by Sergey Lepilov
03/09/2009 | 06:35 PM
It is a widespread opinion among PC enthusiasts and devoted gamers that a graphics card upgrade must be accompanied with an upgrade of the whole platform in order to provide enough CPU power to the graphics subsystem so that the latter didn’t stay idle. As a matter of fact, this opinion is oftener proved to be wrong in practice than confirmed by tests. And I am going to check this out again in today’s review because I’ve got two new graphics cards from XFX, the fastest single-processor GeForce GTX 285 and the fastest dual-processor GeForce GTX 295, and two platforms with quad-core CPUs. So, I will try to find out how important a fast CPU and system memory are for these cards and will also compare them between each other.
The new graphics card is packed into an upright box that shows basic product information on the face side and detailed info on the back.
There is one more box made from thick cardboard inside the external wrapping. The graphics card is fixed within a piece of foam rubber. The following accessories are included into the box:
The product copies the reference design. Covering the entire face side of the card, the cooler’s plastic casing looks very pretty:
Well, tastes differ, and gamers’ tastes differ, too. The card measures 267x111x32 millimeters.
As opposed to its predecessor GeForce GTX 280, the GeForce GTX 285 has no memory chips on the reverse side of the PCB and no cap above them.
The card has the same interfaces as its predecessor. It plugs into a PCI Express x16 2.0 slot and is equipped with two dual-link DVI-I connectors and one S-Video output. There is a vent in the card’s mounting bracket for the hot air to be exhausted out of the system case.
There are connectors for building SLI (or 3-way SLI) configurations in the top part of the card’s PCB.
A 6-pin power connector is now installed instead of an 8-pin one because the GeForce GTX 285 has lower power requirements due to the thinner GPU tech process. According to the specs, the reference GeForce GTX 285 has a peak power consumption of 183W (53W lower than that of the GeForce GTX 280). A 550W or higher power supply is recommended for a computer with this graphics card.
The card’s PCB differs greatly from the GeForce GTX 280. Nvidia’s engineers had had to make the PCB design simpler and cheaper in order to reduce the manufacturing cost and they succeeded.
Besides moving all the memory to the face side of the card, the engineers revised its power section:
The 55nm GT200-350 chip was manufactured in Taiwan on the 48th week of 2008.
Judging by its marking, the chip is revision B3. This GPU incorporates 240 unified shader processors, 80 texture-mapping units and 32 raster back-ends. Thus, it is no different from the GPU installed on the GeForce GTX 280 (GT200-300) but thanks to the thinner tech process its frequencies are increased from 602/1296MHz to 648/1476MHz (+7.6/14.0%). The XFX GeForce GTX 285 (GX-285N-ZDFF) has the same frequencies as the reference sample although XFX already offers accelerated models with specially selected chips and factory overclocking. The GPU frequencies are reduced to 300/600MHz in 2D mode in order to save power and reduce heat dissipation.
The GeForce GTX 285 is equipped with 16 chips of 0.77ns GDDR3 memory from Hynix. The total amount is 1024 megabytes.
These H5RS5223CFR N3C chips are rated for a frequency of 2600MHz but clocked at 2484MHz in compliance with the GeForce GTX 285 specs. Although this is 12.2% higher than the memory frequency of the GeForce GTX 280, the chips can work faster still. I will check this out in an overclockability test shortly. The memory bus is 512 bits wide. The card lowers its memory frequency to 200MHz in 2D mode.
Let’s take a look at the card’s specs:
Take note that the GPU-Z tool reports that the GPU is revision B1 and has a manufacture date of January 2009 although the GPU’s marking clearly says revision B3 and the 48th week of the last year. Otherwise, there are no discrepancies.
The new card’s cooling system has not changed much over the mentioned GeForce GTX 280.
It consists of an aluminum base with a copper core and six 6mm copper heat pipes.
A heatsink consisting of thin aluminum ribs is soldered to the base.
A small blower with automatic control is pumping air through those ribs, cooling the GPU as well as the PCB.
