by Alexey Stepin , Yaroslav Lyssenko
01/19/2007 | 11:58 AM
The Radeon X1950 Pro graphics card, based on ATI’s new RV570 chip announced on October 17, 2006, and described in our review , deservedly earned the title of the best product in the $199 category. It boasted excellent performance for its class in nearly every one of our tests.
But AMD didn’t limit its counter-offensive to one market sector and introduced a cheaper solution in a price range of below $199. To issue this solution they announced the less advanced RV560 chip. As you know, this price category had been dominated by Nvidia’s GeForce 7600 series. The RV530-based ATI Radeon X1600 XT was unable to make a worthy opponent despite its 12 pixel shader processors. Its performance was severely limited by its having only 4 TMUs and ROPs. ATI had to compromise eventually, releasing the Radeon X1800 GTO. That solution was indeed competitive, but also exceedingly complex and temporary due to limited quantities of R520 chips.
Thus, the introduction of the RV560 for the below-$199 segment was just as urgent as the release of the more advanced RV570 for the performance-mainstream category. AMD needed a graphics processor that would successfully compete with the GeForce 7600/7900 series while being as simple and economical as Nvidia’s products. On October 30, 2006, the RV560 was announced officially and the company disclosed the RV560-based Radeon X1650 XT graphics card. The recommended price of the new product was set at $149 which made it a direct market opponent to the GeForce 7600 GT.
To make AMD’s mainstream graphics card line-up consistent, the ATI Radeon X1600 XT was renamed as ATI Radeon X1650 Pro. Now you have to be watchful while shopping: it is only the XT-suffixed model in the Radeon X1650 series that uses the new RV560 chip. The Pro model is based on the slower RV530.
The Radeon X1650 XT seems competitive in terms of price, but is it such in terms of speed? Let’s consider the specifications of the new RV560 chip.
Like its elder brother RV570, the RV560 chip is manufactured on TSMC’s 0.08-micron tech process but has fewer functional subunits and uses a 128-bit memory access. As a matter of fact, the RV560 and RV570 are both the same 330-million-transistor chip, but some functional units and memory controller interfaces are disabled in the former whereas the latter is a full-featured version of the basic chip.
The first thing to be noted is that the Radeon X1650 XT has two times more pixel processors than the GeForce 7600 GT, but has the same number of ROPs. It means it is going to be considerably faster than the Nvidia card at complex math1ematical computations and at least equal to it when it comes to full-screen antialiasing. The Radeon is also equipped with more vertex shader processors than its opponent, which should ensure it an advantage in geometrically complex scenes such as scenes with dynamically generated vegetation.
The graphics cards are roughly equal in terms of memory bandwidth, but the difference in the memory controller architecture should be taken into account. The Radeon X1650 XT uses a ring-bus architecture with two bidirectional buses (the cut-down RV560 may be using unidirectional buses, though) whereas the GeForce 7600 GT features a classic topology. The ring-bus topology is known to ensure lower memory access latencies, so the advantage is on the Radeon X1650 XT side again. It may show up in such cases when the graphics memory performance is of paramount importance like in HDR or full-screen antialiasing modes.
Theoretically, AMD solution’s weakness may come from its fewer TMUs and subunits that process the Z-buffer, yet it’s not a fatal handicap, just as it is not for RV530-based graphics cards. It is unlikely to prevent the new product from competing successfully with the GeForce 7600 GT as we are going to see in our tests soon.
Like the RV570, the RV560 chip incorporates a full-featured Compositing Engine. You can learn more about the changes in CrossFire technology on new ATI Radeon graphics cards from our comparative review of Radeon X1950 Pro and GeForce 7900 GS.
The Radeon X1650 XT printed-circuit board resembles the one of the Radeon X1600 XT, but this is just an appearance. It is a completely new design, largely determined by the peculiarities of the RV560. The two cards are the same size, but that’s in fact the single similarity between them.
You can spot significant differences even without removing the cooler. Note the placement and layout of the power circuit, for example. The bonding pad for a VIVO-supporting Rage Theater chip is placed at an angle of 45 degrees. The VIVO chip is not installed on our sample of the Radeon X1650 XT and we guess it will be missing on many other samples. Such functionality isn’t demanded nowadays, yet it raises the cost of the end-product.
