How well will the Ati Radeon HD 5770 do in the reviews? PC Games Hardware tests the first DirectX 11 card of the Juniper family which is supposed to attract some attention with a price of 140 Euros.
Today Ati's Radeon HD 5770 has to show if it can continue the good launch of the bigger sisters Radeon HD 5870 and Radeon HD 5850. Both models convinced in their reviews due to high performance, a better picture quality in comparison to the previous generation (enhanced Anisotropic Filtering and Supersampling Anti Aliasing as well as the targeted price. By now some cards are actually available and those people who bought one credit the Radeons a good price-performance ratio which has been part of AMD's graphics card philosophies for some time now. Now PC Games Hardware puts the Radeon HD 5770 to the test to see what this mid-range DirectX 11 card is capable of. Radeon HD 5770 reviewed: Specifications
HD5770 06 [Source: view picture gallery]
The graphics chip with the codename Juniper supports DirectX 11 like the progenitor Cypress, but has less functional units. So to put it bluntly: The Juniper is more or less a halved Cypress. Accordingly the graphics processor of the HD 5770 has 10 SIMD units with 16 processors each of which consists of five ALUs (800 ALUs all in all). On the HD 5750 one of the SIMDs is deactivated and thus the number of shaders is reduced to 720 and the chip frequency has been reduced from 850 to 725 MHz.
Each SIMD has access to a quad TMU, so there are 40 respectively 36 TMUs. Like Cypress Juniper is working with shader based Texture Address Interpolation which prevents a bottleneck like the one of the previous card generation. Four quad ROPs - good for a throughput of 16 per cycle - are dealing with most transactions from and to the GDDR5 controller which is integrated twice - buffered by Level 2 caches.
AMD couldn't yet answer the question fully if the caches have the same capacity per unit as on the Cypress chip, but the press office assumes it currently. Some saving measurements have been applied to the display controller, and thus to the Eyefinity feature, too, but it has not been halved. A Juniper chip can deal with up to five instead of six displays.
In the chart below we have summarized all technical details known to us. Radeon HD 5770 reviewed: Power Consumption, Loudness and Cooling
With our equipment we record the acoustic pressure in Decibel (a) and the loudness in Sone at a distance of 50 centimeters. The remaining components of our test system are silent: A AMD Athlon X2 BE 2350 with 2.1 GHz is passively cooled by a Scythe Orochi on a Asrock A780 FDP with AMD 780G chipset. The system is powered by a Nesteq X Zero PSU with 600 watt output and the operating system is stored on a Hama SSD. The ambient noise of our sound lab doesn't exceed 0.1 Sone.
The temperatures and the power consumption are also tested with this test system. We record 2D results on the Windows desktop as well as results at full workload. For this we use Race Driver Grid at 1,920 x 1,200 with 4x MSAA and 16:1 AF. The game is more challenging for the graphics card than Crysis Warhead and other games with a constant workload.
As an absolute worst-case scenario we use Ozone3D Furmark in version 1.6.5. With a renamed exe file this tool stresses the graphics cards to the maximum and is most challenging for the cooling solutions. Important: This scenario reveals what the absolute worst-case is in matters of loudness and power consumption. Such results are most unlikely to be reached during all day usage.
Results: Radeon HD 5770
HD5770 13 [Source: view picture gallery]
Beneath the cover of the Radeon HD 5770 there is a 70 mm radial fan with 0.66 ampere which supplies a very small heatsink - about the size of two matchboxes - with fresh air. Apparently the copper base utilizes the Vapor Chamber technology and has 30 aluminum fins attached.
Like the HD 5870 and the HD 5850 the HD 5770 is quiet and requires requires little energy in idle mode: 0.9 sone and 22 watt are very good values, but are a little higher than the results of the bigger graphics cards - this could be caused by a spread for standard factory models or measuring inaccuracy. Given the temperature of only 41 degrees Celsius in 2D reduction of the fan speed to 20 percent (default 35 percent), which results in almost inaudible 0.2 sone, is no problem. When stressed with games (Race Driver Grid) the temperature of the HD 5770 increases to 70 degrees Celsius and the fan increases the speed which results in audible, but not annoying 2.3 sone. This is more quiet than a HD 4870 or a GTX 260-216. In Furmark the temperature goes up to almost 80 degrees and the noise level reaches 2.9 sone.
Radeon HD 5770 reviewed: Overclocking
HD5770 10 [Source: view picture gallery]
In theory the Juniper should offer better overclocking capabilities than the Cypress since less functional units are run outside the specifications. With 100 percent fna speed (15 Sone!) and an additional 120 millimeter fan our sample ran stable at 930/2700 MHz. This is an advantage of about 9 percent for the chip and about 12.5 percent for the memory frequency. The GDDR5 memory is from Hynix and carries the tag H5GQ1H24AFR T2C 934A - thus it is specified to 2500 MHz.
With those OC clock speeds the performance in Crysis Warhead was increased by about 12 percent - apparently the Juniper longs for additional bandwidth. If you edit the device ID of the HD 5770 into the "Rivatuner.cfg” the Rivatuner is able to read the card's temperatures and fan speed, but overclocking is not possible.