?

    The answer to that question would of course be never. It takes time for the driver teams to squeeze all the performance out of the hardware. In fact, it isn’t unheard of for performance improvements to be found a year after a next-gen GPU launch.

    Again, as we’ve stated from the outset, RV870’s clearly being held back by its drivers. We’re not seeing the full performance potential of the hardware today, and we may not know how well RV870 scales for many more months.

    ATI’s clearly got the best overall GPU on the planet right now though. Sure, ATI’s OpenGL performance still needs more work, but honestly this isn’t as big of an issue as it was 5 years ago. If you’re in the market for a $300 graphics card, the Radeon 5850 delivers the best combination of price, features, and performance in this price segment and we’d recommend it over the GeForce GTX 285, which currently sells for about $325 and up. Until NVIDIA’s able to answer with DirectX 11 hardware of their own, this is going to be a tough comparison for NVIDIA, but rumor has it that we could see something as soon as December. For NVIDIA’s sake, they’re going to need it, but in the meantime, they need GeForce GTX 285 price cuts even more desperately.

    With Radeon 5850 boards hitting shelves now, it’s the card to get in our opinion.



  • © Copyright 2003 FS Media, Inc.
    [ Print Article! | Close Window ]