|ATI Radeon HD5450 HTPC Video Card|
|Reviews - Featured Reviews: Video Cards|
|Written by Bruce Normann|
|Thursday, 04 February 2010|
Page 5 of 16
Radeon HD5450 Detailed Features
The big news about the HD5450 is the reduced size of the GPU die. Once again, the newest member of the HD5xxx family has roughly half the number of transistors as the previous one. The chip in the HD5450 is codenamed "Cedar", and has approximately 292 million transistors on it, compared to 627 million on the Redwood chip, which was released last month with the HD5670. The small size is critical to the cost strategy that ATI is pursuing with all these new releases. After some very lean years, struggling to make it in the graphics chip industry, it appears ATI has finally figured out how to make money.
The Cedar die packaging on the HD5450 GPU is a little bigger than half of the Redwood, because the number of interconnects is roughly the same, with the notable exception of the memory interface, which is 64 bits wide this time. The number of Stream Processors has been radically reduced, from 400 down to 80; only 20% of the number present on the Redwood chips. It's evident there are a lot of transistors consumed in the other functions of the GPU, besides shaders.
The memory specification is going to be somewhat flexible for the HD54xx products. I expect most units will be sold with 512MB of GDDR3, but GDDR2 is a possibility, perhaps for OEM variants. 1 GB versions will also be available, although it may be tough to make the case for any performance advantage to be had with that configuration, with the exception of Eyefinity usage. As mentioned above, the GPU-memory interface is 64 bits wide with a maximum bandwidth of 12.8 GB/s, using GDDR3. That's a major hit, compared to the HD5670, with 64GB/s of bandwidth, but I suspect it's well balanced by the greatly reduced number of Stream Processors.
The K4W1G1646E-HC11 GDDR3 memory chips are sourced from Samsung. They are rated for a maximum clock rate of 900 MHz, and the marketing specs for the card indicate a maximum clock rate of 800 MHz. Version 0.3.8 of GPU-Z reported that the memory on my sample unit was running at 900 MHz. That's engineering samples for you.... I was suspicious that the clock rate is not being reported correctly in GPU-Z, since it has not been updated to support this new chip yet, but ATI confirmed that they had been building some "extra curricular" prototypes for testing, and some of them got into reviewer's hands. Oh well; their loss and my gain, I guess. Too bad they didn't send me some 1 GHz parts, as they are available from Samsung, according to the product specs shown here below.
The power section of the HD5450 video card is simplistic, and optimized for both cost and low power. In this case, all of the dynamic performance scaling is built into the GPU, and the voltage regulators just ride along. I was observing the shader and memory clocks in GPU-Z while using the PC for normal office-type duties, and this card ramped the clocks up and down faster and more dramatically than any card I've used recently. That's where the power savings are going to be made with this card, getting it down very quickly to idle power, which is a miserly 6.4 watts.
Looking at the business end of the passive cooling, I found something that made my mechanical engineering heart shiver. That little square platform, located between the four threaded standoffs, is created by cutting away all the aluminum around it with a milling machine. Then the platform itself is milled to create the flat, smooth surface we all know is essential for good heat transfer from the mirror surface of the GPU. The word you should be thinking is, "Expensive". I'm pretty sure that the way the GPU chip is mounted on the board makes this expensive manufacturing step essential, and I'm also sure that somewhere there are a couple electronics packaging engineers delivering some serious noogies to the PC Board designer that made all this necessary.
The assembly quality is quite good, for an engineering sample. The heat sink had a bit of a gouge in it, but it was put there before the part was anodized, so it doesn't stick out too much. The soldering and surface mount component placement was reasonably well done, and the overall board layout was well designed, with a rational flow.
Before we dive into the testing portion of the review, let's look at one of the most exciting new features available on every Radeon HD5xxx series product, Eyefinity.