by Alexey Stepin , Yaroslav Lyssenko
02/16/2010 | 11:44 AM
The quality and level of detail of 3D graphics are the number one priority of today’s game developers who try to engage the player’s attention with original visual effects even at the expense of the plot. But besides 3D graphics, there are other factors such as sound or a realistic physical model that help immerse the player into the virtual reality. However, even the basic phenomena of our world that we can see each day around us prove to be extremely sophisticated when it comes to modeling and transferring them into the virtual world. Simulating a flow of water or a realistic flying-off of splinters of broken glass calls for a lot of complex mathematical computations and, consequently, processing resources. Most modern games still utilize the central processor for that purpose but game developers have long had a much more powerful computing device at their disposal because each modern graphics card is in fact a set of numerous unified processors capable of working in parallel.
It must be noted that the idea to free the CPU from physics computations and increase the quality of physical effects in games is not new. In 2006 Ageia, which owned a physical effects engine called PhysX, attempted to launch a new class of coprocessors called physics processing units or PPUs. A PhysX chip, manufactured on 130nm tech process, consisted of 125 million transistors and incorporated a general-purpose core controlling an array of SIMD processors. The PhysX was superior to any CPU at simple tasks that required massively parallel computing such as the modeling of a collision of a lot of objects. Theoretically, this provided an opportunity to improve visual effects such as smoke and fire and even invent new, advanced effects like a real-life simulation of liquids and fabrics or a fully destructible environment. We tested one of the first PPUs in our labs (it was the ASUS PhysX P1) but were not impressed much due to the weak support for the device on game developers’ part, even though some special effects were astonishing indeed.
So, it was all up to game developers. Perhaps, the PPU could indeed make it eventually as a standalone and popular class of devices but on the 13th of February 2008 Ageia with all its innovations was bought by Nvidia which, instead of offering the PPU as a separate product, preferred to endow its GPUs with such functionality instead. As a result, the Ageia PhysX remained the first and only discrete physical effects accelerator. Starting from the 15th of August, 2008, it was replaced by Nvidia’s GeForce 8, 9 and 200 series GPUs. Ageia’s original accelerators are still supported by Nvidia in Windows XP and Vista but not in Windows 7. Anyway, Nvidia’s GPUs are currently the only class of devices that can officially accelerate physical effects in games.
It should be noted that the type of computations required to process physical effects in video games is an ideal match for the architecture of modern GPUs from both ATI and Nvidia. There is only one problem. If one and the same computing core is used for both 3D graphics and physics computations, these tasks will conflict, competing for the GPU’s resources and the GPU may prove to be unable to deliver a high enough frame rate in a 3D scene that is graphically complex and also features advanced physical effects. This problem can be solved by installing a second graphics card into the computer and using it as a PhysX accelerator; Nvidia’s drivers offer this opportunity. But it is not always possible or desirable to install a second card whereas the dual-processor GeForce GTX 295, besides its low availability, has such shortcomings as a high price and a high level of power consumption. And what if an advanced and complex GPU is accompanied with a simple and cheap one, serving as a PPU? That must be the reasoning behind EVGA’s decision to develop and market a unique graphics card that comes with a dedicated PhysX accelerator core. We will test the EVGA GeForce GTX 275 CO-OP PhysX Edition in this review and tell you all about all its highs and lows.
But we mustn’t forget that Nvidia is currently in the position of a technological outsider. In terms of pure gaming performance AMD’s Radeon HD 5800 series, especially the ATI Radeon HD 5850 model, look preferable at a comparable price. But as that model does not support PhysX, it may be hopelessly weak in games that make use of hardware physics effects acceleration, notwithstanding its tremendous computing potential. Nvidia keeps AMD away from the secrets and beauties of PhysX although AMD’s GPUs could be used as physics processing units just as successfully as their GeForce counterparts. This is no good for the PhysX technology itself because, being exclusive, it cannot become really widespread. Game developers just won’t bother to implement innovations that will only be enjoyed by half of the potential audience. We can’t influence Nvidia, of course. But we can go another way and check out if we can endow AMD’s solutions with PhysX support by adding a PPU card from Nvidia. We will also compare the resulting tandem with the EVGA GeForce GTX 275 CO-OP PhysX Edition. Let’s first take a look at the cards we are going to use in this test session.
EVGA hasn’t revised its product packaging for a long time already, and that’s good as this creates a feeling of consistency. The packaging design itself is neat and elegant: a black box with a colorful stripe at the top.
We have been unable to grasp EVGA’s rules for choosing the color of the stripe for the particular product. For example, boxes of different versions of GeForce GTX 285 may have either red or an orange stripe. It looks like dark-red is reserved for exclusive or dual-processor models, though. All versions of EVGA GeForce GTX 275 use a yellow-orange stripe and the CO-OP PhysX is no exception, although it is a highly unusual GTX 275. The packaging quality is high as is typical of EVGA products. It will protect the card from various incidents that may occur on its way to the shop shelf.
Besides the card, the box contains the following accessories:
As you can see, there is everything you need to use the card. There is a nice bonus. You can download the stealth shooter Batman: Arkham Asylum using the Stream distribution system. With the current speeds of Internet connections, downloading the game shouldn’t be a problem. The choice of the title is not accidental: the game uses a PhysX engine for creating physical effects. Theoretically, the dedicated PhysX/GPGPU processor on board the EVGA GeForce GTX 275 CO-OP PhysX Edition should boost the card’s performance in Batman: Arkham Asylum as well as in other PhysX-enabled games. We’ll check this out in our practical tests soon.
