ATI Radeon HD 5550 GDDR5 Review

While the Radeon HD 5550 may not be a new card, the inclusion of GDDR5 is.  This should add a small boost in performance over the GDDR3 and DDR2 models while retaining the low-power profile, perfect for inclusion in a Home Theater PC (HTPC) build.

Introduction

The Radeon HD 5550 is an entry level card aimed at the HTPC market.  The most notable features are it’s very low power draw (39W maximum) and hardware acceleration for video.  If you throw a Bluray Disc Drive in with it, then you can also stream the uncompressed audio directly to your receiver without the need for another audio card.  All in all, the specs add up to a decent option for a HTPC and possibly even worthy of a low end gaming computer with it’s DirectX 11 support.  When you consider this card has a MSRP of $75, it begins to look even better.

Notable Technologies

• ATI Stream technology – hardware acceleration for video
• ATI Eyefinity technology – run up to 3 monitors off one card
• Bitstream Audio – pass uncompressed audio from Bluray Discs straight to a receiver via HDMI

Specifications (as seen by GPU-Z)

	HD 5550 (as tested)	HD 5570	GT 220
Core Clock	550 MHz	650 MHz	625 MHz
Memory Clock	1000 MHz	1000 MHz	790 MHz
Memory Type	GDDR5	DDR3	DDR3
Memory Size	512 MB	1024 MB	1024 MB
Memory Bus Width	128 Bit	128 Bit	128 Bit
Direct X Support	11	11	10.1
ROPs	16	16	8
Shaders	320	400	48
Texture Fillrate	22.9 GTexel/s	29.4 GTexel/s	14.9 GTexel/s
Pixel Fillrate	11.4 GPixel/s	11.8 GPixel/s	7.4 GPixel/s
Memory Bandwidth	73.6 GB/s	33.3 GB/s	31.2 GB/s
GPU Technology	40 nm	40 nm	40 nm
Transistors	627 M	627 M	486 M

You can see in the images below that ATI was kind enough to send us a reference card to test.  I was told that the retail models will be low-profile cards so they can be used in Small Form Factor (SFF) cases.

Testing Platform

To test this puppy, I’m going to throw it in my relatively old Core2Duo/P45 based system.  I say “relatively old” because it’s not the most recently released hardware, obviously, but is still readily available and still performs well for gaming and multimedia.

• Processor: Intel E8400 @ 4.23 GHz
• Motherboard: Asus P5Q
• RAM: 4 GB GeiL Ultra DDR2 @ 470 MHz
• Hard Drive: Western Digital 74GB Raptor
• Power Supply: Enhance ENP-5150GH 500W
• Heatsink: Zalman CNPS9900
• Operating System: Microsoft Windows 7 x64, Phoronix Test Suite Live (Ubuntu)
• Drivers: Catalyst 10.5 and 10.6

Benchmarks

• Cyberlink Mediashow Espresso (encoding)
• Youtube 1080p and 4k resolution (video playback)
• Hulu Desktop (video playback)
• 3DMark 2001, 2003, 2005, 2006, and Vantage (gaming)
• Aquamark3 (gaming)
• Unigen (DirectX 11 gaming)
• Alien vs. Predator (DirectX 11 gaming)
• Linux: JXRendermark
• Linux: Warsow

Hardware Acceleration Performance

First, I have to mention that hardware acceleration will not work without the Avivo package, which is separate from the standard Catalyst Control Center package. It looks like they’ve already fixed this problem, but their download page used to say the Avivo package was not needed for 5000 series cards and I spent a great deal of time searching the internet to find out why I could not get the hardware acceleration to work.  However, I still don’t think it is shown prominently enough on the page to show that this package is highly recommended, if not mandatory, for proper use of this card.

I actually didn’t notice that much of an improvement in processor usage while using hardware acceleration, but there was a huge difference in encoding time.  To encode in Mediashow Espresso, I created a 2 minute video (640×480, AVI) using my camera then used the iPhone 3g preset (h.264 video, AAC audio).

Encoding Time

HW Accel. On	HW Accel. Off	Difference
24 seconds	43 seconds	19 sec. (44%)

Processor Usage

UPDATE (7/22/2010):

Our ATI representative brought a few issues to my attention and apparently I missed a setting or two while testing the processor usage.  So, I re-ran the Cyberlink Mediashow Espresso and Youtube tests to see if I would get different results this time.  For the record, I did.  I’m not sure exactly what I missed in the previous tests, but in this round of tests I did see a good difference in processor usage between having the hardware acceleration turned on an off.  For the YouTube testing, I used a 4K Resolution playlist to play two videos in succession at both 1080p and original 4k resolution, Life in the Garden and Surf NYC.  These are the same videos, played in the same manner for the original test.  When I changed the Hardware Acceleration setting in Flash, I made sure to refresh the page several times to make sure the setting took.