As opposed to the Radeon X1600 XT that comes without an additional power connector, the Radeon X1650 XT can have one, but the connector is not installed on our sample of the card. Some power elements are not installed, either. There are only four power transistors of the voltage regulator on the PCB instead of eight. In the same place on the reverse side of the PCB a number of elements are missing, including a second controller, as it seems. This is indicative of rather low power consumption of the card. This also suggests that there may appear versions of the card with improved frequency or functional characteristics. On the other hand, the PCB might be developed to be future-proof and the empty seats in the power circuit may mean that this PCB is going to be used in ATI’s forthcoming mainstream products that will have more powerful graphics processors with much higher power consumption.
The RV560 chip is much larger than the RV530. Its die size is comparable to that of the RV570, but its packaging is smaller, without a protective frame. The Radeon X1650 XT’s cooler isn’t large or heavy, so the lack of the frame is not a problem if you handle the device properly. The main markings of the RV560 and RV570 differ by only one letter, G and D, respectively. So, this is indeed one and the same GPU in different packaging and with a different memory bus width. Our sample was manufactured on the 34th week of 2006 which corresponds to late August. Following the official specification, the GPU is clocked at 575MHz. There are fewer active subunits in the RV560 than in the RV570, so we can expect it to have somewhat lower heat dissipation and heat consumption.
Unlike the senior Radeon X1950 Pro, the Radeon X1650 XT is equipped with Infineon’s memory. There are four 512Mb 16Mx32 GDDR3 chips on board (HYB18H512321AF-14) for a total of 256MB accessed across a 128-bit bus. The rating frequency of the chips is 700 (1400) MHz, but they are actually clocked at 675 (1350) MHz on this card.
The Radeon X1650 XT has the same configuration of connectors as the Radeon X1950 Pro: two DVI-I and a universal S-Video/VIVO. There is a pair of new CrossFire connectors in the top left of the PCB – the two cards are joined with special flexible cables via these connectors.
The cooler installed on the Radeon X1650 XT should be familiar to you from our Radeon X1600 XT review. Its design is very simple: a copper base to which a folded metallic sheet is glued. The cooler has a small fan with straight blades that drives air through the ribbing and is covered with a black plastic casing. The fan uses a two-wire connection, it is not equipped with a tachometer as more advanced ATI Radeon models are. There are no advanced cooling technologies, like heat pipes or something, here.
This cooling system used to be criticized, and quite justly so, for being noisy. Perhaps the fan speed management works better on the Radeon X1650 XT – we’ll check this out in the next section. The cooler contacts the GPU die through a later of traditional dark-gray thermal grease. The memory chips are not cooled.
We measured the level of noise produced by the graphics card’s cooler with a digital sound-level meter Velleman DVM1326 (0.1dB resolution) using A-curve weighing. At the time of our tests the level of ambient noise in our lab was 36dBA and the level of noise at a distance of 1 meter from a working testbed with a passively cooled graphics card inside was 40dBA. We got the following results:
So, the Radeon X1650 XT is by far not as loud as the Radeon X1600 XT is. Its fan is working at its max speed for only a few seconds after the system starts up. Then the fan speed is lowered and the noise subsides. This cooler can’t be said to work silently, yet it is a great improvement over the Radeon X1600 XT’s reference cooler. We could not make the card increase its fan speed even by long testing it in 3D applications, so 42.4dBA is the final result. This is much better than the results of the Radeon X1900 XT and GeForce 7900 GT. Subjectively, the Radeon X1650 XT is comfortable to deal with. Its noise is hardly distinguishable against the noise from other PC components.
The power consumption of the Radeon X1650 XT was measured according to our traditional method on the following testbed:
The mainboard in this testbed is modified: measuring shunts are connected inline into the power lines of the PCI Express x16 slot. The shunts have connectors for measuring instruments. The 2xMolex → PCI Express power adapter is also equipped with such a shunt. We perform our measurements with a Velleman DVM850BL multimeter that has a measurement accuracy of 0.5%.