Summing it up, the EVGA GeForce GTX 275 CO-OP PhysX Edition is packaged very well and there’s nothing we can find fault with, especially considering that this product, for all its uniqueness, is not a top-speed luxurious solution. EVGA values it at $350, which is the price quoted on the product page of the company’s website.
The EVGA GeForce GTX 275 CO-OP PhysX Edition is almost the same in face view as the single-card GeForce GTX 295 we described in an older review, but there are a few differences at the back:
EVGA GeForce GTX 275 CO-OP PhysX (left)
and Inno3D GeForce GTX 295 Platinum (right)
However, you can still see that the PCB design of the GeForce GTX 275 CO-OP PhysX is largely copied from the reference PCB of the GeForce GTX 295. This practical approach must have saved EVGA some money on developing a completely new PCB. We didn’t have any problems dismantling the cooling system except that we had to unfasten a lot of screws. With the cooler removed, it became clear that the EVGA card and the single-card version of GeForce GTX 295 do differ although EVGA’s engineers had tried to utilize as much of the existing design as possible.
EVGA GeForce GTX 275 CO-OP PhysX (left)
and Inno3D GeForce GTX 295 Platinum (right)
EVGA GeForce GTX 275 CO-OP PhysX (back) without the cooler
It is only in the front part of the card, where the place of a G200b is occupied by a G92b chip that serves as a PhysX/GPGPU accelerator, that a completely new wiring design is used. The power subsystem is almost the same as on the GeForce GTX 295: each processor is powered by an independent 3-phase regulator based on Renesas R2J20651 packs managed by an ADP3193A controller from Analog Devices.
The memory power subsystem has changed somewhat: a Renesas R2J20651 with an unknown chip marked as N12 VGF932 are responsible for the memory belonging to the main GPU.
The memory chips related to the G92b coprocessor are powered by a regulator based on ordinary MOSFETs and managed by an uP6161 chip from uPI Semiconductor.
The rest of the card’s components are powered by regulators based on uP6161 and APW7142 chips.
Like the GeForce GTX 295, the EVGA GeForce GTX 275 CO-OP PhysX has two power connectors, one of which is of the 8-pin variety. We guess two 6-pin connectors would be quite enough, however, because the G92b chip has a much lower power draw than the main G200b. The card won’t start up if you connect a 6-pin plug into its 8-pin connector, reporting a power problem with the LED indicator on its mounting plate, unless you bypass this protection.
There is only one NVIO2 chip here because the G92b contains all the necessary logic inside. Moreover, it does not need this video output logic since its purpose is limited to PhysX and CUDA applications. An nForce 200 chip is responsible for switching the PCI Express bus. It is marked differently than on the GeForce GTX 295: “NF200-SLI-A3” as opposed to “NF200-P-SLI-A3”.
The main core, which works as a graphics processor, is marked as G200-105-B3. The second number in the marking means that this model is manufactured out of the same chips as the ordinary versions of GeForce GTX 275 whereas the GeForce GTX 295 uses chips marked as G200-400/401-B3. Our sample was manufactured on the 41st week of 2009 and is clocked at 633/1296MHz. Thus, its shader domain frequency is lowered for some reason (it is 1404MHz with the ordinary GeForce GTX 275). This means that the GeForce GTX 275 CO-OP PhysX is going to be somewhat slower in games despite its physics processor.
The GPU’s memory bank has the same specs as on the GeForce GTX 275: 896 MB, 448 bits, 1134 (2268) MHz. The peak memory bandwidth is 127 GBps. The EVGA GeForce GTX 275 CO-OP PhysX uses GDDR3 chips from Hynix (H5RS5223CFR-N2C, 512 Mb) with a rated frequency of 1200 (2400) MHz, so there is some overclocking headroom left.
The dedicated PPU installed on this card is an ordinary G92b chip manufactured on 55nm tech process. Our sample was made on the 37th week of 2009 and is marked as G92-421-B1. It is clocked at the same frequencies as the core of the ordinary GeForce GTS 250, i.e. 738/1836 MHz. If the engineers had any apprehensions regarding the heat dissipation of the card, they should have reduced the clock rates of the coprocessor rather than of the main GPU, which is responsible for 3D rendering. This would help avoid the performance hit in games. We don’t know why the engineers preferred to slow down the main GPU but we will measure the performance hit in our tests.
Interestingly, one RBE section is disabled in the coprocessor due to the reduced amount of memory: 384 MB instead of 512 MB. The memory bus has become narrower. This memory bank consists of Hynix H5RS5223CFR-N2C chips, too. With a 192-bit bus and a clock rate of 1100 (2200) MHz, the memory bandwidth is 52.8GBps. EVGA’s engineers must have thought this enough for the coprocessor that does not have to do any 3D rendering and is limited to PhysX computations. Considering that the inexpensive GeForce GT 220 is also promoted as a physics effects accelerator, the GeForce GTX 275 CO-OP PhysX seems to have more than enough resources to accelerate physics.
