by Alexey Stepin , Yaroslav Lyssenko
01/19/2010 | 09:57 AM
We used to be rather skeptical about the prospects of multi-GPU subsystems built out of entry-level graphics cards. Their reliance on the game engine’s and driver’s support made them unsuitable for most gamers who were likely to prefer a single top-end graphics card instead. Although this statement is largely true to this day, ATI’s CrossFire and Nvidia’s SLI technologies should be given credit for having gone a long way since their early days and problems with compatibility and performance. Today, these multi-GPU technologies work just fine in a majority of modern and old games and if they fail to do so in a new game, the compatibility issue is usually resolved very quickly.
So, our position on this issue had to be adjusted and the global economic recession impacted it, too. As a result, we have published a series of reviews dedicated to entry-level multi-GPU solutions based on such cards as Radeon HD 4670, Radeon HD 4770 and Radeon HD 4830.
Recently we also reviewed CrossFireX subsystems built out of AMD’s inexpensive Radeon HD 5700 series products, describing symmetric tandems of two Radeon HD 5770 and two Radeon HD 5750 cards. But we know that, unlike the capricious Nvidia SLI technology that might require not only that you use identical graphics cards but also that these cards have identical BIOSes, the ATI CrossFireX technology in its current implementation works just fine on asymmetric configurations. Therefore we want to check out what we can expect from a multi-GPU configuration assembled out of a Radeon HD 5770 and a Radeon HD 5750.
We think there is a practical point in this. CrossFireX-compatible mainboards are widely available and comparable to mainboards with only one PCI Express x16 slot in price. The Radeon HD 5750 can be bought for less than $150 in retail. Thus, if you don’t have $400 or more to spend for a Radeon HD 5870, you can equip your gaming subsystem with a rather cheap but fast Radeon HD 5750 and later add in a more expensive Radeon HD 5770. This combination would have a total of 1520 ALUs, 76 TMUs and 32 RBEs, which is more than what you get from a Radeon HD 5850 but at a price lower than the Radeon HD 5870. The opportunity of building this graphics subsystem in two steps makes it even more appealing.
We’ve got a couple of cards necessary for such a test, an ASUS EAH5770/2DIS/1GD5 and PowerColor PCS HD5750 1GB GDDR5 Premium Edition, and are going to show you how appealing an asymmetric CrossFireX tandem can be. First, let’s take a look at each card of the tandem.
This product comes in a rather small box that looks familiar to us. We’ve seen the same knight on the package with the ASUS EAH4890.
The design is far from original, but doesn’t look ugly, either. There is no useful info on the face side of the box, save for the type and amount of graphics memory. The interior of the box is divided into a few cardboard compartments. Besides the card wrapped into an antistatic pack, there are the following accessories here:
The disc with software contains drivers and ASUS SmartDoctor, a utility for software-based control over the graphics card’s GPU voltage. The ASUS EAH5770/2DIS/1GD5 supports this feature fully.
The accessories are not gorgeous, but sufficient for a graphics card priced below $200. You even get a power adapter you can use if your power supply doesn’t have a PCIe 1.0 connector. We don’t have any serious complaints about the packaging or accessories of this product.
The first member of our asymmetric CrossFireX tandem is a copy of the reference card. While this is normal for top-end solutions like Radeon HD 5870 or GeForce GTX 285, inexpensive cards are often custom-designed. Anyway, this product only differs from the reference Radeon HD 5770 with the sticker on the cooler’s casing.
Asus EAH5770 (left) and ATI Radeon HD 5770 reference (right)
A few jumpers and smaller elements that we saw on an engineering sample of Radeon HD 5770 are missing. They must have been utilized by AMD for testing and certifying their new design and are not necessary on a serially produced product. We could find no other difference between the ASUS EAH5770 and the reference card, so you can refer to our review of the latter for details.
To remind you, the reference Radeon HD 5770 has a rather advanced “3+3” power system which is redundant for a 40nm RV830 chip but will allow reusing this PCB for the more voracious successors to the Juniper GPU. There is only one external power connector. With the introduction of the 40nm tech process, even the Radeon HD 5870 can do with two 6-pin PCIe 1.0 connectors whereas 8-pin PCIe 2.0 connectors are only installed on topmost dual-processor solutions like ATI Radeon HD 5970 or Nvidia GeForce GTX 295.