The efficiency of the graphics card’s cooler was evaluated in a small test. The card was installed into a closed system case and the room temperature was 24°C. The card’s frequencies and temperatures were monitored with RivaTuner 2.22 and 2.23. There were two test modes: 3DMark06’s Firefly Forest test running at 1920x1200 with 16x anisotropic filtering ten times (this is a typical load on the graphics card in 3D games) and FurMark 1.6.5 running in the stability check mode at 1280x960 (this is similar to Linpack but for graphics processors; the GPU is loaded very heavily here). Each test ran for 13-15 minutes until the GPU temperature stabilized at its maximum. Since I had dismantled the card before the tests, the default thermal interface of the GPU was replaced with Arctic Silver 5 thermal grease.
Let’s see what temperature the graphics card has when working at its default frequencies and with the automatic management of its blower.
|
|
As you can see, the GPU is never hotter than 85°C irrespective of the test mode. The only difference is that the card’s blower was working at a moderate 2200rpm in 3DMark06 and at a very loud 3250rpm in FurMark. The temperature of the PCB is 3°C lower in the latter case, though.
It is even more interesting at the maximum speed of the blower.
|
|
The higher speed of the cooler helps reduce the temperature greatly, by as many as 12°C, under typical 3D load (3DMark06). When running FurMark, the card has the same temperature as before, and there is no benefit from the higher speed (but this load is far different from real-life games).
Next I checked out the card’s overclockability using its default cooler in automatic mode. The card proved to be stable at GPU frequencies of 691/1571MHz (+6.6/6.4%) and at a memory frequency of 2760MHz (+11.1%).
So, the XFX GeForce GTX 285 has but modest overclocking potential. I guess the best chips go into the production of XFX’s pre-overclocked models.
And now let’s see what temperature the overclocked card has:
The temperatures are lower than at the default frequencies because I had replaced the Ascot 6AR-B with an Antec Twelve Hundred system case just before that test. I hope you’ll forgive me that.
You can email me for this card’s BIOS.
The second card from XFX comes in a box that is larger but designed in the same way as the box of the above-discussed GeForce GTX 285.
1792 megabytes of memory is twice indicated on the face side of the box, but it is just marketing bait for people who are not versed in computer hardware. In fact, the card’s memory is 2x896 megabytes: the multiple sets of memory do not add up in SLI and CrossFireX configurations.
The box contains the same accessories as are included with the XFX GeForce GTX 285 with the addition of an audio cable.
The GeForce GTX 295 looks most impressive. It is a heavy, solid brick measuring 267x111x32 millimeters. The face side of the card is covered with a rubberized casing, and it is a pleasure just to hold it in your hands.
The reverse side has no casing, and we can make sure there are no memory chips on it.
The card is equipped with a PCI Express 2.0 interface, one HDMI port and two DVI-I connectors installed on one of the two PCBs.
There is a vent grid in the mounting bracket for exhausting the hot air out of the system case but some of the air goes into the gap between the PCBs and stays within the computer.
There is a MIO connector on one of the PCBs. It allows combining two GeForce GTX 295 cards in a Quad SLI subsystem. Two power connectors and an S/PDIF header are located at the opposite end of the PCBs.
So, the GeForce GTX 295 is powered by one 8-pin and one 6-pin connector and has a peak power draw of 289W. Nvidia recommends a 680W or higher PSU for a system with a GeForce GTX 295. For comparison, the peak power draw of the Radeon HD 4870 X2 is specified to be 260W.
When the plastic casing is removed, we can see that the PCBs face each other.
It was easy to take the card apart: I only had to undo the screws on both sides of it and unglue the GPUs from the cooling system.
There is a cutout at the back of the PCB for the cooler’s blower to get fresh air from.
In the front part of each PCB there is a NVIO chip. An nForce 200 chip, responsible for SLI mode, can be found there, too.
The PCBs are connected with two flexible bridges.