As for interfaces, the card is equipped with two dual-link DVI-I ports (which can transfer S/PDIF audio over HDMI) and two MIO connectors for SLI technology. Thus, you can add two ordinary GeForce GTX 275 cards to this one to build a 3-way SLI subsystem in which three G200b processors are used for 3D rendering and a dedicated G92b processor with its own memory is responsible for PhysX computations, without utilizing any resources of the main GPUs. This looks appealing in theory, but practically, there are faster solutions available for demanding gamers, for example the Radeon HD 5970. Besides, our EVGA GeForce GTX 275 CO-OP PhysX refused to work together with a GeForce GTX 275 if installed into the mainboard’s top PCIe x16 slot. It is only when we installed it into the bottom slot (and the ordinary GeForce GTX 275 into the top one) that we could make this SLI configuration work.
The cooling system of the EVGA GeForce GTX 275 CO-OP PhysX is almost a copy of the one installed on the single-card GeForce GTX 295 and consists of two individual heatsinks each of which cools its own core.
This cooling system is usually quiet and effective. Its only drawback is that the hot air is not exhausted out of the computer case but stays inside. Here, this drawback is aggravated by the fact that the main G200b core, being much hotter than the simpler G92b chip, is located at the back of the card. That is, it is going to heat up the air inside the computer. Why didn’t the engineers put the G200b at the front to ensure better cooling for it? We don’t have an answer. Well, the EVGA GeForce GTX 275 CO-OP PhysX is indeed an odd and original solution and we know no other product similar to it. Now we just need to check out the practical tests. But before that, let’s take a closer look at its competitor – XFX Radeon HD 5850 Black Edition.
XFX often ships its products in original packaging like the X-shaped box we once described to you. Perhaps this packaging is too extravagant to be practical, but it surely fulfils the purpose of catching the customer’s eye. This time the box is unusual, too:
It is narrow, long and rather compact, so you shouldn’t have any problems carrying it home. The design is not as minimalistic as that of the EVGA card’s box but looks demure and neat. The font of the Black Edition label reminds us of the disco 80’s. The text on the box tells you the type and amount of graphics memory and that a copy of the racing sim Colin McRae: Dirt 2 is included with the card. The fact that the card’s frequencies are pre-overclocked is only indicated by the Black Edition label. Inside, there is another box made from thick black cardboard with XFX logo. And in this box, there is a flat one with accessories. And it is at the very bottom that you will find your XFX Radeon HD 5850 Black Edition wrapped into an antistatic pack.
The product comes with these accessories:
So, the XFX Radeon HD 5850 Black Edition is packed up just as well as the EVGA GeForce GTX 275 CO-OP PhysX Edition. A copy of a modern game is included into this kit, too. There is one more common thing between the two products: the free game is selected in such a way as to emphasize the advanced features of the respective card, namely the EVGA’s PhysX and the XFX’s DirectX 11. But again, Nvidia’s PhysX is deliberately exclusive without any fundamental technical reasons for that whereas the DirectX 11 support in the new generation of ATI Radeon HD processors is indeed unique and only due to their more progressive architecture.
Summing it up, the packaging and accessories of the XFX Radeon HD 5850 Black Edition deserve as high a mark as those of its opponent from EVGA. XFX’s product has the advantage of price: the Black Edition Radeon HD 5850 can be found selling at less than $350 whereas the mentioned GeForce GTX 275 version from EVGA will cost you $349 or more. Thus, the XFX Radeon HD 5850 Black Edition has only one drawback. It lacks PhysX support. Perhaps we can get rid of this drawback. Let’s first see how the card is designed, though.
This version of Radeon HD 5850 is based on the reference PCB design and uses a reference cooler developed by AMD’s graphics department. As a matter of fact, this is true for all other Radeon HD 5800 series cards available today. We described this design in detail in our article called "Hidden Threat": ATI Radeon HD 5850 Graphics Card Review. The only distinguishing feature of the XFX product is the sticker on the cooler’s casing. The PCB itself is almost 4 centimeters shorter than the Radeon HD 5870’s one as we already wrote in an earlier review.
XFX Radeon HD 5850 Black Edition (left) and ATI Radeon HD 5870 reference (right)
So, you shouldn’t worry that this card may be too long for your system case. The XFX Radeon HD 5850 Black Edition can be easily installed into any computer enclosure, save for the most compact models. The power circuit follows a 3+2 design. Thus, the GPU power section has one phase less than on the Radeon HD 5870 but this should be more than enough for the RV870 core considering the Radeon HD 5850 specs.
The GPU voltage regulator is based on a Volterra VT1165MF controller that provides the option of software voltage control. Theoretically, this should help overclockers avoid volt-modding the card, which is a dangerous thing to do. But practically, the most reckless of overclockers still combine both methods to squeeze some more megahertz and a few hundred 3DMark points from the poor graphics card. Such experiments often end disastrously, but the risk cannot stop an inveterate overclocker!
As opposed to the Radeon HD 5870, the Radeon HD 5850 has only one VT1156MF controller. Additionally, there is a VT237WF chip that combines the control and power sections of the voltage regulator which seems to be responsible for the RV870’s memory controller. The memory chips are powered by the pair of tiny chips from the same maker marked as VT243WF. The configuration and placement of the power connectors is different, too. Here, we have two 6-pin connectors installed at the shorter edge of the PCB whereas the Radeon HD 5870 has them at the top, longer edge. This may make it easier to connect the PSU cables if the card is already installed into the system case.