The GPU voltage regulator is based on an L6788A controller from STMicroelectronics. We thought that the memory chips were serviced by the pair of uP7701 chips but they are actually powered by the tiny VT242WF chips from Volterra. Technical documentation on this manufacturer’s products is hard to find in public domain, but we guess that a VT242WF combines three components necessary for a switching regulator: a controller, a driver and a power section. This is indicated by the active cooling of these tiny chips.
The EAH5770/2DIS/1GD5 has the same specs as the reference sample. It has a memory frequency of 1200 (4800) MHz. Coupled with the 128-bit bus, this provides a peak memory bandwidth of 76.8GBps. The total amount of memory is 1 gigabyte, which is a standard amount even for such inexpensive solutions as the Radeon HD 5700 series. The GPU is clocked at 850MHz and works in its full configuration with 800 ALUs, 40 TMUs and 16 RBEs. By the way, ATI’s previous-generation single-processor flagship Radeon HD 4890 has the same specs and only surpasses the Radeon HD 5770 in terms of memory bandwidth.
The graphics card has two DVI-I connectors and one DisplayPort. Coupled with the Eyefinity video controller implemented in all fifth-generation Radeons, it means you can connect up to three monitors simultaneously (one monitor has to support DisplayPort or you can use a DisplayPort à DVI-D adapter). There are also CrossFireX connectors on the PCB – we are going to need them today.
The reference Radeon HD 5770’s cooling system is a simplified and light version of the cooler installed on top-performance Radeon and GeForce series cards: a blower is cooling a heatsink connected with heat pipes to a base that contacts with the GPU die. The hot air is exhausted out of the system case through slits in the graphics card’s mounting bracket. The memory chips on the front side of the PCB are cooled while the chips on the reverse side of the PCB are not. This is not a problem as they do not get very hot at work.
The cooler is overall effective and quiet although many graphics card makers come up with more original solutions with the main purpose of singling their product out among identical copies of reference cards. Sometimes original coolers do ensure lower temperatures and noise and facilitate overclocking. The second graphics card is an example of that.
This card’s packaging is far from original, too. The box is vertically oriented and embellished with a picture of a warrior in gaudy armor.
There is little information on the box. Besides the type and amount of graphics memory, the I/O connectors are indicated, but wrongly. The text “Dual DVI” implies two DVI ports, but the card has one port of each type: DVI, HDMI and DisplayPort. There is also a sticker reading “Professional Cooling System”. We will see shortly what is so professional about it. The product is accompanied with the following accessories:
The accessories seem to be scantier than those of the ASUS EAH5770/2DIS/1GD5, but the owner of a PowerColor PCS HD5750 1GB GDDR5 Premium Edition can download a full version of the racing sim Colin McRae: Dirt 2 for free. The game supports DirectX 11 and can vividly demonstrate the unique capabilities of the ATI Radeon HD 5x00 series. PowerColor should be praised for including a popular game with a graphics card costing less than $150.
This graphics card looks interesting than the previous one as it seems to be an original product. However, the red color of the PCB should not mislead you. The card has the standard PCB wiring developed by AMD. The color and the cooler are the only original things about it.
PowerColor PCS HD5750 Premium (left) and ATI Radeon HD 5750 reference (right)
It is easy to see that the PowerColor PCS HD5750 Premium is akin to the reference Radeon HD 5750 even without removing the cooler although an inexperienced user may be confused by the tilted chokes of the voltage regulator. They are installed neatly enough and do not look like a defect of assembly, though. The kinship becomes even more obvious with the coolers removed:
The circuit design is somewhat simplified. PowerColor’s version lacks smaller chokes and some capacitors that were present on the original Radeon HD 5750 but the power circuit is a copy of the reference one. It follows a 3+1 design in which the 3-phase GPU voltage regulator is managed by an uP6201 chip from uPI Semiconductor and the 1-phase memory voltage regulator, lacking one choke, is based on an uP6101 controller.
