Tech-Hounds.com

Internal Bridge Crossfire with Radeon X1950 Pro

The internal bridge Crossfire with the Radeon X1950 Pro marks the third effort from AMD in the realm of multi rendering solutions. By using internal and dedicated connectors, installation should be easier for end users not to mention it will also reduce the potential for display quality degradation. These are the two main concern gamers and users experience with the first version of,Crossfire that uses a Crossfire cable to connect master and slave cards..

Another improvement is the absence of master and slave cards, though actually this is not new. ATI introduced bridgeless / cableless Crossfire with the Radeon X1650 and X1900GT. In this second version of Crossfire, ATI chose to utilize the PCI Express bus to transfer data back and forth between the two cards. There is one problem with this setup - at higher resolutions and with more powerful GPUs, there is a chance the card will transfer more data than the PCI Express bus can handle - effectively saturating the bus.But more importantly, in this type of situations the Crossfire setup will likely not be able to reach its maximum potential.This is one of the reasons why more powerful card such as the Radeon X1900XTX and X1950XTX retains the use of a cable (and master / slave cards) for Crossfire.

They say the third time is a charm and hopefully this is true for AMD Crossfire. Taking example from NVIDIA's SLI, AMD introduced a Crossfire internal bridge and connectors with the Radeon X1950 Pro. Connectors? Yes, each Radeon X1950 Pro supports two internal bridge connectors, most likely to support expansion in the future (three, four cards Crossfire setup). Since Crossfire enabled motherboard didn't bundle the necessary Crossfire bridges to support this new version of Crossfire, ATI chose to bundle one bridge for each Radeon X1950 Pro. However, as we saw during our Radeon X1950 Pro round up, not all vendors adhere to this policy.

The Radeon X1950 Pro is very likely the most popular card in AMD's lineup right now. Though AMD decided to cull shader processing power by equipping the card with 12 pixel pipelines with 3 pixel shader units for a total of 36 shader units, they wisely retain the 256 bit memory bus from the card's more powerful brethen. This allows the card to provide performance close to NVIDIA's GeForce 7950GT, though its selling price is actually closer to GeForce 7900GS levels.

Obviously, we are very interested in seeing what a pair of Radeon X1950 Pro in a Crossfire setup has to offer. In addition to performance, we also like to see whether or not AMD kept their promise about ease of installaion and use. 

Installation

Installation was definitely easier than the old Crossfire with cables and dongles. Of course, you still need a Crossfire capable motherboard - the Intel P965 chipset on our Gigabyte P965-DS3P works just fine. Simply put the two cards in the corresponding PCI-E x16 slot, then install the two required bridges.  Let Windows install the drivers for the two cards, afterwards you can enable Crossfire from Catalyst Control Center.



Should be pretty easy right? Well, at least in theory. Even with identical cards, there are times when we had to reinstall the bridges (switching them) to enable Crossfire - Catalyst Control Center kept complaining we didn't install the bridges properly.

Mixed Vendor / Cards Setup

We first started to test Crossfire with mixed vendor / cards setup, using cards from three different vendors - Gigabyte, MSI and Sapphire. Since the Gigabyte was the odd one out (it didn't use the same board design as the other two), we opt to start with the Sapphire Radeon X1950 Pro and MSI RX1950 Pro. First, we put the Sapphire card in the first PCI-E x16 slot and the MSI card on the second and then install the two Crossfire bridges. After Windows detected the second card and install the appropriate driver, we restarted the PC. Then we enable Crossfire and started testing,  however we we immediately noticed display corruption everytime we entered full screen 3D mode. We tried switching Catalyst AI mode and even swapping the card but that didn't fix the problem.

So, we decided to use the Gigabyte Radeon X1950 Pro instead of the Sapphire card. After some initial testing, we found the only workable Crossfire setup is to put the Gigabyte card in the first slot and the MSI card in the second slot. Using the MSI card on the first slot will trigger a BSOD everytime we enter Windows. It's pretty likely this issue is driver related, the BSOD basically occured when Windows detects the new card. With the above setup, everything seems to be in order.

At least, that's what we thought. During testing, we encounter some very odd behaviour with our Crossfire setup, mainly with Quake 4 and The Elder Scrolls IV: Oblivion. We saw a huge performance drop in Quake 4 at 1280 x 1024 and 1600 x 1200, both with AA and AF and without. On some parts of the benchmark, the frame rate will slow to a crawling 30 fps at precisely the same point. In Oblivion, the Crossfire setup seems to be leaking (graphics) memory like crazy after about half an hour of gameplay.

	1024x768	1280x1024	1600x1200
Min	141	32	31
Avg	206.206	134.419	110.845
Maxi	248	331	215

We e-mailed an ATI representative about these problems and he asked us what cards are we using so that they can reproduce the problem. We sent a full description of the problem and information about the setup and tests we used. Unfortunately, we haven't heard any response about these issues to this day. However, searching the web, we found one possible explanation for the display corruption we saw with the Sapphire card in TechPowerUp's forums (here.). The problem seem to be caused by different BIOS used by the cards - using the same BIOS supposedly fix the problem - we didn't try flashing the BIOS for two reasons. One, ATI did promise there won't be any problem with using mixed vendor / cards and two. flashing the BIOS does involve some risks, more so if you're using cards from different vendors.