There are eight Samsung K4G10325FE-HC04 chips on the PCB, the same as you can find on the Radeon HD 5870. These GDDR5 chips have a capacity of 1 Gb (32 Mb x 32), a voltage of 1.5V, and a rated frequency of 1250 (5000) MHz. The Radeon HD 5850 has a specified frequency of 1250 (4500) MHz which raises its peak bandwidth to 144 GBps. The memory bus is 256 bits wide and the total amount is a standard 1024 megabytes.
The GPU marking says that this sample of RV870 was manufactured on the 39th week of the last year. As usual, there is no other practically useful information in it. Two out of the 20 SIMD arrays available in the RV870 are turned off on the Radeon HD 5850, so the total amount of active ALUs and TMUs is 1440 and 72 as opposed to the Radeon HD 5870’s 1600 and 80. Like the memory chips, the GPU of the XFX card is pre-overclocked from the reference frequency of 725 MHz to 765 MHz, which is a promise of a nice performance boost in games. But considering that Radeon HD 5850 cards use those RV870 samples that did not pass the frequency test and/or have defective subunits, we don’t expect the card to be able to overclock much above its default clock rate.
Like the senior cousin Radeon HD 5870, the XFX Radeon HD 5850 Black Edition can output video content to three monitors simultaneously but one of them has to have a DisplayPort. Besides two DVI-I connectors and one DisplayPort, the card has a native HDMI port, so you don’t have to use an adapter to connect a plasma or LCD TV-set. The card can output audio over HDMI in various formats including multi-channel HD formats, which is so far the unique feature of the ATI Radeon HD 5000 series. The two CrossFireX connectors allow uniting up to four cards in multi-GPU mode, but there are few mainboards that would allow installing as many as four dual-slot graphics cards simultaneously.
As for the cooling system, the XFX is equipped with the Radeon HD 5850’s reference cooler which is a smaller and slightly revised version of the Radeon HD 5870’s cooler.
The heatsink is smaller and has only two heat pipes instead of the four pipes in the Radeon HD 5870’s cooler. Besides, the heatsink is not part of the base and is not physically connected to it. A layer of dark-gray thermal grease ensures proper thermal contact between the heatsink’s copper base and the GPU die. The rest of the components that require cooling contact with the aluminum frame through two types of thermal pads: 3-layer elastic sandwiches with a filling of gel-like white grease for the memory chips and dry, brittle green pads for the power circuit elements.
The heatsink is cooled by a FD9238H12S blower from NTK (HK) Limited. The 12V fan has a power consumption of 0.8A, which is quite a lot. The fan is very noisy at its maximum speed, but it may only reach that maximum in the hands of extreme overclockers. In ordinary mode, its speed is PWM-regulated automatically and the card is overall not very loud. Like with the Radeon HD 5870, the partitions in the cooler’s casing turn a part of the air flow by 90 degrees to drive it through the slits in the card’s side and into the system case. The rest of the hot air is exhausted outside through the slits in the card’s mounting bracket. The latter is populated so densely that this compromise is unavoidable.
Overall, the cooling system of the XFX Radeon HD 5850 Black Edition uses a reliable and time-tested design that has proved effective on earlier graphics cards from both AMD and Nvidia. We have no worries that it may not cope with its job or be too loud, especially as we covered this issue in a previous review. However, the XFX Radeon HD 5850 Black Edition is pre-overclocked by its manufacturer, so its GPU and memory chips produce more heat. Therefore, we should check out how well this cooler works under these harsher conditions.
It is the first time that we have ever had an asymmetric dual-processor graphics card and we want to know how much power it would need in different applications. We measured its power draw on a testbed configured like follows:
The new testbed for measuring electric characteristics of graphics cards is uses a card designed by one of our engineers, Oleg Artamonov, and described in his article PC Power Consumption: How Many Watts Do We Need?. This device 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.
Here are the obtained results for EVGA and XFX solutions:
The EVGA GeForce GTX 275 CO-OP PhysX Edition is far from economical as it needs two times as much as an ordinary GeForce GTX 275 even in idle mode. Both external power lines are loaded in the same measure, which means that both cores, the main G200b and the auxiliary G92b, are active even though the latter does not have any work to do in this mode. The G92b is identified by the OS as the GPU of a full-featured GeForce GTS 250 and thus receives the full ration of juice with the ensuing consequences.
Only the main core is working during video playback while the current on the power line of the auxiliary core remains the same. As a result, the peak power consumption is over 100 watts in this mode, which is not good, especially compared to AMD/ATI’s solutions. In the gaming mode the peak power draw of the EVGA GeForce GTX 275 CO-OP PhysX is the same as that of the reference GeForce GTX 275 because Crysis Warhead does not support PhysX. The 6-pin connector is loaded heavily, however. It means that this connector is not only limited to powering the auxiliary GPU. As for OCCT: GPU, the EVGA card has a lower power draw in that test than in Crysis Warhead.
As opposed to the EVGA card, the XFX Radeon HD 5850 Black Edition makes use of the advanced power-saving features implemented in the ATI Radeon HD 5000 series. Yes, it is less economical in the idle mode than the Radeon HD 5870 but the difference is a mere 1 watt. Its peak power consumption is 17 watts, which is much better than the EVGA card’s 52 watts. The red team wins in the HD video playback mode where XFX is two times as economical as its opponent. The difference in the gaming mode is obvious: 122 against 219 watts. We should note that the XFX has a power draw of over 150 W under the synthetic OCCT: GPU test, but the EVGA card consumes more than 200 W then, anyway. That’s the effect of the outdated technologies employed in the latter.