Like in the senior model of the Radeon HD 5700 series, there is one 6-pin PCIe 1.0 power connector here.
The GPU was manufactured on the 37th week of 2009 and works at a clock rate of 700MHz in full compliance with AMD’s official specs. It is technically an RV830 with some subunits turned off. To be specific, one of its 10 SIMD cores is turned off, with its 80 ALUs and 4 texture processors. Thus, like any Radeon HD 5750, the PowerColor PCS HD5750 Premium has 720 ALUs (grouped into 144 universal shader processors) and 36 texture-mapping units. The number of RBEs is 16, like in the Radeon HD 5770. This card has a memory frequency of 1150 (4600) MHz, providing a peak memory bandwidth of 73.6GBps. Like the senior card of the series, this model has 1 gigabyte of memory. We could not read the marking on the GDDR5 chips as they had small aluminum heatsinks which we dared not tear off as we might damage the card. We guess they are Hynix H5GQ1H24AFR-T2C, the same memory as you can see on the reference Radeon HD 5750 and 5770.
Like the senior model of the Radeon HD 5700 series, the junior one supports up to three monitors simultaneously: two with DVI interface and one with DisplayPort. The card also offers an HDMI connector that supports multi-channel HD audio formats such as Dolby TrueHD and DTS-HD Master Audio. This is only available with the fifth-generation Radeons as yet. No other graphics card, not even Nvidia’s GeForce GT 220/240, offer this functionality. Of course, the card supports CrossFireX and has two appropriate connectors that we are going to use today.
So, notwithstanding its brighter appearance, the PowerColor PCS HD5750 Premium is in fact one more copy of AMD’s reference sample. Its cooling system is a different story, though. PowerColor’s website says the company collaborates with Arctic Cooling, but the cooler installed on this model was developed and manufactured by ZEROtherm, a well-known maker of cooling solutions. A brief search on the ZEROtherm site produced the GX810 model that is the same as the cooler on the PowerColor card except for the shape of the base which contacts with the GPU die and for the lack of the thermal sensor which regulates the fan’s speed depending on the temperature.
This solution is technically very much alike to the Zalman VF900-Cu but has different and smaller ribbing. It also has a blower that blows sideways whereas Zalman’s cooler also blows downwards to cool the card's PCB. And finally, there is only one heat pipe here. Anyway, this ZEROtherm solution is obviously superior to the reference Radeon HD 5750 cooler which is an ordinary aluminum "flower" copying the old coolers of Thermaltake's Orb series. And we expect PowerColor’s cooler to be more effective and, probably, quieter. We’ll check this out right now.
As we have transitioned to a more accurate and progressive method of measuring power consumption of modern graphics cards, we will retest the Radeon HD 5700. The testbed is configured as follows:
The new testbed for measuring electric characteristics of graphics cards is based on the measurement tool developed by Oleg Artamonov and described in his report PC Power Consumption: How Many Watts Do We Need?. The tool facilitates and automates the measurement process.
We used the following benchmarks to load the graphics accelerators:
Except for the maximum load simulation with OCCT, we measured power consumption in each mode for 60 seconds. We limit the run time of OCCT: GPU to 10 seconds to avoid overloading the graphics card's power circuitry.
We got the following data using the new method:
Notwithstanding its abundant resources, the graphics card is very economical in the Windows 7 desktop mode but decoding a VC-1 stream from a Blu-ray disc calls for a serious increase in power consumption. The Radeon HD 5770 should be given credit for not reaching a load of 4 amperes on the external power connector even under the demanding Crysis Warhead. The peak power draw is less than 80 watts. It is only under the synthetic OCCT: GPU load, which cannot occur in real-life applications, that the card needs more than 100 watts.
The junior model of the series behaves in the same manner. But as it has a lower GPU frequency and a few disabled subunits, its power consumption is lower: 30 watts when decoding HD video, less than 60 watts in modern games and only about 70 watts in the synthetic OCCT: GPU test.
* - estimated
The Radeon HD 5700 looks brilliant in terms of power consumption compared to Nvidia's G200b-based products as well as to AMD's previous-generation cards. Even the GeForce GTS 250, based on the outdated G98b core and hardly capable of competing with the Radeon HD 5700 series, consumes much more.
