Sapphire Radeon HD 6670 Ultimate Video Card Review

Product: Sapphire Radeon HD 6670 Ultimate
Company: Sapphire
Author: James Prior
Editor: Pete Vagiakos
Date: July 29th, 2011

Vision Engine Analysis

Image Quality

Both the Turks and Sumo graphics cores are based on the Redwood design from the Evergreen product stack, but they differ in the improvements applied to them. Turks, like Barts and Caicos, got improved Anisotropic Filtering which addressed the kernel transistion issue identified in the Evergreen series. With the Radeon HD 6670 (based on Turks) as the primary GPU, we should get the benefit of the improved filtering to reduce the possibility of shimmer, vs. the Sumo and Juniper cores (click the images for full quality png images):

Radeon HD 6550D

Sumo Anistropic Filtering Angle Dependency
Sumo Anisotropic Filtering Kernel Transitions

Radeon HD 6690D2 (Turks) + Sumo

Sumo Anistropic Filtering Angle Dependency
Sumo Anisotropic Filtering Kernel Transitions

Radeon HD 6770 (Juniper)

Sumo Anistropic Filtering Angle Dependency
Sumo Anisotropic Filtering Kernel Transitions

The Sumo, Turks and Juniper cores all share angle independence when applying anisotropic filtering, something NVIDIA and Intel have yet to demonstrate.

Gaming performance will be demonstrated with two titles using benchmark tools from John Mautari, of RadeonPro fame.

Aliens Vs. Predator

Two resolutions benchmarked; 1600x900 with medium texture quality and medium shadow quality, tessellation enabled, advanced shadow sample disabled, and SSAO off. 1920x1080 with low texture quality and low shadow quality, tessellation enabled and advanced shadows & SSAO off. Anisotropic Filtering level and FSAA level was altered in an attempt to see the nature of AA/AF performance hit at a resolution, and for different graphics hardware:

1xFSAA, 1/4/8/16x AF:

At 1600x900 with medium settings the Sapphire HD 6670 is just under being in the enjoyable playable range (average 45fps), with no AA/AF. The performance hit going to 1920x1080 and low settings is just under 24%, dropping towards 20% as AF and AA are added. All the GPU configurations do this, showing nearly the same impact on performance by increasing resolution. Enabling Dual Graphics increases performance significantly, 47-48% increase in average FPS and now almost to the 60fps sweet spot and only 1-2% behind the 6770. At 1920x1080 the increase is still in the 45-46% increase range, now 2-3% behind the 6770. At these numbers though, that translates into identical performance.

The Anistropic Filtering performace hit at 1600x900 is 6.4% (4xAF), 11.2% (8xAF), 13.5% (8xAF). This reduces at 1920x1080 to 4.4% (4xAF), 6.9% (8xAF), 9.1% (16xAF). Switching to Dual Graphics the 1600x900 hits are reduced (16x9: 4.9%, 8.9%, 11.5%;19x10: 2.6%, 5%, 7%) pretty consistently. This brings the HD 6690D2 inline with the 6770 in terms of hit for applying Anistropic Filtering.

Again, but with 2xFSAA

Adding 2xFSAA is a big performance hit for the 6670, 25-28% performance loss vs. 1xFSAA at 1600x900, 28-30% at 1920x1080. The performance hit of adding AF is smaller than at 1xFSAA (16x9: 4.3% [4xAF], 8.3% [8xAF], 10.6% [16xAF]; 19x10: 2.6% [4xAF], 5.2% [8xAF], 7% [16xAF]). As we increase the resolution and anti-aliasing, Anistropic Filtering is getting 'cheaper' - it's costing less and less performance to add it.

Moving to Dual Graphics, the 6690D2 performance hit when applying AF stays about the same percentages as the 6670 alone, but we get a 41-45% increase in performance to go with it. HD 6690D2 is still within 1-2% of the 6770 performance, and both are pretty close to the enjoyable playable range at 1600x900 with 39fps for 2xFSAA/16xAF and cruising along at the bottom end of playable at 1920x1080, keeping above 30fps with 2xFSAA/16xAF.

Now with 4xFSAA

At 4xFSAA we're into the frustrating and slideshow FPS territory, taking a ~40% performance hit from 1xFSAA at 1600x900 and 35% at 1920x1080. This is indicating that the bottleneck is moving, more towards the number of poly's being processed. The 6670 AF hit at 1600x900 is 4.2% [4xAF], 7.6% [8xAF], 10.2% [16xAF], about the same as when 2xFSAA was applied. At 1920x1080 the hit is smaller, 2.5% [4xAF], 5% [8xAF], 7% [16xAF]. Moving to Dual Graphics bumps performance back up to playable at 1600x900, hitting an average FPS of 3-5fps over the 30fps marker. The 6770 is starting to keep a consistent 1-2% performance advantage, but at these FPS numbers its not noticeable in gameplay.