Encoding Time
Hardware Acceleration On	Hardware Acceleration Off	Difference
25 Seconds	44 seconds	19 seconds (44%)

Processor Usage
	MediaShow Espresso		Youtube 1080p		Youtube 4k resolution
	Core0	Core1	Core0	Core1	Core0	Core1
Hardware Acceleration Off	93%	85%	39%	39%	87%	86%
Hardware Acceleration On	43%	52%	27%	20%	84%	83%
Difference	50%	33%	12%	19%	3%	4%

Overclockability

Surprisingly, I was able to work a 30% overclock (715MHz) on the core, which was the maximum that the MSI Afterburner program would allow.  Not as spectacular, the memory would not go past 1150 MHz.  There was more room on the slider bar, but ever time I hit apply it would revert back to 1150 MHz.  This happened on both the 10.5 and 10.6 drivers.  Also, the exact same clocks were stable on both the 10.5 and 10.6 drivers.

Gaming Performance

Without any further delay, let’s get to the numbers:

3DMark

	10.5 Stock (Marks)	10.5 Overclocked (Marks)	10.6 Stock (Marks)	10.6 Overclocked (Marks)
3DMark Vantage	3849	4849	3875	4683
3DMark06	8358	10217	8435	10291
3DMark05	15902	19477	15988	19477
3DMark03	25308	31736	25388	32013
3DMark01	50702	53137	51505	53999

Aquamark3

	Score
10.5 stock	144920
10.5 oc	154694
10.6 stock	144679
10.6 oc	157406

Alien vs. Predator  – DirectX 11

	Avg. FPS
10.5 stock	11.3
10.5 oc	14.5
10.6 stock	11.3
10.6 oc	14.5

Unigine – DirectX 11

	Avg. FPS	Min. FPS	Max. FPS	Scores
10.5 stock 1600×1200	9.7	6	22.9	243
10.5 stock 1024×768	15.2	8.4	46.6	384
10.5 oc 1600×1200	12.4	7.6	29	313
10.5 oc 1024×768	19.5	10.7	59	491
10.6 stock 1600×1200	9.7	6	22.3	244
10.6 stock 1024×768	15.2	9	45.7	382
10.6 oc 1600×1200	12.3	7.6	28.5	310
10.6 oc 1024×768	19.4	10.7	58.5	490

Overall, this low-end card did pretty well with gaming.  It’s not setting any world records, that’s for sure, but it can easily handle older games like Counter Strike: Source and World of Warcraft.  I even got to use it with the StarCraft II beta, and I had no problem running the game at 1600×1200 with all settings at Medium.  As for the comparison between the 10.5 and 10.6 drivers, it’s almost a toss-up but the 10.6 drivers did have a slight edge when the card was overclocked.

For a quick and dirty overview of performance, you can check out this hwbot compare link to see what people are able to obtain with the various competing cards.  As of this writing, I’m the only one with scores submitted for the HD 5550 GDDR5.

Linux Performance

Since this card is aimed at the HTPC market and my personal favorite software to run is XBMC for Linux, I decided to run a  few tests using the Phoronix Test Suite Desktop Live DVD.

JXRenderMark

	32×32 (Ops./sec)	128×128 (Ops./sec)
Rects Composition	110067.89	26996.69
Put Composition	170497.09	20728.61
Simple Blit	482905.75	138167.72
Transformed Blit Linear	98998.78	8309.44
Transformed Blit Bilinear	24891.39	2017.26
Transformed Texture Paint	26365.21	1909.9
Linear Gradient Blend	53296.33	5047.67
Radial Gradient Paint	31678.78	2363.81
Gradient Temp Texture	45277.17	4796.62
12pt Text Grayscale	18277.98	5445.58
12pt Text LCD	19292.26	5587.61

Warsow

	Avg. FPS
800×600	266.53
1024×768	230.43
1280×960	184.93
1440×900	184.06
1280×1024	176.2
1400×1050	162.5
1600×1200	142.53

These benchmark runs are also viewable online: JXRendermark and Warsow.

Conclusion

In conclusion, I think this card packs a lot of punch for a sub-$100 price.  If you are building a HTPC or low-end gaming system, certainly take this GPU into consideration.

– splat