There are matching marks and numbers at the ends of each bridge and on the PCBs for you to connect everything properly. Otherwise, the card won’t start up or won’t work in SLI mode.
The GPUs are both revision B3. Their manufacture dates differ by one week only: 48th and 47th week of 2008. The third lines of the markings differ greatly but their meaning is only clear to Nvidia’s engineers.
Each GPU incorporates 240 unified shader processors, 80 texture-mapping units and 28 raster back-ends. Thus, the GPUs of the GeForce GTX 295 are in between the GeForce GTX 280/250 and GeForce GTX 260 (216SP) in terms of resources. The XFX card has GPU frequencies of 576/1242MHz, i.e. exactly those of the reference sample. The company also offers faster, pre-overclocked GeForce GTX 295 models. The GPU frequencies are dropped to 300/600MHz in 2D mode.
Each PCB carries 896 megabytes of GDDR3 memory.
These Hynix H5RS5223CFR N0C chips have an access time of 1 nanosecond and a rated frequency of 2000MHz. The card’s memory frequency is indeed 2000MHz in 3D mode and 200MHz in 2D mode.
Here is a summary of the card’s specs:
The XFX GeForce GTX 295 has a dual cooling system with a blower that drives air in between the halves of the heatsinks.
This cooler design has been discussed in our earlier reviews, so let’s just move on to its efficiency.
|
|
Contrary to my expectations, the card’s GPU temperature was not higher than 84°C under 3DMark06 but the blower was working at a noisy 2700rpm. Under the heavier load of FurMark, the GPUs were as hot as 94°C while the card’s power components were nearly 100°C hot. The blower was working at 3400rpm then, which was too loud.
And here are the results at the maximum speed of the blower (about 4000rpm):
|
|
The card is cooler in both test modes, but its blower is much too loud.
The new card pleased me with its overclockability as I managed to increase its GPU frequencies to 685/1477MHz (+18.9/18.9%) and graphics memory frequency to 2380MHz (+19.0%), its cooler working in automatic mode.
The card is not much hotter when overclocked:
I didn’t test the overclocked card from XFX but I can show you the result it achieved in 3DMark06 with an Intel Core i7 processor clocked at 4GHz:
You can email me for the card’s BIOSes.
The graphics cards were benchmarked in a closed system case with the following configuration:
Both platforms were tested in their nominal mode:
|
|
…and overclocked to 4GHz:
|
|
We tested the platforms stability in two cycles of OCCT test that took total one hour per processor. We disabled SMT mode of the Intel Core i7 CPU, because with this mode activated, the CPU was unstable at the 4GHz frequency without additional Vcore increase. Moreover, later on we found out that SMT activation in the games and applications we used for testing this time results in an insignificant framerate increase in Lost Planet: Colonies game and in 30~35% framerate improvement in Far Cry 2 (as tested on GeForce GTX 295 in 1280x960 resolution). We detected no performance changes in any other games and benchmarks.
The tests were run under Windows Vista Ultimate Edition x86 SP1 (plus all critical updates as of 02.21.2009). We started testing on 02.21.2009, so these are the driver versions we used:
Once the drivers had been installed, I made the following changes in their control panels: image quality set from “Quality” to “High Quality”; Transparency Antialiasing set to “Multisampling” instead of “Disable”; vertical synchronization set to “Always Off”. There were no other changes. I turned on full-screen antialiasing and anisotropic filtering in the menu of each game. If the game didn’t provide such options, I enabled FSAA and AF from the GeForce control panel. During the tests of GeForce GTX 295 I enabled any special settings responsible for activation of multi-GPU configurations support, if they were available.
The graphics cards were tested in two resolutions: 1280x960 and widescreen 1920x1200. The tests were performed in two modes: “High Quality” without any image quality enhancements and “HQ+ AF16x+AA4(8)x” with enabled 16x anisotropic filtering and 4x full screen anti-aliasing (or 8x FSAA with sufficient framerate for comfortable gaming experience).