After some time tinkering, we found the problem with Quake 4 (and Oblivion) is that the Crossfire setup (or more precisely, the Catalyst drivers) seem to have a bug of 'flushing' or clearing the graphics memory. In Quake 4, the trigger is applying system changes or in our case executing the command 'vid_restart'. The workaround to this problem is simple - after applying system changes, quit the game to the desktop and then restart the game.

	With 'vid_restart'			With application restart
	1024x768	1280x1024	1600x1200	1024x768	1280x1024	1600x1200
Min	141	32	31	138	108	82
Avg	206.206	134.419	110.845	213.927	170.964	136.903
Maxi	248	331	215	311	255	204

A very huge difference in minimum fps and significant difference in average fps. So, we tried using the same workaround in Oblivion and - it works! Just like Quake 4, the trigger was entering / exiting maps, like you do when you enter / exit a town or a building (or fast travelling). Frame rates will drop to the teens, but after a save (on the same spot) and a restart, frame rates went up to where it's supposed to be (30 to 50 fps depending on the scene).

Before using this workaround, we also tried using identical cards (two Gigabyte Radeon X1950 Pro), but the problem still persists.  Clearly, the performance drops we're seeing are driver related. Using the latest Catalyst 7.3 didn't help, so hopefully this will be addressed in the next driver release. We certainly hope ATI will also address the issues with Crossfire installation with mixed vendor / cards. At the moment, we don't think you should use mixed vendor / cards setup , so if you want to use Crossfire, you better use two identical cards from the start. The question is - is it worth the hassle? We'll examine the performance Crossfire has to offer next.

Performance

Graphical settings for these games are set to their highest possible values. Notable differences are F.E.A.R where we disable "Soft Shadows", Quake 4 where we use High Quality - not Ultra Quality - but anisotropic filtering was set to "1" for default tests and "16" for AA / AF tests. We used Serious Sam II built in "Maximum" quality preset then change resolution and AA / AF settings accordingly. All of these settings can be seen here. Our TES IV: Oblivion settings are as close as we can get to Ultra High Quality - you can see our test settings here. For Need for Speed: Carbon, we used the game's "Maximum" video quality preset, but change the filtering to "Trilinear" for default tests - AA AF tests are run with "4x" and "Anistropic". NFS: Carbon test were done with the Sprint Race on Lincoln Boulevard. You can see the settings we used for GTR 2 here. Clear weather was used for our run of Monza GP tests.

Our test setup
Intel Core 2 Duo E6300 socket LGA-775 (running at 7 x 400 MHz)
4 x 512 MB A-DATA Vitesta 5-5-5-18 PC6400 DDR2-SDRAM
Gigabyte Radeon X1950 Pro 256 MB graphics card
Gigabyte P965-DS3P Intel P965 motherboard
Maxtor DiamondMaxPlus9 80 GBs Serial ATA 8 MB buffer
LiteOn 1673S DVD-RW
Tagan TG530-U15 530 watts ATX/BTX power supply

Windows XP Professional with Service Pack 2 installed
ATI Catalyst 7.2 and 7.3 reference driver
Intel Chipset Software Installation Utility 8.1.0.1006
DirectX 9.0c
all respected games used for benchmarks have been updated to their latest, final builds.

We'd like to thank Gigabyte for providing the Gigabyte P965-DS3P motherboard and the Gigabyte Radeon X1950 Pro graphics card for this test. We'd also like to thank Tagan for the Tagan TG530-U15 power supply.

The results:

Obviously, Crossfire has some performance issues with F.E.A.R. Although on average we see an increase in frame rates, we also notice much lower minimum fps - 26 to 28 fps compared to 31 fps we saw with single card setups. Gameplay testing sessions confirms that Crossfire has trouble with some parts of the game. The frame rates near the checkpoint in the Afterimage level (between ATC and Replica forces with mobile armor) are very much unplayable.


FEAR Performance Test 1024
(click here for a more detailed graph)

	Crossfire	Single X1950 Pro
<30 fps	1	0
30-45 fps	4	10
45-60 fps	7	15
60-90 fps	23	18
90-120 fps	13	3
>120 fps	7	9
Total	55	55

results are in seconds

Click at the detailed graph and you'll see see the problem lies near the end of the performance point - the explosion on the last room. As we have said repeatedly, minimum frame rates when you play the game tend to be much lower than what you saw in F.E.A.R Performance Test. You'll want minimum frame rates around 35 to 40 fps for a fluid gameplay experience. A Crossfire setup is unable to deliver such experience, though a single card setup can. Obviously, AMD have issues to sort out with Crossfire and F.E.A.R.


Because of the bug we mentioned earlier, we had to restart the game everytime we change graphical settings - AA, AF, and resolution. We noticed no performance loss with level reloads, so thankfully there's no bug there. You should be able to get the most out of Crossfire with normal gameplay. Just remember you need to restart the game after applying system changes or executing 'vid_restart'.