The top power connector is always loaded more than the bottom one and its peak load is 90 W, which is higher than the recommended maximum of 75 W. Such a high load is unrealistic, though, whereas the power draw of 57.6 W we observe in Crysis Warhead is within the required limits. So, there is nothing to worry about. Every high-quality power supply should cope with the XFX Radeon HD 5850 Black Edition. It does not even have to have a very high wattage rating. Stable output voltages are more important.
The EVGA card is overall just as good as an ordinary GeForce GTX 275 in terms of power consumption, but only when it comes to gaming. It is less economical under low loads. The GeForce GTX 275 CO-OP PhysX is not a good choice if you care about how much power your computer consumes when playing video or working in Windows applications. The old Radeon HD 4890 looks more power-efficient while the new Radeon HD 5770 leaves no chance to the EVGA card when it comes to power consumption.
The XFX Radeon HD 5850 Black Edition looks better in this test. Its consumption is never higher than 170 W even under an extremely high synthetic load. Under real-life loads it will consume no more than 130 watts. And its power draw in 2D mode is very low, too.
The temperatures are all right. The main GPU of the EVGA card is no higher than 75°C under load while the auxiliary GPU is not hot at all. So, EVGA may have been right in picking up the GeForce GTX 295 cooler for this card. The XFX Radeon HD 5850 Black Edition is somewhat disappointing. Its GPU is as hot as 80°C under load. This result is due to the pre-overclocked frequency and the smaller heatsink compared to the Radeon HD 5870 cooler.
And here are the results of our noise measurements.
Alas, the acceptable temperature of the EVGA card is achieved at the expense of silence, especially in 3D mode. At a background noise of 37dBA, the noise-level meter reported 49dBA at a distance of 1 meter from the working testbed! The card is even louder than a Radeon HD 4890. The rather low level of noise in 2D mode cannot save the day because the main purpose of this graphics card is gaming. The XFX is quiet in ordinary mode but becomes audible under load. It remains much quieter than its opponent, though. The difference between them is 3 dBA (at a distance of 1 meter from the testbed) which means a twofold difference in sound pressure because the decibel is a logarithmic unit. Thus, the XFX wins one more test.
As for overclocking, you can overclock both the main GPU and the physics coprocessor of the EVGA GeForce GTX 275 CO-OP PhysX. The memory banks can be overclocked individually or together, too. We managed to make our card work at frequencies of 700/1515 MHz for the GPU and 1200 (2400) MHz for the memory; the auxiliary subsystem was stable at 750/2000 MHz for the processor and 1150 (2300) MHz for the memory.
This is a rather good result but we guess that overclocking the PPU can hardly be rewarding in real-life games. We have already tested an overclocked GeForce GTX 275 in earlier reviews, so we won’t benchmark the EVGA card at the overclocked frequencies. Instead, we will pit it against other graphics subsystems that support hardware PhysX acceleration.
It was harder to overclock the XFX card. We quickly reached the top GPU frequency available in the Catalyst Control Center, raising it from the default 765 MHz to 775 MHz and there was no headroom for memory overclocking at all. Other tools refused to work with the card. It is only with the AMD GPU Clock Tool that we managed to boost the GPU frequency to 900 MHz and the memory frequency to 1200 (4800) MHz.
This is a very good result, considering that the Radeon HD 5850 is made out of lower-grade RV870 chips but you should not expect every other sample of the XFX Radeon HD 5850 Black Edition to be that good at overclocking. Now, let’s move on to gaming tests.
We are going to investigate the gaming performance of EVGA GeForce GTX 275 CO-OP PhysX Edition and XFX Radeon HD 5850 Black Edition using the following universal testbed:
The graphics card drivers were configured in the following way:
Below is the list of games and test applications we used during this test session:
First-Person 3D Shooters
Third-Person 3D Shooters
Semi-synthetic and synthetic Benchmarks
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. This time we tried to include as many gaming titles as possible that use PhysX engine and support hardware acceleration of physical effects.
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.
Since the primary goal of this review was to determine the efficiency of discrete PhysX accelerators, we decided to compare EVGA GeForce GTX 275 CO-OP PhysX and XFX Radeon HD 5850 Black Edition against the following solutions:
Unfortunately, we had to give up the interesting idea of checking out a combination of GeForce GTX 275 + Asus PhysX P1: trying to do whatever it takes to promote PhysX acceleration by their GPUs, Nvidia decided to bury discrete PPUs built on Ageia chips. Although these cards are still supports for Windows XP and Windows Vista operating systems, there are no plans to introduce official support for Windows 7 and this is exactly the OS that has already become a de facto standard among gamers. With a few tricks you can install the drivers and enable the original PhysX accelerator in Windows 7, but this approach is not that obvious and absolutely unintuitive, although you can easily Google it. We could recommend resorting to it only if you have already bought an Ageia based physics accelerator and don’t want your money to go to waste, despite Nvidia’s insistent suggestions to buy another card. Unfortunately, even if you succeed, no one can guarantee stability of this card combination under Windows 7.