The cards were benchmarked in the following set of applications including two synthetic benchmarks, one techno-demo and 12 games of various genres:
Here I’d like to add that if the game allowed recording the minimal fps readings, they were also added to the charts.
We’ve got very interesting results in 3DMark06. The GeForce GTX 295 is clearly more CPU-dependent than the GeForce GTX 285 just as we might have expected. The dual-processor card adds to the overall performance even in the hardest graphics mode when the CPU or the whole platform is changed or overclocked. The performance growth is nearly linear in the easy mode (1280x960 without FSAA and AF). Then, at the default frequencies the Core i7 is faster than the Core 2 Quad and increases the gap even more when they are both overclocked to the same frequency.
3DMark Vantage refused to output the total score for the resolution of 1280x960, so I will only publish the results of the GPU tests and the full results in the Extreme mode.
3DMark Vantage being a heavier test, we can only see it depend on the platform’s speed with the GeForce GTX 295 – and only in the Performance mode. The GeForce GTX 285 is indifferent to the platform performance altogether.
Both cards react eagerly to the acceleration of the platform in World of Conflict. Of course, this is more conspicuous with the GeForce GTX 295 than with the GeForce GTX 285. The Core i7 is astonishing again, being far faster than the 4GHz Core 2 Quad.
Enemy Territory: Quake Wars is not a new game but its speed is limited by the graphics subsystem in the FSAA + AF mode. Moreover, the faster GeForce GTX 295 loses to the GeForce GTX 285 at 1920x1200. Considering that SLI technology works quite normally in this application, I can suppose that the dual-processor card just lacks graphics memory since the GeForce GTX 285 has an extra 128 megabytes.
Comparing the platforms in the easier graphics mode, the Core i7 is faster than its predecessor at the default as well as overclocked frequencies. The GeForce GTX 285 becomes indifferent to the change of the CPU at 1920x1200 without FSAA and AF whereas the GeForce GTX 295 really needs a fast CPU for this game.
The graphics cards react well to the acceleration of the whole platform in Call of Duty: Modern Warfare, too. It is the first time that the Core i7 920 is somewhat slower than the Core 2 Quad QX9650 at the default frequencies, but only with the GeForce GTX 285 graphics card. We don’t see the same thing with the more advanced GeForce GTX 295 – the platform with the newer CPU is ahead in every mode. Each graphics card delivers a comfortable frame rate in every test mode but the GeForce GTX 295 shows better scalability than the GeForce GTX 285.
Unreal Tournament 3 produces interesting numbers. Despite the high frame rate in each mode, the GeForce GTX 295 shows excellent scalability on the Core i7 platform. Although such a high frame rate has no practical value, the Core i7 enjoys an impressive 22% advantage over the overclocked Core 2 Quad.
I have the same comment for all the three previous games: the graphics cards are not limited by the speed of the CPU or platform at large. The frame rate depends on the graphics card’s performance only.
The GeForce GTX 295 shows good scalability at 1280x960 without FSAA and AF on the Core i7 platform. We don’t see that in the other modes. The results of the GeForce GTX 285 are interesting because the Core i7 is somewhat slower than the Core 2 Quad (I rechecked this illogical fact but to the same effect).
Save for one test mode, the graphics cards both show excellent scalability on the two platforms in this game. The Core i7 beats its predecessor at the default as well as overclocked frequencies.
You have to manually specify the AFR2 rendering method in the driver in order to enable SLI mode in this game:
Unfortunately, we didn’t do that for our comparative review of CrossFire and SLI technologies. Hopefully, this problem will be corrected in the next version of the driver by means of a special settings profile. Here are the results:
The GeForce GTX 295 shows good performance scalability in Left 4 Dead until the resolution of 1920x1200 in the high-quality graphics mode. The GeForce GTX 285 is almost not limited by the platform at 1920x1200 in the low-quality graphics mode. The newer processor and faster memory wins in the competition between the Core 2 Quad QX9650 and Core i7.