Crossfire really shines, offering significant frame rate gains in graphically limited situations. We can see gains between 40 to 60 percent in mininum fps with default and AA AF enabled, respectively. On average, the gains are smaller, but still significant - 30 to 50 percent and we're only talking about 1024 x 768 results. With higher resolutions, the gains become more significant - basically 60 to 70 percent on minimum fps (under default and when AA AF is enabled) 40 to 60 percent with average fps. At 1600 x 1200, we practically have double the performance of a single card.


Quake 4 - Data Processing Terminal 1024
(click here for a more detailed graph)

	Crossfire	Single X1950 Pro
<30 fps	0	0
30-45 fps	0	0
45-60 fps	0	2
60-90 fps	0	9
90-120 fps	3	16
>120 fps	22	9
Total	25	36

results are in seconds

The fps progress graph and table above shows how staggering the frame rate we gain with Crossfire. Instead of dropping to about 50 fps, we only dropped to around 100 fps with two Radeon X1950 Pro. For the most part of the benchmark, we saw frame rates around 150 fps compared to 100 fps with single card setups.


Notice the differences between setting Catalyst AI to Advanced and Standard. Although using Advanced allows us to get higher average fps, minimum fps are about 10 percent lower than Standard. So, if you want less drastic drops in frame rates, you should set Catalyst AI to Standard, that should eliminate most stutters you see when you're playing the game.

Using a fast processor like the Core 2 Duo E6300 at 2.8 GHz really allows us to experience the game at super high frame rates. We still have some graphical head room with default settings at 1024 x 768, though not much. It's only at higher resolution and with AA and AF enabled do we see significant differences between running Crossfire and single card setups. Even at 1600 x 1200, though the gains are significant, they are nowhere near the gains we saw in Quake 4. Here, we are only seeing about 50 to 60 percent increase. We couldn't help but wonder what kind of frame rates we can expect from Crossfire with more powerful cards - 80 to 90 fps seems to be the ceiling with our overclocked E6300 processor.


Serious Sam II - Greendale 1024
(click here for a more detailed graph)

	Crossfire	Single X1950 Pro
<30 fps	0	0
30-45 fps	0	0
45-60 fps	0	4
60-90 fps	36	59
90-120 fps	27	0
>120 fps	0	0
Total	63	63

results are in seconds

The graph and the table speaks for itself. Crossfire allows you to have a much higher average fps overall. Instead of spending time mostly between 60 to 90 fps, we can experience about half the time above 90 fps. Minimum frame rates are also up from around 60 fps to just between 75 to 80 fps. Of course, that's at 1024 x 768. At higher resolutions, the differences should be more significant though the nominal frame rates are lower.


It's a pity the same bug we saw with Quake 4 also plagues Oblivion - it would seem Crossfire have trouble 'releasing' memory whenever you switch between levels - like for instance entering / exiting a town or building. Frame rate drops can be ver significant. At one time, we experience frame rates of around 15 to 19 fps, but exiting the game and loading the save game at the very same spot would fis the problem - frame rates jumped to about 40 to 50 fps.

Oblivion is a very graphically intensive game when you use Ultra High Quality settings, more so if you max out all the sliders. This is already apparent at 1024 x 768 with AA and AF enabled - the difference is about 30 percent. At 1280 x 1024, Crossfire can still offer very playable frame rates on every part of Tamriel, with minimum frame rates of about 40 fps in this benchmark (and about 30 fps on heavily dense vegetationo areas like the Great Forest). Frame rates at 1600 x 1200 may appear to  be comfortable enough, but you'll definitely see lower frame rates when you enter the Great Forest (below 30 fps).


Oblivion - Grass Test 1024
(click here for a more detailed graph)

	Crossfire	Single X1950 Pro
<30 fps	0	0
30-45 fps	0	13
45-60 fps	5	8
60-90 fps	15	0
90-120 fps	0	0
>120 fps	0	0
Total	20	21

results are in seconds

Need we say more? Well, ok since we need a filler paragraph. Instead of seeing 30 to 45 fps, with Crossfire we experience much more fluid frame rates (45 to 60 fps) and we mostly see frame rates above 60 fps. Hopefully, AMD will be able to quickly fix the problems with this game and Quake 4. Maybe in the next Catalyst release and oh yes, overall better drivers for both Windows XP AND Vista.


Ouch. Updating the game to the latest patch did not fix the problem, so we suspect the Catalyst drivers is to blame here. Needless to say, you don't want to enable Crossfire when playing this game. Changing Catalyst AI settings didn't help, as you can see. Of course, disabling Catalyst AI means you disable Crossfire so that's no solution at all.


GTR 2 - Monza GP 1024
(click here for a more detailed graph)

	Crossfire	Single X1950 Pro
<30 fps	2	0
30-45 fps	4	0
45-60 fps	10	0
60-90 fps	28	10
90-120 fps	71	39
>120 fps	18	83
Total	133	132

results are in seconds

The fps progress graph and table above does not tell us anything new. If Crossfire was able to function properly, we could see areas where we are system or graphically limited. We did notice a trend though - the performance difference between Crossfire and single card setup seems to stay fairly constant - about 30 fps through the whole benchmark.