To create a 3D load on the graphics card, we ran the first SM3.0/HDR test from Futuremark 3DMark06 in a loop at 1600x1200 resolution with 16x anisotropic filtering. A Peak 2D load was created by launching the 2D Transparent Windows test from Futuremark PCMark05. Here are the results:
Click to enlarge
Just as we had suspected, the Radeon X1650 XT breaks no records when it comes to power consumption under load. Its power draw in this mode is about 10.5W lower than that of the Radeon X1950 Pro, but much higher than that of the GeForce 7600 GT and 7600 GS and even of the GeForce 7900 GS. The +3.3V line is loaded hard as is typical of a lot of other Radeon-based graphics cards.
So, the use of 0.08-micron tech process didn’t help AMD reduce the power consumption and heat dissipation of its new products. They are still inferior to same-class Nvidia solutions in this parameter.
Let’s see if the Radeon X1650 XT is competitive in terms of performance it provides.
We tried to overclock our Radeon X1650 XT but failed. The latest version of RivaTuner (16.2) identified the card correctly, but the system would hang up as soon as we increased the memory frequency. And when we set a higher GPU frequency, RivaTuner’s monitoring module would report that the frequency was dropped to the default value on launching a 3D application.
So we abandoned all attempts to overclock our sample of the Radeon X1650 XT to return to this matter later. We’ll check the Radeon X1650 XT out at overclocking as soon as we get a retail version of this card.
We performed out test session for ATI Radeon X1650 XT and its competitors on the following platforms:
The drivers were configured for maximum texture filtering quality.
We selected the highest possible graphics quality level in each game. We didn’t modify the games’ configuration files. Performance was measured with the games’ own tools or, if not available, manually with Fraps utility. We also measured the minimum speed of the cards where possible.
We tested the cards in three standard resolutions according to our testing methodology: 1280x1024, 1600x1200 and 1920x1200. Since the performance of mainstream solutions including ATI Radeon X1650 XT is not high enough for enabling FSAA, we ran the tests only in two modes: only with anisotropic filtering 16x and with FSAA 4x and anisotropic filtering 16x. We enabled FSAA and anisotropic filtering from the game’s menu. If this was not possible, we forced them using the appropriate driver settings of AMD Catalyst and Nvidia ForceWare.
We ran the tests with disabled FSAA only for those games that do not support FSAA due to the specifics of their engine or use HDR (FP16). The thing is that the GeForce 7 family cannot perform FSAA together with floating-point HDR.
Besides the ATI Radeon X1650 XT, we included the following graphics cards into this review:
Since we didn’t have an original GeForce 7600 GS graphics card at the time the review was conducted, we simulated it by dropping the clock speeds of the GeForce 7600 GT down to the necessary level.
For our tests we used the following games and applications:
First-person 3D shooters:
Third-person 3D shooters:
Synthetic gaming benchmarks:
The Radeon X1650 XT beats the GeForce 7600 GT in our first test, but can’t match the GeForce 7900 GS that boasts 20 TMUs, 16 ROPs and a 256-bit memory interface. The gap isn’t bigger than 10-13%, though.
It should be noted that the graphics card is fast enough only at 1280x1024 without full-screen antialiasing, and even in this case its average frame rate is lower than the 60 frames per second which you need to play a first-person shooter with comfort.
Although the game employs a new renderer with support for cutting-edge shader-based DirectX 9 effects, its engine is a modified version of the Quake 3 engine. This is perhaps the reason why the AMD cards are so slow: the Radeon X1650 XT is everywhere slower than the GeForce 7600 GT while the Radeon X1950 Pro, having more pixel processors and a higher core frequency, cannot surpass the GeForce 7900 GS.
The game is obviously too demanding for graphics cards like Radeon X1650 XT and GeForce 7600 GT, and you can’t get a comfortable speed from them even by disabling full-screen antialiasing. You’ll probably have to switch to 1024x768 or use lower graphics quality settings.
The Radeon X1650 XT is successfully competing with the GeForce 7600 GT throughout this test. The more advanced memory controller gives an advantage to the AMD solution in high resolutions of the 4x FSAA + 16x Aniso mode.
The resolution of 1600x1200 pixels plus 16x anisotropic filtering is the highest playable mode for both graphics cards. They aren’t strong enough to perform full-screen antialiasing even in such an undemanding game as Far Cry (well, you can try to enable 4x FSAA at 1280x1024, but don’t complain about occasional slowdowns).