Users don’t have too many original Ageia PhysX accelerators, but the option of combining a Radeon HD 5970, 5870 or 5850 with an inexpensive card from Nvidia and use the latter as a physics effects accelerator is far more urgent. At the current moment, Radeon HD 5900 and 5800 series cards are far superior to their opponents in technical specs and performance, hardware PhysX acceleration being their only downside. And as we’ve already said, there are no fundamental obstacles to adding this feature to AMD’s cards except for Nvidia’s stubborn desire to keep the PhysX technology exclusive.
Enthusiasts have found a way to bypass the protection implemented by Nvidia on the level of the GeForce driver. The method is rather simple. You only have to install an unofficial PhysX mod patch that comes in both 32-bit and 64-bit versions. The latest version (1.02) is for all systems and can be downloaded from here.
Here is the full instruction for installing and enabling PhysX support for ATI Radeon HD cards:
Although these instructions may look simple, the installation process is not trivial and may not work at first attempt. Besides, our configuration with an XFX Radeon HD 5850 Black Edition and a GeForce GT 220 (as a PhysX accelerator) was very unstable. For example, it would disable PhysX acceleration automatically when the game changed its display resolution or prevented us from using the ATI Catalyst Control Center settings. Despite all these difficulties, we were still able to benchmark this unusual tandem in every game from our list. Let’s take a look at the results.
The game does not use PhysX, so all GeForce GTX 275 based solutions behave in the same manner here. The dual-processor card from EVGA has a somewhat lower result due to the reduced shader domain frequency, but the difference is practically negligible. All of the Nvidia cards deliver a playable frame rate at resolutions up to 2560x1600 whereas the two junior Radeon HD models cannot do that at the highest resolution. The XFX Radeon HD 5850 is beyond competition and shows a blistering speed even at 2560x1600.
The CryEngine 2 is known for its advanced shader effects and, consequently, for its high GPU requirements, but it does not use PhysX. Therefore, if you disable the appropriate option in the driver, you get a small performance gain because the freed GPU resources are utilized for 3D rendering. The game is not playable on most of the graphics cards as their bottom speed is too low with 4x FSAA even at 1280x1024. The XFX is unrivalled again. Comparable in price to the EVGA, its performance is high enough for comfortable play at 1280x1024.
This game was tested using its integrated benchmark at the highest graphics quality and physical effects settings.
Here, hardware PhysX acceleration determines the very possibility of playing Cryostasis because the game slows down to a slideshow without it. It is thanks to the dedicated PPU that the EVGA card delivers a playable speed at 1280x1024. The ordinary GeForce GTX 275 cannot do that as it lacks computing resources, some of which are busy processing physical effects. Its bottom speed is below playable.
However, as soon as you add a cheap GeForce GT 220 into the system and assign it the job of PhysX acceleration, the graphics subsystem speeds up, ensuring a comfortable frame rate even at 1920x1200! Despite the humble specs of the GT216 core, this hybrid solution is preferable to the EVGA GeForce GTX 275 CO-OP PhysX, even though it requires a second PCI Express x16 slot. There is a lot of mainboards with two graphics slots nowadays. The XFX Radeon HD 5850 Black Edition + GeForce GT 220 tandem looks even better, outperforming the mentioned configuration at every resolution. However, it cannot cope with 2560x1600, either.
The game does not allow using advanced physical effects unless the computer has PhysX acceleration enabled. Therefore we only tested graphics subsystems with PhysX support. We used the PhysX High mode.
The single GeForce GTX 275 turns in the worst result as it has to do both 3D rendering and physics processing at once. The configurations with a dedicated PPU show better performance. The EVGA card, although ahead of the GeForce GTX 275 + GT 220 subsystem in terms of bottom speed, is inferior to the latter in average performance, especially at 1920x1200. Perhaps this is due to the larger amount of graphics memory available to the PPU in the GTX 275 + GT 220 configuration. We guess the latter subsystem is more appealing even though it occupies two PCI Express x16 slots. Note also that neither system can deliver a comfortable frame rate at 2560x1600. The hybrid solution with the XFX Radeon HD 5850 Black Edition as the main graphics card is very, very fast. Its performance remains at about the same level irrespective of the display resolution. It’s a shame that this tandem cannot boast high stability. Otherwise, it would be just perfect.
We used the integrated benchmark at highest settings of graphics quality and physical effects.
The picture looks similar to what we saw in Cryostasis. The dedicated PPU ensures a hefty performance gain that may make the difference between whether the game is comfortable to play or not. When PhysX acceleration is turned off, the game is reduced to a slideshow. For some reason, the use of a standalone GeForce GT 220 card as a physics accelerator does not improve the performance much except for 2560x1600 where the GeForce GTX 275 + GT 220 overtake the EVGA solution. This behavior is not what we see in other PhysX-supporting games. Perhaps Darkest of Days does not utilize the PPU’s resources effectively or the EVGA GeForce GTX 275 CO-OP PhysX is faster due to the higher frequency of its PPU's computing subunits.
Users of AMD cards have to forget about this game altogether or use the PhysX mod patch. Although less stable, the hybrid tandem with the XFX Radeon HD 5850 Black Edition delivers an acceptable frame rate, although cannot overtake the EVGA GeForce GTX 275 CO-OP PhysX. We are still wondering at Nvidia’s stubborn desire to keep PhysX all to themselves. Being exclusive, the technology can never really take off.