The temperature data are somewhat unexpected. The ASUS EAH5770 is 5°C hotter than the reference card under load, which may be due to the difference in the ambient temperature during the tests. And the advanced ZEROtherm cooler installed on the PowerColor PCS HD5750 1GB GDDR5 Premium Edition is not any better than the reference card’s cooler in 3D mode and even inferior to the latter in 2D mode. Perhaps the contact between the cooler's base and the GPU was poor, yet the cooling efficiency is overall high.
We measured the noise with a noise-level meter Velleman DVM1326. The reference point for our noise measurement tests is 37dBA which is the level of ambient noise in our test lab as measured at a distance of 1 meter from the testbed with a passively cooled graphics card inside. When we installed the tested graphics cards, we got the following results:
The cooling system of the PowerColor PCS HD5750 1GB GDDR5 Premium Edition is louder than the reference one, but we could only hear it at a very short distance. At a distance of 1 meter it was as quiet as the reference cooler of the GeForce GTX 275 and much quieter than the Radeon HD 4890's cooler. The ASUS EAH5770 does not irritate the ear at all with its soft hiss of airflow.
We tried to overclock both cards we used for this review.
ASUS EAH5770 overclocked
PowerColor PCS HD5750 1GB GDDR5 Premium Edition overclocked
PowerColor PCS HD5750 1GB GDDR5 Premium Edition overclocked well enough considering that the Radeon HD 5750 is based on RV830 chips which did not pass the frequency test and/or were found to have defective subunits that made them unsuitable for Radeon HD 5770. It means the Radeon HD 5750 is handicapped when it comes to overclocking, but you can still speed one up if you are lucky with your specific sample.
Asymmetrical CrossFireX up and running
Now, we are going to unite these two cards into a CrossFireX tandem and pit it against a Radeon HD 5870.
We are going to investigate the performance of asymmetric CrossFireX tandem using the following universal testbed:
The graphics card drivers were configured in the following way:
Since GeForce GT 240 graphics cards can be quite fit for contemporary games we used the complete set of gaming benchmarks including the following titles:
First-Person 3D Shooters
Third-Person 3D Shooters
We selected the highest possible level of detail in each game using standard tools provided by the game itself from the gaming menu. The games configuration files weren’t modified in any way, because the ordinary user doesn’t have to know how to do it. We made a few exceptions for selected games if that was necessary. We are going to specifically dwell on each exception like that later on in our article. Unfortunately, this time we had to give up Colin McRae: Dirt 2 auto simulator because the current game version was unstable on multi-GPU systems, which resulted into system hanging when we tried to change the screen resolution.
We ran our tests in the following resolutions: 1280x1024, 1680x1050, 1920x1200 and 2560x1600. Everywhere, where it was possible we added MSAA 4x antialiasing to the standard anisotropic filtering 16x. We enabled antialiasing from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of ATI Catalyst and Nvidia GeForce drivers.
Performance was measured with the games’ own tools and the original demos were recorded if possible. We measured not only the average speed, but also the minimum speed of the cards where possible. Otherwise, the performance was measured manually with Fraps utility version 3.0.2. In the latter case we ran the test three times and took the average of the three for the performance charts.
Our asymmetric tandem shows its worth right from the start, performing as fast as the single Radeon HD 5870. The symmetric Radeon HD 5770 subsystem is but slightly faster and does not offer any advantages. Thus, it does not justify the higher financial investment. The tandem built out of two Radeon HD 5750 cards is not so fast, its bottom speed falling below 30fps at 2560x1600. So, the asymmetric configuration looks the most appealing of all in this game.
The junior symmetric and the asymmetric CrossFireX tandems built out of Radeon HD 5700 series cards are roughly equal to each other in terms of average performance and inferior to the Radeon HD 5870. The Radeon HD 5770 pair looks better, although its frame rate plummets at 2560x1600 for some reason. The other CrossFireX configurations also suffer a performance hit at that resolution, though.