The graphics cards are largely indifferent to the change of platform and CPU overclocking in this game. The only exception is that the GeForce GTX 295 delivers a higher average speed in the first scene of the test at 1280x960 without FSAA and AF.
The GeForce GTX 295 accelerates on the Core i7 platform at 1280x960. There is nothing interesting in the other test modes.
First of all, I suggest that you take a look at the advantage of the dual-processor GeForce GTX 295 over the fastest single-processor graphics card GeForce GTX 285. The results are shown for the overclocked Core i7 platform.
Since these two diagrams compare fast graphics cards, I guess it would be correct to evaluate them by the least CPU-dependent modes, i.e. at high resolutions and with high graphics quality settings. For example, the GeForce GTX 295 has an average advantage of 47.5% over the GeForce GTX 285 at 1920x1200 and in the FSAA mode, but the gap is smaller (33.5%) at 1280x960. Anyway, you can see that the dual-processor card is far faster despite its lower frequencies and smaller amount of graphics memory.
The next two diagrams compare the platform with a Core i7 920 processor overclocked to 4GHz (and with 3GB of DDR3 memory clocked at 1.6GHz) and the platform with a Core 2 Quad QX9650 overclocked to 4GHz (with 4GB of DDR2 at 1GHz). The results are shown for the GeForce GTX 295 card.
It is logical: the easier the game is for modern graphics cards, the higher the performance growth from the change of the platform, and vice versa.
I performed the power consumption test with a multifunctional Zalman ZM-MFC2 panel. This panel measures the overall power draw of the computer (without the monitor) rather than of a single component. There were two test modes: 2D (Word and Web surfing) and 3D (a double run of the DM-ShangriLa demo from Unreal Tournament 3 at 1920x1200 with 8x FSAA and 16x anisotropic filtering). I chose Unreal Tournament 3 because it can utilize all of the CPU’s four physical cores, which is not a typical capability of 3D games as yet. Here are the results:
It turns out that platforms with a GeForce GTX 295 consume about 80W more under load than the same platforms with a GeForce GTX 285. Comparing platforms with different CPUs (and mainboards and memory), the Core i7 920 consumes about 40W more than the Core 2 Quad QX9650 at the default frequency as well as at 4GHz. The difference is smaller in 2D mode but the newer platform still has higher power consumption.
What did we learn from today’s test session? Against my expectations, a fast platform with a Core i7 processor will come in handy for a GeForce GTX 295 graphics card in order to increase gaming performance. On the other hand, you should be aware that there are games and modes in which speed is limited by the graphics subsystem only. Particularly, these are Devil May Cry 4, S.T.A.L.K.E.R.: Clear Sky, Crysis, Far Cry 2 and the synthetic 3DMark Vantage. Considering that such cards are bought with the purpose of playing games at high settings and high resolutions, the role of the CPU becomes insignificant. As for the GeForce GTX 285, it is about 40% slower than the dual-processor GeForce GTX 295 and is a less CPU-dependent graphics card.
Then, I have also found out that the new platform with a Core i7 processor and fast memory is ahead of the Core 2 Quad platform in most of my tests at the default CPU frequencies as well as when both are overclocked to 4GHz. The Core i7 enjoys an impressive 35-40% advantage when not limited by the graphics card. The tests suggest that the new platform is worth the investment, and you can even hope for further performance benefits from SMT technology in the future. Well, Intel surely thought through every aspect of the launch of the new platform in the current economical situation. I guess I’ve offered enough data for owners of GeForce GTX 295 cards to make up their mind as to whether to replace the platform and overclock it for higher speed.
There is little I can say about the graphics cards I used for the tests. Both products from XFX are clones of the reference samples with sufficient accessories including a copy of Far Cry 2. The GeForce GTX 285 has a quieter cooler but the GeForce GTX 295 showed better overclockability. The single-processor model consumes less power, costs less, and is free from the drawbacks of the AFR rendering mode, but the dual-processor card is beyond competition if you are looking for maximum performance without compromises.