The EVGA GeForce GTX 275 CO-OP PhysX has lower bottom speed than the ordinary GeForce GTX 275.
Like in the majority of games, the PhysX coprocessor is not needed here as is indicated by the results of the hybrid tandem built out of the XFX and the GeForce GT 220 cards. The reduction of the shader domain frequency affects the performance of the EVGA card even though its frame rate is high even at 2560x1600.
The game’s integrated benchmarking tools are far more accurate than Fraps but cannot report the bottom frame rate.
The EVGA is just a tiny bit slower than the ordinary GeForce GTX 275. This has no effect on the gamer’s experience.
The frame rates are high at 2560x1600. Even the Radeon HD 5770 with its 128-bit memory bus copes with that resolution.
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 G92b core on board the EVGA GeForce GTX 275 CO-OP PhysX doesn't do anything good in this game whereas the reduced frequency of the main GPU's shader domain leads to a performance hit. We guess the developer should have indeed reduced the coprocessor’s frequency and keep the main GPU’s clock rates intact.
The average frame rate of the Radeon HD 5800 series cards is high but their bottom speed is only comfortable at 1280x1024. At 1680x1050 the gameplay may be jerky in graphically complex and action-heavy scenes.
The game was tested in OpenGL-based multiplayer mode. Unfortunately, the integrated benchmark cannot report the bottom frame rate.
Like in the previous test, the EVGA is a little slower than the reference card at 1024x768 and higher. The gap is negligible, though, and the game’s system requirements are not harsh. Therefore, you won’t feel any difference even at 2560x1600. Both Radeon HD 5800 solutions are barely standing out in this test.
We tested the game using its integrated benchmark and highest graphics/physics settings.
It is in Batman: Arkham Asylum that PhysX effects are at their best. They raise the game’s visuals to a whole new level, so a dedicated PPU is indeed justifiable here. The best result is delivered by the EVGA GeForce GTX 275 CO-OP PhysX but the modest GeForce GT 220 with an ordinary GeForce GTX 275 follows the leader slowly. The latter configuration is inferior in bottom speed but allows playing at 2560x1600 anyway. Moreover, it is quieter than the EVGA card. Users of Radeon HD cards have to think about switching their platform or enabling PhysX in unofficial ways. If you succeed, the XFX Radeon HD 5850 Black Edition can deliver as high a speed as the GeForce GTX 275 based tandem at resolutions up to 1920x1200. But the hybrid tandem slows down suddenly at 2560x1600 for some reason.
The game lacks a PhysX engine for physical effects, so the PhysX accelerator is useless again, but we can see the negative effect of the reduced shader domain frequency on the performance of the EVGA card. This effect isn’t too serious, though. You can recall the price factor, though. You can spend a comparable sum of money for a Radeon HD 5850 and get much higher performance, even though it is not really called for in this game.
Here is one more PhysX-free game and it shows us an already familiar picture. While a second GeForce GTX 295 would help increase the card’s performance through SLI technology, the G92b coprocessor on board the EVGA GeForce GTX 275 CO-OP PhysX just is just wasting away electric power doing nothing. It would make more sense to spend the same amount of money or even less on one of Radeon HD 5850, such as the XFX solution, for instance.
For the reasons we’ve repeatedly stated above, the EVGA GeForce GTX 275 CO-OP PhysX is somewhat slower than the other G200b-based solutions in this comparison. On the other hand, its performance is high enough for a comfortable play at any resolution including the rather exotic 2560x1600. We wouldn’t credit Nvidia or EVGA for it, though. It is rather the game developer’s merit that Fallout 3 has such modest system requirements and runs at a playable speed even on the humble Radeon HD 5770 (which also needs far less power than its more advanced opponents, by the way).
We enable DirectX 11 mode for graphics cards that support it.
The game has an advanced physics engine but it has nothing to do with PhysX and cannot utilize hardware acceleration. Thus, the EVGA card has no advantage in this game while its reduced shader domain frequency makes it slower than the ordinary GeForce GTX 275: the gap is 15% at 2560x1600. The rather low results of the Radeon HD 5800 series shouldn’t be regarded as a failure. Their speed is high for playing at 2560x1600 and they offer the visually appealing DirectX 11 mode which is not supported on Nvidia’s cards.
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.
The EVGA GeForce GTX 275 CO-OP PhysX falls quite far behind the ordinary GeForce GTX 275 with reference frequencies. The gap is as large as 15% at 2560x1600. However, their bottom speeds are identical at the highest resolution and the frame rates are overall high enough for comfortable play. The Radeon HD 5800 series cards are more appealing at 2560x1600 than the EVGA card which is comparable to the Radeon HD 5850 in price. In other words, the XFX Radeon HD 5850 Black Edition will be a better choice for a 30-inch monitor than the EVGA GeForce GTX 275 CO-OP PhysX.
We turn on DirectX 11 support for the Radeon HD 5000 series in this game.
Like in any other PhysX-free game, the physics processing unit is useless here. There is no visible effect from turning PhysX acceleration on and off with the single GeForce GTX 275, either. The EVGA cannot compete with the XFX card which has a much higher average frame rate. If it were not for the low bottom speed, the resolution of 2560x1600 would even be playable.