The Radeon HD 5770+5750 tandem is roughly equal to the Radeon HD 5870 here in terms of average frame rate. The Radeon HD 5770 pair is competitive to the GeForce GTX 295 at resolutions up to 1920x1200 but then Nvidia’s card goes ahead, enjoying a large lead at 2560x1600. The tested solutions all deliver about the same level of playing comfort at the lower resolutions, though.
The game’s integrated benchmarking tools are far more accurate than Fraps but cannot report the bottom frame rate.
A single Radeon HD 5770 will suffice for this game but you can unite it with another Radeon HD 5700 series card to get the performance of a Radeon HD 5870 or even higher (if you add in a junior or senior model of the series, respectively). What is important, the power consumption and noise of such a graphics subsystem are going to be within reasonable limits.
We use the highest settings in the new S.T.A.L.K.E.R. but without antialiasing. We turn on the DirectX 10.1 and DirectX 11 modes products that support them.
The symmetric Radeon HD 5770 CrossFireX tandem is the only opponent to the Radeon HD 5870 here. The cheaper asymmetric tandem with a Radeon HD 5750 is a little worse, but we should note that the bottom speed of every tested solution sinks below comfortable as soon as 1680x1050.
The game was tested in OpenGL-based multiplayer mode. Unfortunately, the integrated benchmark cannot report the bottom frame rate.
There is almost no difference between the asymmetric Radeon HD 5700 CrossFireX configuration and the Radeon HD 5870, and neither can compete with the GeForce GTX 295 at resolutions of 1920x1200 and higher. It must be noted that AMD’s solutions are fast enough even at 2560x1600 to ensure comfortable and smooth gameplay.
One Radeon HD 5750 is more than enough here. We guess even a Radeon HD 5670 would make this game playable. However, we can see that the Radeon HD 5750+5770 tandem is somewhat faster than the Radeon HD 5870 while the two Radeon HD 5770 are ahead of the GeForce GTX 295.
The struggle is tough here. The CrossFireX configurations built out of Radeon HD 5700 series cards win at 1280x1024 and 1680x1050 but give way to the Radeon HD 5870 at 1920x1200. And at 2560x1600 the GeForce GTX 295 comes out the winner. The frame rate of each solution is comfortably high at any resolution, though.
There is no difference between the top-end graphics subsystem in this game as they all hit the performance ceiling even at 2560x1600. We’d still prefer the Radeon HD 5870 for playing this game as it supports all modern industry standards and is free from problems associated with multi-GPU technologies.
We use the in-game benchmarking tools that do not allow to measure the bottom frame rate. We also enable DirectX 10.1 mode for graphics cards that support it.
Nvidia’s products are good as usual, but the GeForce GTX 295 is not so far ahead of its AMD rivals at 2560x1600 and AMD’s solutions subjectively ensure the same playing comfort. The asymmetric Radeon HD 5700 tandem copes with it job just as successfully as the single-chip Radeon HD 5870.
We turn on DirectX 11 support for the Radeon HD 5000 series in this game.
First place goes to the GeForce GTX 295. The Radeon HD 5870 is second and the Radeon HD 5770 CrossFireX is only third, following the leaders closely. On the other hand, the CrossFireX platform is better than its opponents in terms of bottom speed, which is important for scenes with large amounts of units.
The CrossFireX configurations based on Radeon HD 5700 series cards are far more confident in the other real-time strategy we use: the most advanced variant even beats the Radeon HD 5870 although cannot overtake the GeForce GTX 295. The bottom speed of AMD’s solutions is no worse than that of Nvidia’s flagship which consumes over 200 watts. Unfortunately, none of the tested graphics subsystems can maintain a frame rate higher than 25-30fps at 2560x1600.
We minimize the CPU’s influence by using the Extreme profile (1920x1200, 4x FSAA and anisotropic filtering). We also publish the results of the individual tests across all display resolutions to provide a full picture.
The GeForce GTX 295 supports PhysX, so AMD’s solutions are handicapped in this test. Still, we can note that the Radeon HD 5700 CrossFireX platforms are not fast even in comparison with the Radeon HD 5870. The Radeon HD 5770 tandem barely touches the 8,000-point mark while the single-chip Radeon HD 5870 takes the barrier easily.
