Rage3D Hotfix Schmotfix - A Rage3D Look at the Vantage Hotfix Rage3D Hotfix Schmotfix - A Rage3D Look at the Vantage Hotfix
Releasing hotfix drivers for synthetic benchmark is evil. EVIL! Synthetic benchmarks are useless and irrelevant - or something. Is everyone wondering what kind of idiots would open an article about a hotfix driver for a synthetic benchmark in that manner? Good, that was the goal in order to create the setting for this little news-flash: the recently released Vantage Hotfix drivers fix quite a few things actually ... they just have a very uninspired name.
You might recall from Part I of our HD3870X2 investigation, the one focused on DX10 titles, that we encountered some issues, issues we penned on drivers based on our exchange with ATi. Since the writing of that piece, a full 8.4 Catalyst release and a slew of hotfixes have made their way into our hands, none of them changing the outlook on the aforementioned front in any significant way. This changes with the Vantage Hotfix.
Test setup:
Given the nature of this article, we'll jump right to the nitty gritty details, with only a short stop necessary to detail our test configuration.
The same testing methods outlined in Part I are employed. Unless otherwise specified, the CPU's clock was 3.2GHz. Now, let's fix Vantage with heat, shall we?
Editor's Note: Be advised that the values in the following graphs are rounded. As a consequence, differences of up to 0.9 fps can exist between the base framerates, albeit on the graph they'll be displayed as the same value(19.5 vs 20.4, for example, will both show up as 20). While this method was used to make the graphs more viewer friendly, we'll continue to monitor feedback so that, if necessary, we can revise our methodology in future articles.
From our earlier testing you might recall that GoW+3870X2+DX10 mode=extremely poor performance. This is one game we've kept an eye on, without any significant improvements until these new drivers hit the Rage3D labs.
Gears of War - 3870X2 vs. 3870X2 CF:
Yummy! We've checked the DX10+AA scenario as that was the worst offender in our earlier tests; the improvements we're seeing are nothing short of amazing. What once was a slow mess has become a very playable experience. Something else is also apparent though - currently, Gears appears to be capped at 2 GPU scaling, at least through its DX10 pathway. Let's synthesize the improvements the new hotfix brings:
Gears of War - Vantage vs Official:
The increases are rather huge, mainly due to the very low level of performance that prior drivers induced. Capping Crossfire to 2 GPUs for the moment also cured the issue with QuadCF causing even worse performance than DualCF - we would've loved to see scaling to 4 GPUs actually, but this will to do for the time being.
Lost Planet wasn't in the same boat as Gears DX10 (namely, bugged), it just suffered from less than stellar performance and no support for more than 2 GPU Crossfire. Is this still the case?
Lost Planet - Snow - 3870X2 vs. 3870X2 CF:
Lost Planet - Cave - 3870X2 vs. 3870X2 CF:
Notes: we've tested only the 4X and 8X AA settings as these were the ones most in need of improvement. Also, no AF is used in order to maintain comparability with the results obtained in Part I.
Lost Planet - Vantage Hotfix vs Official:
- the Cave setting, which was the worst performer in our experience, gets a boost both on Single and Dual X2s - probably some bug got squashed and/or some optimization got implemented that affected this particular scenario. Snow doesn't move all that much with the new drivers on the Single X2.
Lost Planet - Scaling:
The Catalyst crew seems to have cracked QuadCF scaling in this title, with both Snow and Cave settings benefiting with the move to Dual X2s, as opposed to prior behavior where scaling was capped at 2 GPUs. Cave+8X AA sees the greatest amount of scaling (nearly 50%).
Summing up, it seems that most improvements in Lost Planet are aimed at folks with more than 2 GPUs, albeit the general jump that the Cave setting experiences suggests that other tweaks and fixes were implemented as well.
World in Conflict treated us fairly well during our initial benchmarking, as long as AA remained disabled. Enabling AA seemed to cause a greater than expected performance drop, and a somewhat odd behavior when it came to Maximum FPS. Has this been addressed with the Vantage Hotfix?
We've tested with 4X AA and no AF (again, for comparability with earlier results), and it seems the AA situation has indeed been resolved, with performance results more in line with what we'd expect from these cards. It's also worth noting that Minimum FPS (one of the sore spots with multi-GPU setups in general) seem improved.
World in Conflict - Vantage vs Official:
World in Conflict - Scaling:
Scaling doesn't appear improved (remember, WiC was one of the games that scaled to 3 GPUs at most, with QuadCF not being available yet). It seems we'll have to wait some more until 4 GPUs can be effectively exploited by World in Conflict.
