By: Hilbert Hagedoorn | Edited by George Panayiotou and Ian Barling | Published: September 30, 2009
Veni, vidi, vici (I came, I saw, I conquered) -- That all classic Latin phrase applied to ATI last week when they released the Radeon HD 5870 graphics card; the first ever DX11 class graphics card with a lot of features and raw brute horsepower to ensure you can play your games not only in the best quality, but also with a lot of performance headroom.
There is however one slight disadvantage with the Radeon HD 5870, its price tag. Surely everybody agrees that 399 USD might be a fair price for this all new 40nm GPU based graphics card, however... it's also a sum of money that most of us can't afford to spend, especially in these harsh economic times.
Well, ATI of course will address that issue with a truckload of graphics cards that will be cheaper. But in-between now and Christmas 2009 in the high-end segment two products will make an appearance. The dual-GPU based Radeon HD 5870 X2, which will be even more expensive, but there was another product launched last week that did not get enough attention as... well the cards had not been distributed just yet. That's right, it is of course the Radeon HD 5850.
As such today we'll look into the Radeon HD 5850, we'll check both the single GPU performance but also will check it out in CrossfireX multi-GPU mode. This card promises to go neck-to-neck with the 349 USD GeForce GTX 285 in terms of performance. The Radeon HD 5850 however comes with full DirectX 11 compatibility, Eyefinity and some sheer brutal rendering performance in both gaming and compute features. It's everything the Radeon HD 5870 really is, just a slight step slower but more affordable.
Meet that mean little sister of the Radeon HD 5870... The 259 USD Radeon HD 5850 1024MB.
The Radeon HD 5850
Last week we gave you a pretty deep insight on what the Radeon HD 5870 is all about. What you guys need to remember is that the Radeon HD 5850 is 100% similar to the feature set of that card. As such we'll be going a little less deep in this article, yet focus more on specifications, performance, layout and board design.
To understand the Radeon HD 5850 and 5870 we need to explain a thing or two though.
The new Radeon Series 5000 products are tagged under the family name 'Evergreen' and within that Evergreen family over the months to come we'll see several products released, what you see below is unconfirmed information, though:
Hemlock - dual-chip (X2) flagship graphics solutions powered by two RV870 chips (originally known as R800).
Cypress - single-chip high-performance graphics solution based on the RV870 chip that will replace the ATI Radeon HD 4890; this is the Radeon HD 5870 and as tested today the Radeon HD 5850.
Juniper - single-chip performance graphics solution that will replace the remaining Radeon HD 4870 and 4850 graphics boards. We expect Radeon HD 5750 and 5770.
Redwood - single-chip mainstream graphics solution(s) based on the RV830 chip that will replace both RV730 and RV740-based solutions.
Cedar - single-chip entry-level graphics solution(s) based on the RV810 chip.
The Radeon HD 5870 has gone on sale for $379, while the cheaper entry in the 5800 series, the Radeon HD 5850 is priced at $259. Both cards are fast, really fast. In the 5850's case it outperforms a GeForce GTX 285 in nearly any scenario as our tests will show.
So we have just established product positioning and price point, now let's have a peek at some of the key features for this product series:
1GB GDDR5 memory
ATI Eyefinity technology with support for up to six displays (four on default reference cards)
ATI Stream technology
Designed for DirectCompute 5.0 and OpenCL
Accelerated Video Transcoding (AVT)
Compliant with DirectX 11 and earlier revisions
Supports OpenGL 3.1
ATI CrossFireX™ multi-GPU support for highly scalable performance
ATI Avivo™ HD video and display technology
Dynamic power management with ATI PowerPlay technology
DL-DVI, DL-DVI, DisplayPort, HDMI
PCI Express® 2.0 support
We'll address the preponderance of features in our article, but let's focus on the sheer technical specifications, transistor count for example.
The number of transistors on the 5850 is not at all different from the 5870. Both products have the same GPU, yet one cluster of shader processors simply has been disabled.
So while the Radeon HD 4870 had 956 million transistors embedded onto its die, the new Radeon HD 5800 GPUs have 2.15 billion transistors. Yes Sir, that is 2150 million transistors tucked away in a small chip. The fabrication node is 40nm for this product, resulting in a die size of 334 mm², which for AMD is monolithic, yet thanks to the 40nm fabrication process only (roughly) a third bigger than the previous 4890 GPU.
Now you'd think with so many transistors high clock frequencies would be an issue. Incorrect, the high-end Radeon HD 5850 will be clocked at a steady 725 MHz on its core and shader processor domain. The gDDR5 memory is clocked at 1000 MHz (4000 MHz effective). Both can be overclocked way higher though... more on that later, of course.
Shader processors then, we went from 800 Shader processors on the Radeon HD 4850/4870/4890 to 1440 shader processors (also called stream processors) on the Radeon HD 5850. The ROPs went up from 16 to 32 as well and sure... texture units from 40 to 72 as well.
Radeon HD 5850 has 9x 160 Shader clusters = 1440 Shader processors
Radeon HD 5870 has 10x 160 Shader clusters = 1600 Shader processors
But before you get blinded by all the specs in a few lines of text, let's break down the two new cards announced in comparison to last year's Radeon HD 4870.