Like all games without PhysX support, World in Conflict is indifferent to the availability of a PPU in the computer. The single GeForce GTX 275 shows a tiny growth of average performance at low resolutions when PhysX acceleration is off, but this may just as well be due to some measurement inaccuracies. The XFX Radeon HD 5850 Black Edition is but slightly better than the EVGA GeForce GTX 275 CO-OP PhysX.
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 the full picture.
To our amazement, 3DMark Vantage is virtually indifferent to the availability of a dedicated PPU in the system. The only effect we could observe is that the single GeForce GTX 275 had a higher result in the GPU test when we forced its PhysX support off. We did not spot anything interesting in the individual tests, either. Even the reduced shader domain frequency of the EVGA card had but a very little influence on its performance in this popular benchmarking suite. The hybrid tandem made of XFX Radeon HD 5850 Black Edition and GeForce GT 220 behaved almost the same way here.
Our tests of the unique dual-processor graphics card from EVGA prove that a dedicated PhysX accelerator is indeed useful and even necessary for comfortable play, but only if the game itself uses the appropriate physics engine. Otherwise, the additional G92b core installed on the EVGA card won’t do anything and won’t have any effect on the card’s performance. As opposed to the GeForce GTX 295, the G92b cannot be used in SLI mode together with the main G200b core.
As we can see in the summary diagrams, the EVGA GeForce GTX 275 CO-OP PhysX is expectedly inferior to Nvidia’s reference card in PhysX-free games due to the reduced frequency of the shader domain. We don’t know why the EVGA card’s shader domain clock rate is lowered. Fortunately, the difference is very small and has no effect on the level of gamer’s comfort. In other words, wherever the ordinary GeForce GTX 275 ensures smooth gameplay, the EVGA GeForce GTX 275 CO-OP PhysX does the same, too. As for the comparison between the unique EVGA solution and Radeon HD 5850, nothing like that is even remotely possible, because the latter is considerably faster in most contemporary games.
There are, however, games (Cryostasis, Dark Void, Darkest of Days and Batman: Arkham Asylum) in which the EVGA card did very well because the auxiliary G92b processor took up all the processing of physical effects, freeing the main GPU’s resources for rendering the 3D scene. The performance gain varies from a modest 13% to an impressive 70%, depending on the game and display resolution. In some situation, this made the difference between whether the game was comfortable to play or not.
You must be aware, though, that there are still rather few PhysX-supporting games due to the exclusiveness of that technology and the efforts Nvidia takes to keep it exclusive. The last and questionable move was the decision to give up supporting PhysX accelerators based on the original Ageia processor in Windows 7. At the time of our writing this review, the list of games that benefit from a hardware PhysX accelerator published at nZone.com names only 16 titles, one of which, Nurien, is not a true game and one more, Unreal Tournament 3: Extreme Physics Mod, is but a modification of an existing game. Then, PhysX effects do not improve in-game visuals as much as to justify the purchase of a PPU, the only exception being the stealth shooter Batman: Arkham Asylum. And finally, if you are really interested in one of the few PhysX-supporting games, you can achieve the same effect by buying and installing an inexpensive Nvidia GeForce GT 220 card which sells for less than $70 in retail. In some cases, a configuration of GeForce GTX 275 and GeForce GT 220 may prove even more effective than an EVGA GeForce GTX 275 CO-OP PhysX. Particularly, it has a considerably higher bottom performance in Cryostasis, making the gameplay smooth and comfortable at resolutions up to 1920x1200 whereas the GeForce GTX 275 CO-OP PhysX cannot do that, notwithstanding its formally more advanced PPU. Besides, this tandem is much quieter than the original product from EVGA.
Moreover, if you use the trick described in this review and combine a GeForce GT 220 with a Radeon HD 5800 series card, you will have even higher performance in PhysX-supporting games. You will also have a much higher speed than with any version of GeForce GTX 275 in all other games. DirectX 11, an HDMI audio core with support for HD formats, and much lower power consumption come as an extra bonus. If this hybrid tandem were more stable, it would easily beat the EVGA GeForce GTX 275 CO-OP PhysX as well as any tandem like GeForce GTX 275/285 + GeForce GT 220/240. Alas, our method is far from trivial and has stability issues, so we can’t recommend it for casual gamers whereas Nvidia keeps on hiding their rather good technology from the masses.
Summing it up, the EVGA GeForce GTX 275 CO-OP PhysX is an exciting product. We have never tested anything of its kind before. But, like with many other solutions with proprietary technologies, its applications and target audience are limited. And in this particular case, the user can achieve the same effect in an alternative way by purchasing any entry-level graphics card from Nvidia that works as a physical effects accelerator, e.g. a GeForce GT 220.
In the majority of games the EVGA GeForce GTX 275 CO-OP PhysX is going to perform just like any other GeForce GTX 275. So, we can only recommend it to users who need PhysX acceleration but cannot afford a GeForce GTX 295 or do not have a free PCI Express x16 slot. The EVGA GeForce GTX 275 CO-OP PhysX will be just perfect for them. But if you don’t play games with PhysX effects, the XFX Radeon HD 5850 Black Edition is going to be a better buy. It is faster, quieter, smaller, more economical and has more advanced functionality than the EVGA card. Additionally, the XFX outperforms the ordinary Radeon HD 5850 due to factory overclocking. And if you want to experiment, you can endow it with PhysX support, but again, we won’t guarantee that such a tandem will be stable and reliable.