The Radeon X1650 XT feels much more confident on the Research map in the FSAA mode than in the previous case, enjoying an advantage of 20% over the GeForce 7600 GT. The graphics cards deliver higher speeds in this test on average, but it is still arguable whether FSAA is an option. The speed may bottom out below comfortable level as soon as you go out into an open scene.
The numbers suggest that the FP HDR mode of this game is totally unplayable on graphics cards like Radeon X1650 XT or GeForce 7600 GT. The speed is too low.
The competing solutions from AMD and Nvidia deliver similar performance with a certain advantage on the side of the former in the FSAA mode and vice versa. However, this performance is insufficient for this game. It’s only in 1280x1024 resolution and with anisotropic filtering (no FSAA!) that you can get a more or less comfortable frame rate.
Using the deferred rendering technique, this game is incompatible with full-screen antialiasing. There are only anisotropic filtering results here. Moreover, the game is too heavy to be run on mainstream graphics card and we decided to benchmark them in 1280x1024 only.
Anyway, the numbers are very low. Graphics cards from the below-$199 category do not suit much for playing Ghost Recon Advanced Warfighter . The more advanced Radeon X1950 Pro and GeForce 7900 GS look unassuming in this test, too, but always manage to keep the frame rate above 25fps at least.
It’s the opposite of what we’ve seen in F.E.A.R. Extraction Point : the Radeon X1650 XT is for some reason slower than the GeForce 7600 GT in the 4x FSAA mode in spite of its low-latency memory controller architecture.
In the FSAA mode, the GeForce 7600 GT is ahead by about 10% at 1280x1024, which is the single resolution the mainstream cards provide comfortable speeds in. Well, neither of the two cards can deliver an average frame rate of over 50fps even in 1600x1200 without FSAA.
The Radeon X1650 XT isn’t much slower than the GeForce 7600 GT in this OpenGL application, but these are purely theoretical data. If you try to play Prey on either of these cards, you’ll have to lower the level of detail or switch into a resolution lower than 1280x1024, or perhaps do both things at once.
The Radeon X1650 XT has the same speed as the GeForce 7600 GT here: the AMD solution compensates its fewer TMUs (8 against 12) with its pixel processors (24 against 12).
But when it comes to practical gaming, it’s only the Radeon X1950 Pro of all the participating graphics cards that ensures a comfortable gameplay. It delivers an average frame rate of 45fps and a min speed of 25fps.
It is impossible to run this game on the Radeon X1650 XT at the highest graphics quality settings. It’s somewhat better with the GeForce 7600 GT because its min speed is never lower than 25fps in 1280x1024. This makes playing the game possible considering the lower requirements of third-person shooters in comparison with first-person shooters. However, some jerkiness and loss of control accuracy may occur, and you may want to step down to 1024x768 to avoid such problems.
Note also the characteristic thing in the contest of the Radeon X1650 XT and GeForce 7600 GT: the former may be slower in FSAA-less modes, but then is often ahead when FSAA is turned on, especially in high resolutions.
The average speed of the contending graphics cards is similar when there’s no FSAA, but the Radeon X1650 XT is preferable in terms of minimum speed due to its having more pixel shader processors. A smaller difference between the average and minimum speed always means a higher level of comfort in gameplay, so the Radeon should be considered the winner of this test.
When it comes to the 4x FSAA + 16x AF mode, the Radeon X1650 XT has a better minimum as well as average speed in high resolutions, but the numbers are too low for this mode to be playable. Like in a majority of other games, we do not recommend enabling FSAA on graphics cards of this price category.
The Radeon X1650 XT is always slower than the GeForce 7600 GT regardless of FSAA mode, although it gets closer when FSAA is enabled thanks to its more advanced memory controller. We can’t pinpoint the reason for this defeat. Perhaps the game engine puts a very high load on the graphics card’s TMUs. The results of the Radeon X1950 Pro without FSAA agree with this supposition.
There’s no talking about comfortable play here. Every graphics card included in this test provides too low performance. The GeForce 7900 GS has the best average frame rate (when FSAA is off), but slows down to 15fps at times.
The current version of this game doesn’t support FSAA, so we had to use anisotropic filtering only.