Radeon HD 4870 Radeon HD 5850 Radeon HD 5870
Process 55nm 40nm 40nm
Transistors 956M 2.15B 2.15B
Die Size 263 mm² 334 mm² 334 mm²
Core Clock 750 MHz 725 MHz 850 MHz
Shader Processors 800 1440 1600
Compute Performance 1.2 TFLOPs 2.09 TFLOPs 2.72 TFLOPs
Texture Units 40 72 80
Texture Fillrate 30.0 GTexels/s 52.2 GTexels/s 68.0 GTexels/s
ROPs 16 32 32
Pixel Fillrate 12.0 GPixels/s 23.2 GPixels/s 27.2 GPixels/s
Z/Stencil 48.0 GSamples/s 92.8 GSamples/s 108.8 GSamples/s
Memory Type GDDR5 GDDR5 GDDR5
Memory Clock 900 MHz 1000 MHz 1200 MHz
Memory Data Rate 3.6 Gbps 4.0 Gbps 4.8 Gbps
Memory Bandwidth 115.2 GB/s 128.0 GB/s 153.6GB/s
Maximum Board Power (TDP) 160W 151W 188W
Idle Board Power 90W 27W 27W
As you have been able to observe, the 5850 only has one shader processor partition less than the 5870. However its clock frequency is a good amount slower. But that's nothing we can't tweak out we dare to say.
One of the biggest accomplishments of the series 5000 graphics cards is the enhancement in the power design, the implementation of voltage and clock regulation is even more dynamic -- power management at a new level.
So we'll look purely at the Radeon HD 5850 now, in IDLE the GPU will clock down and lower its voltages on both GPU and memory. Have a look:
GPU Radeon HD 4870 Radeon HD 5850 Radeon HD 5870
Max. Board Power (TDP) 160W 151W 188W
Idle Board Power 90W 27W 27W
The card obviously achieves a low 27W IDLE power consumption by clocking down with several power states. Thus a low engine (core) clock frequency with lowered voltages and lower GDDR5 memory power. It's amazing though as your generic high-end graphics card would normally consume 50~60 Watts when it idles in Windows.
Things get even better though, the performance of the graphics card opposed to the last generation products has nearly doubled up in performance and design, yet the 5850 has a TDP (peak wattage) of only 151 Watts.
For architecture, PCB and voltage design we like to recommend you to read our reference Radeon HD 5870 review. But talk is cheap, let's have a look at the product.
DirectX 11
Aah finally, a new DirectX. It's funny how most game developers skipped DX10 really. Face it, if there are not enough changes over DX9 then why should software houses invest in a new code path and thus spend extra money on development? This literally was a problem with DX generation 10. Next to that add the stupendous limitation from Microsoft to limit DX10 to Windows Vista only. Probably the most horrendous call Microsoft ever made for an operating system.
Good news though, DirectX 11 is an extensive step upwards for both developers and gamers. Developers can speed up their games and improve it with more complex shaders and a few new tricks like tessellation. Gamers on their side can have faster running games with some really cool new eye candy. This is the new shader palette for developers to use: Vertex, Hull, Domain, Geometry, Pixel and Compute Shaders. With the compute shader comes DirectCompute as well, allowing Windows Vista or Windows 7 to utilize the GPU directly from within Windows. It's a first step but quite a number of applications that would benefit from GPU computing now can make use of it. This really is a revolutionary step in development, as parallel processing can be really helpful in specific situations and this software.
Here are the most prominent new features of DirectX 11 (and I'm keeping this as simple as possible) that will effect you directly:
Shader model 5.0
Multi-threading
DirectCompute 11 - Physics and AI
Hardware Tessellation
Better shadows
HDR Texture compression
Let's highlight and discuss the five more important features that will effect you the most.
Multi-threaded rendering
Much like modern day applications and processors, it is now possible to fire off code and datasets directly towards the GPU multi-threaded, we call this multi-threaded rendering. Your gain here is efficiency. If an instruction or shader has to be queued up (single threaded) that creates latency, a delay. The GPU as such can now handle all the data completely threaded. And that means better overall performance.
Think of a hundred cars that have to move over a single lane road from point A to B.
Now imagine a hundred lane road where all hundred cars have a lane available.
Which approach do you think would get all cars to point B the quickest? Exactly. I'll probably receive a few emails from programmers and developers for this oversimplified explanation though.
Fact is, and this you need to remember with multi-threaded rendering, DirectX will take better advantage of all the available processing cores.
DirectCompute 11
Another new feature in DirectX 11 I find very exciting is DirectCompute. It allows Windows 7 and Vista developers to make use of the parallel processing power of modern video cards; software developers will have access to the GPU and can use it to help out the system processor with tasks that involve say, high-quality video playback or high performance transcoding.
In its most simple explanatory form DirectCompute allows access to the GPU for Stream Computing (acceleration, post processing, whatever). As such DirectCompute allows for more easy access to the GPU’s many cores for parallel processing, thus utilizing the GPU for stuff other than gaming. Examples here are Stream Computing, transcoding videos over the GPU (which is something we'll be testing today as well later on in this review).
What about games you ask? Well, you could implement and use DirectCompute 11 for image processing and filtering (post processing), order independent transparency (really cool feature where you could see through an object like it was made out of glass), shadows rendering, physics, Artificial Intelligence and sure... Ray Tracing as well (though very limited).
I just touched order independent transparency (OID) and quickly wanted to show you that feature through a little video.
Now ATI will very likely release this footage at high quality somewhere this week, but I made a recording of an OTD technology demo. The quality is poor as it is recorded HD camera footage from a regular monitor. But you'll get the idea, in this demo we'll use a "Mech" and apply the OID technology (Proper rendering of sorted transparent geometry). Look closely at how you can see through objects like that Mech as it where a 3D X-ray. You can actually use this feature in smoke, fire, hair, foliage, fences, grates and so on. In this particular demo DirectCompute is utilized to enable single pass transparent pixel sorting
Some stats: the environments is build out of of 343 thousand triangles, the Mech is built out of 262 thousand triangles.
MORE.http://www.guru3d.com/article/radeon-hd-5850-review-crossfire/5
Marked Topics