The AF performance hit for the Dual Graphics 6690D2 is 2.5%/4.4%/6.6% at 1600x900 and 1.5%/2.2%/4% at 1920x1080 (4xAF/8xAF/16xAF, respectively). This is similar to the 6770, which shows 3.2%/6.4%/8.8% and 1.4%/3.5%/5.3% (1600x900; 1920x1080; 4xAF/8xAF/16xAF respectively).

Now fixed 1xAF but varying FSAA:

The 6670 at 1600x900 resolution incurs a performance hit of 39% at 2xFSAA, and 58% for 4xFSAA. Interestingly the delta between 2x and 4x is only 14%. The percentages are nearly indentical at 1920x1080, (39%, 58%; 13%). Dual Graphics 6690D2 is 45% faster at 1xFSAA, 41% faster at 2xFSAA, and 37% faster at 4xFSAA, for the 1600x900 resolution. These increases are maintained at 1920x1080 resolution. The 6770 is 1-3% faster than the 6690D2 configuration, with nearly identical AA hit.

For all graphics configurations, FSAA incurrs a high performance hit. The DG 6590D2 42%, 67%


Benchmark average FPS results from the Adrenaline Codemasters benchmark tool, Finland stage. Two resolutions benched; 1600x900 in DirectX 11 with medium preset and varying AA; 1920x1080 in DirectX 11 with low preset and varying AA.

For the 6670, the performance hits are 7%/14.4%/35% [2/4/8xAA, 1600x900] and 8.6%/18.8%/55.8% [2/4/8xAA, 1920x1080). Performance for both resolutions is good at the selected in game settings, hitting around the 60fps average sweet spot for fluid smooth gaming with no AA and 2xAA. The ~15% and 19% hit incurred by 4xAA still leaves performance at a decent level for enjoyable play. 8xAA tanks performance but still in the target area for playable gaming, close to the 45FPS marker. Enabling Dual Graphics gives a 35% boost in average FPS, at 1xAA. The increase drops to 32% at 2xAA for 1600x900, but stays at 35% at 1920x1080. At 4xAA, the increase is still in the 33-35% range, and increases to 38%/40.7% at 4xAA [1600x900,1920x1080].

The 6770 starts to stand out more in these tests. The 6770 is 45% faster at 1600x900 and 44% faster at 1920x1080, 1xAA. At 1600x900 with 2xAA, the 6770 is 45% faster than 6670 and 9.4% faster than 6690D2. This scaling remains as AA increases: 46%/9.8% vs. 6670/6690D2 at 4xAA, 52%/10.2% at 8xAA for 1600x900 resolution; 47%/8.4% at 4xAA and 59%/13% at 8xAA for 1920x1080.

Compared to the HD 6670 alone, adding APU graphics in CrossfireX to the $105 Sapphire HD 6670 Ultimate gives you free 8xAA at 1600x900 and 8xAA costs 10% at 1920x1080. Spending $10 more (difference after rebate, 6770 Vapor-X is currently ~$135USD with $15MIR) gets you free 8xAA at 1920x1080. Both the dual graphics option and the 6770 will allow you to run higher quality settings in game at 1080p, to tune performance and quality around the 60fps average mark.


Using a Kill-a-watt power meter we recorded total system power draw for three use cases, for different conditions. Idle represents Windows 7 idle with aero enabled, MS Office 2010 Word and Excel running and a PDF document open in Adobe Reader X. DiVX playback is during watching a fullscreen 1080p DiVX HD video in Cyberlink PowerDVD 10 MkII with AMD Picture Perfect enabled. 3DMark11 is recorded during 3DMark11's GPU Test 2 running with Extreme settings at 1920x1080 resolution.

Below you can see AMD Picture Perfect HD in action as Cyberlink PowerDVD 10 MkII is used with AMD Video enhancements enabled, for playback of a high definition DiVX movie file. Click the pic for full size (1920x1080).

DiVX HD Playback

In operation the Sapphire HD 6670 Ultimate idled at a low 34C, a nice temperature given the 25C ambient room temperature, with video playback around 40C. Under load we never saw temperatures above 60C, indicating the dual direct contact heatpipe design is very effective in our case. We expect higher temperatures in less airflow orientated cases, such as HTPC or SFF desktop.

GPU-Z Info
GPU-Z Idle

The 6670 adds 25-30% to the idle power draw of the Lynx platform, regardless of if the APU graphics are enabled or not. The 6770 increases idle power by 10% over the 6670, and 45% over the APU alone. Using Dual Graphics for gaming adds a 45-50% perfomance increase, and the 6770 can increase peformance by 55% over the 6670 alone, at times. Our gaming load power reading indicates that dual graphics uses 22% more power than the APU alone. The 6770 uses 40% more power to deliver a little more than 50% performance increases in our tests against the 6670, where the Dual Graphics option uses 30% more than the 6670 for a similar increase in performance.