The game makes wide use of shader-based special effects, and the 24 pixel processors of the Radeon X1650 XT card lift it up to the same level with the GeForce 7900 GS. Without full-screen antialiasing, the 256-bit memory bus of the Nvidia card doesn’t give it any advantage.
The current version of Gothic 3 is a highly demanding application. The mainstream graphics cards run it too slow. You can’t have a smooth gameplay on them.
The game looks best with enabled HDR, so we benchmarked the cards in this mode. Moreover, the Nvidia GeForce 7 series cannot use FP HDR together with FSAA.
The Radeon X1650 XT is slower than the GeForce 7600 GT in 1280x1024, but then overtakes it in 1600x1200 and leaves it behind in 1920x1200. The speeds are low, however. Moreover, the reflection and refraction effects do not work on Radeon X1000 cards in the current version of the game.
The Nvidia GeForce 7 architecture doesn’t permit to use FP16 HDR along with full-screen antialiasing, but without HDR the game loses much of its visual appeal. That’s why we test the cards here with anisotropic filtering only.
In closed environments, houses and dungeons, the Radeon X1650 XT and GeForce 7600 GT deliver similar performance, but the Radeon offers a higher minimum of speed, which makes it preferable for playing this game.
The Radeon X1650 XT has a bigger advantage in open scenes. Although the two opponents have comparable GPU and memory frequencies and the same amount of ROPs, the AMD card features a ring-bus controller (it minimizes memory access latencies), uses improved invisible surface removal algorithms, and has more pixel and vertex processors than the opponent. This is all called for in this game, so it’s no wonder the Radeon X1650 XT is only slower than the higher-class products in this test, which is one of the most difficult on our list.
Well, the speed of the Radeon X1650 XT isn’t high by itself, yet its opponent is even slower. So, the choice is obvious here.
Radeon X1000 series cards do not support vertex texturing, so they can’t use the appropriate technology to render the water surface with the highest quality in this game. This is only available on Nvidia GeForce 7 GPUs.
The Radeon X1650 XT is no match to the GeForce 7600 GT in both performance and image quality here because the game is optimized for Nvidia’s architecture. Anyway, you have to lower the resolution and/or the level of detail to play the game with comfort.
If you like this very beautiful space simulator, you can now run it on a mainstream card that provides the performance of a GeForce 7900 GS at a lower price. The new Radeon is limited to 1280x1024, but you can’t expect more from a graphics card of its class.
The Radeon X1650 XT is about 9-10% slower than the GeForce 7600 GT in 1280x1024 irrespective of FSAA, but the gap is getting smaller in high resolutions. Unfortunately, the speed is low here and you may want to switch to 1024x768 or lower the level of detail.
The progressive nature of the Radeon X1650 XT architecture shows up in the pure speed mode: it is a second test when the AMD performs as fast as the higher-class GeForce 7900 GS.
It can’t happen in the FSAA mode due to the Radeon X1650 XT’s 128-bit memory bus, but it is successfully competing with the GeForce 7600 GT here. Its average frame rate is lower in 1280x1024, but it has a higher minimum of speed.
The Radeon X1650 XT scores only 115 points less than the GeForce 7900 GS. That’s an excellent result considering that the AMD solution has worse technical characteristics, particularly a 128-bit memory bus, and belongs to a lower price category. Let’s discuss the results of the separate tests now.
The GeForce 7600 GT is in the lead in the “pure speed” mode thanks to its 12 TMUs, but the Radeon X1650 XT overtakes it in the 4x FSAA mode.
The Radeon X1650 XT triumphs in the second test. This is expectable considering its 8 vertex processors as opposed to the GeForce 7600 GT’s 5. The scene is rather small and the memory load is low here. This allows the AMD card to get very close to the GeForce 7900 GS.
The third test is like the first one: the Radeon X1650 XT is somewhat slower with disabled FSAA and equal to its opponent with enabled FSAA. The GeForce 7600 GT is better in high resolutions of the FSAA-less test mode, just like in the first test.
The overall outcome is correct since the Radeon is just a little slower than its opponent in two out of three tests, but has a considerable advantage in the second test. This is true for 1280x1024 resolution and should also be true for 1024x768 the benchmark defaults to.