The individual tests confirm our point. The Radeon HD 5770 duo can compete with the single Radeon HD 5870 in the second test only. The cheaper CrossFireX configuration with a Radeon HD 5770 and a Radeon HD 5750 is only competitive at 1280x1024.
This test serves to estimate the computing capabilities of GPUs that can be used for GPGPU applications. We select OpenCL mode which is common for AMD and Nvidia solutions.
At single-precision computations the junior Radeon HD 5700 CrossFireX tandem is almost as fast as the GeForce GTX 295 that has much fewer ALUs whereas the senior CrossFireX configuration is close to the Radeon HD 5870. However, the RV830-based products are hopeless in double-precision mode which is used for serious scientific and engineering computations. They do not have native FP64 support and emulate such computations in a very slow way whereas the RV870 Cypress supports FP64 natively and delivers a highly impressive result. The failure of the GeForce GTX 295 is due to an extremely inefficient implementation of FP64 support in the G200 chip: there is only one FP64 unit per each 30 shader processors.
FP32 mode is quite enough for home use, though. Besides, currently available GPGPU applications are mostly based on Nvidia’s CUDA technology rather than on OpenCL or DirectCompute. The GPGPU concept has not yet fully taken off.
Can you add an expensive Radeon HD 5750 to your Radeon HD 5770 and get the performance of a Radeon HD 5870? Our test session answers in the affirmative. As opposed to Nvidia’s SLI, AMD’s CrossFireX technology works perfectly on such asymmetric multi-GPU tandems as Radeon HD 5770+5750 and we had no stability or compatibility issues whatsoever.
At a resolution of 1280x1024 this asymmetric tandem is an average 2% slower than the Radeon HD 5870, the largest gap of 14% being observed in BattleForge. The Radeon HD 5770+5750 pair is occasionally even faster, enjoying an advantage of nearly 15% in Resident Evil 5.
The overall picture remains the same at the higher resolution. The Radeon HD 5770+5750 tandem never falls behind the Radeon HD 5870 by more than 15%. Its biggest advantage lowers to 7%, though. The CrossFireX tandem beats or equals its single-chip opponent in eight tests.
The total number of draws and wins is six now and the asymmetric CrossFireX tandem falls behind its opponent in Resident Evil 5, yet it still looks appealing, especially as you can hardly buy a Radeon HD 5870, let alone a GeForce GTX 295, for less than $400.
It is in Crysis Warhead that the asymmetric CrossFireX configuration fails noticeably at 2560x1600. In the other games it delivers as much comfort as the premium Radeon HD 5870 and GeForce GTX 295, though.
Thus, the idea of using a Radeon HD 5750 together with a Radeon HD 5770 in CrossFireX mode does make sense. The former card can be found in retail for $130-150. Considering its performance, it can make a good starter for an inexpensive graphics subsystem. And when you need to boost your computer’s graphics performance later on, you can add $170-190 more, buy a Radeon HD 5770 and play games like on a Radeon HD 5870 which would cost you $400. A Radeon HD 5850 would cost about the same money as the asymmetric CrossFireX tandem but would be slower.
This upgrade method can help you save your money and can also be done in two steps, without investing a large sum right away. The only shortcoming of the asymmetric Radeon HD 5770+5750 configuration is that it depends on software support on the side of the driver and game engine, but this is a common feature of all modern multi-GPU solutions.
As for the specific graphics cards from ASUS and PowerColor, they both leave a good impression. The ASUS EAH5770/2DIS/1GD5 is a copy of the reference card from AMD with rather scanty accessories but it showed good overclockability, almost reaching a GPU frequency of 1GB with its default cooling. The PowerColor PCS HD5750 1GB GDDR5 Premium Edition has an advanced cooler, even though slightly noisier than the cooler of the ASUS card, and boasts good overclockability, too (for an HD 5750). So, the PowerColor PCS HD5750 is quite a good option for a starter. If you want to use the described upgrade method later on and add an ASUS EAH5770 or other similar card to it, do not forget that you will need a CrossFireX-compatible mainboard.