Remember that the Maximum FPS values obtained with QuadCF with earlier drivers were anomalous, and that Minimum FPS with a single X2 were much lower in the same conditions, and you'll realize that the huge gains seen in those 2 departments with the Vantage Hotfix aren't that earth-shattering, and that relative scaling to 4 GPUs remains unchanged.
It's another story with absolute values - the Vantage Hotfix brings significantly better performance with AA enabled, thus addressing another of the previous weak points found.
Hellgate was one of the most troublesome games we toyed with. Performance wasn't worth writing home about, and some bugs made using Quad Crossfire a non-feasible alternative. Since we've had such positive experiences until now with the Vantage Hotfix, perhaps it also brings some Hellgate goodness:
Hellgate: London - 3870X2 vs. 3870X2 CF:
Note: 4X AA is used, with no AF added to the mix.
If we were to strictly analyze performance, nothing significant seems to have happened, except for the massive improvement the QuadCF configuration sees in its Minimum FPS, and the reduction of Maximum FPS happening for it and the Single X2:
Hellgate: London - Vantage vs Official:
Going beyond that, we do have some good news: the stuttery-hitchy experience that used to be typical for the Quad is no longer here! You'll still suffer through some low performance upon first loading the game and levels (an issue associated with multi-GPUs, caused by the need to upload data to more than one GPU), but afterwards it becomes a fluent playing experience.
Hellgate: London - Scaling:
Scaling seems to have improved a bit as well, albeit not in a significant manner.
Of course we would've liked to see even more performance being extracted from the game, but even so, being able to actually play the game without going through the hitch-stutter-hitch syndrome is nice.
Having written quite a bit about this game recently, we had hoped to let it rest until something significant happened with regard to it. Since the Vantage Hotfix brings some changes affecting Assassin's Creed, we'll be mentioning it here before sending the game on a well-deserved vacation:
Assassin's Creed - 3870X2 vs. 3870X2 CF:
We're testing 4X AA and 16X AF, using the same run through Masyaf we used for our main AC article. In this case we're comparing the Vantage Hotfix to the 8.4 based Hotfix we used in that same main article. The changes that the new driver brings amount to including an official Crossfire profile (no need to use the exe renaming trick anymore) and enabling application controlled AF when using the DX10 and DX10.1 pathways (so no longer is it necessary to force AF through the CCC with these drivers).
The official CF profile is limited to 2 GPU scaling for now, so the numbers we got on the 3870X2 CF configuration with the older driver and the renamed exe are actually a bit better (there was a slight amount of scaling beyond 2 GPUs with that arrangement, if you recall). The tables are turned when we move to the Single X2 scenario, with the official profile being a better match for the game.
Assassin's Creed - Vantage Hotfix vs 8.4 Hotfix:
The improvements are nice, if not immense, and it's rather encouraging to see ATi reacting rather rapidly to the new release, as opposed to prior times when it took some waiting in order to get official CF support (some older games are waiting for it to this day).
With our single HD3870 card still in hiding (search parties are out), it's hard to evaluate the weights DX10 fixes and CF fixes hold in the noted performance improvements. What is certain is that we're dealing with a mix of the two.
What was sent out as something not that useful (synthetic benchmark hotfix, yuck), ended up being a pleasant surprise and, if Terry Makedon and his Catalyst team keep it up, many other good things are likely to arrive in the coming months, as there is always room for improvement. A first (major) improvement would be to employ more suggestive names for Hotfix releases, or at least hint at all of the improvements included in them, as opposed to what happened with the Vantage driver!
In wrapping up, this article exists for two reasons: the improvements offered by the Vantage Hotfix were certainly significant enough to be worth mentioning, and it also provides us the opportunity to clear something up with regards to our own work.
Generally speaking, we will reserve full driver reviews for releases that bring new and/or improved things to the table, so no monthly Rage3D Catalyst release reviews - unless each monthly release is as significant as this one (of which we certainly wouldn't complain!).
Our readers may remember that a Part III to the 3870X2 investigation was promised. Given the time that has passed since the writing of it, and based upon what the new drivers have shown us, we've elected to delay publishing it in order to redo the testing - this makes even more sense because Part 3 focuses on less commonly benchmarked games, and as such, games that are more sensitive to driver improvements. This had to be cleared up, lest you believe we had scrapped it.