The latest version of 3DMark makes full use of the capabilities of today’s graphics cards, so it shows quite clearly that the Radeon X1650 XT and the GeForce 7900 GS belong to different product classes. However, the Radeon scores 131 points more than the GeForce 7600 GT.
The Radeon X1650 XT is a mere 59 points ahead of the GeForce 7600 GT in the SM2.0 tests. That’s right as these tests are limited to the Shader Model 2.0 framework. The smaller number of texture-mapping units in the AMD card is not a limiting factor here.
In the tests that make use of Shader Model 3.0 and FP HDR the difference is 130 points, or 10%, in favor of the Radeon X1650 XT. This is not a twofold advantage, yet it’s clear that ATI’s 3-to-1 concept is more relevant today than the 1 pixel processor per 1 TMU concept implemented in the GeForce 7 series.
The results of the first test aren’t good for the Radeon X1650 XT in the FSAA-less mode. That’s the only situation when the GeForce 7600 GT enjoys a big advantage due to its 12 TMUs. In the 4x FSAA + 16x Aniso mode this advantage is negated by the memory subsystem bandwidth and the number of ROPs (to remind you, the two cards have the same number of ROPs).
In the second test the “pure speed” of filling the scene is not a decisive parameter. Like in the similar test from 3DMark05, the Radeon X1650 XT wins here, but by a small margin, especially when there’s no FSAA.
The Radeon X1650 XT gains the upper hand again in the first of the SM3.0/HDR tests, although the gap is a mere 0.5fps.
Strangely enough, it is the GeForce 7600 GT that wins in the second SM3.0/HDR test, although the Radeon X1650 XT supports the Fetch4 feature that accelerates the rendering of shadows with Cascaded Shadow Maps. Judging by the separate tests, the overall scores of the two graphics cards – the Radeon X1650 XT has just a little higher score than the GeForce 7600 GT – are correct.
Following the successful RV570, the RV560 graphics processor has made a good debut and the RV560-based ATI Radeon X1650 XT graphics card has done well in real-life tests. The new card offer much higher performance in comparison with the Radeon X1600 XT, which is now called Radeon X1650 Pro, and is quite competitive against the GeForce 7600 GT and sometimes even against the more technically advanced GeForce 7900 GS. There are few cases when the Radeon X1650 XT is far slower than the GeForce 7600 GT. It is such games as Call of Duty 2, Hitman: Blood Money, Tomb Raider: Legend, and Pacific Fighters . And as a rule, both solutions are too slow in these games to prove a comfortable gameplay.
Alas, today’s 3D games are generally extremely demanding to be run at an acceptable speed on mainstream graphics cards at resolutions higher than 1280x1024 and with the maximum graphics quality settings. There’s no talking about full-screen antialiasing, of course. We’ve only got acceptable or near-acceptable results in the 4x FSAA + 16x Aniso mode in such games as Far Cry , Half-Life 2: Episode One and Splinter Cell: Chaos Theory . So, the optimal choice for mainstream-class graphics cards seems to be resolutions below 1280x1024 (or below 1600x1200 in some cases) coupled with anisotropic filtering. Today’s graphics cards can perform such filtering without losing much in speed whereas it can considerably improve the image quality. The ATI Radeon X1650 XT has an advantage over the GeForce 7600 GT in this respect as it supports High Quality AF mode which provides high-quality anisotropic filtering irrespective of the texture’s angle of inclination.
The ATI Radeon X1650 XT is good from a technical point of view. It is small, consumes little power, dissipates a decent amount of heat, and uses a simple cooler. The defect of the Radeon X1600 XT – the fan speed management system worked incorrectly, producing too much noise – is now removed and the new Radeon is rather quiet. We should acknowledge that the Radeon X1650 consumes more power than the GeForce 7600 GT. The new GPUs from AMD don’t break any records in this area, despite the use of 0.08-micron tech process.
The design of the Radeon X1650 XT is simpler in comparison with the Radeon X1800 GTO, so we can expect the new card to become even cheaper eventually. Although the GeForce 7600 GT has been long on the market, the Radeon X1650 XT still has a chance to win customers. Today, this solution is the best offer indeed in the mainstream graphic card class.