Original Link: http://www.anandtech.com/show/1997



Whenever a PC game pushes the limits of what current hardware can do, it generally ends up being fairly GPU bound. In the past, as long as you had pretty much any Socket-939 Athlon 64 you had enough CPU power to drive even the fastest single GPU video cards. You would typically be running at fairly GPU-bound graphics settings - even if you were CPU-bound, frame rates would be high enough that it wouldn't really matter. However, every now and then there comes a game that is an equal opportunity stress test on your system, requiring an extremely fast CPU as well as a high end GPU. Bethesda Softworks' latest hit title, The Elder Scrolls IV: Oblivion, is such a game.

In our initial article on Oblivion performance we compared high end and mid range PCI Express GPUs, discovering that we had finally found a game that was stressful enough to truly demand more GPU power than what is currently available on the market. Today's article uses the same benchmarks that we used in our first article, but focuses on finding the right mix of CPU and GPU performance for the best Oblivion experience.

It's worth stating up-front that we are not going to attempt to find ideal settings for every possible CPU/GPU configuration available. There are many tweaks that can be made that will dramatically improve performance on slower CPUs. Reducing the height of the grass as well as the density - or turning off grass entirely - will help a lot. Running without HDR, using medium textures, turning off shadow filtering... you can easily get performance to a level that many people will find acceptable, but it always comes at the cost of reducing the quality of the graphics - or at least the complexity of the graphics. We're interested in characterizing CPU performance under identical configurations for this article, providing an apples-to-apples look at how the Oblivion engine runs on a variety of processors.



The Test

Thankfully ATI's CrossFire runs on both ATI chipsets as well as Intel's 975X, so we were able to use our ultra high end GPU of choice to compare CPU performance under Oblivion. Remember that, just like in our first Oblivion article, we're manually walking through portions of the game and using FRAPS to generate our results, and thus the margin for error in our tests is much higher than normal; differences in performance of 5% or less aren't significant and shouldn't be treated as such.

While we tested with a number of AMD CPUs, we had issues with our Intel test bed where we couldn't adjust clock multipliers to give us the full spread of Intel CPU options, and thus we were only able to highlight the performance of a handful of Intel CPUs. However, with what we had we were able to adequately characterize the performance offered by Intel solutions under Oblivion. We also didn't have an Extreme Edition 965 on hand, so the EE 955 is the fastest offering from Intel in the test. The EE 965 should offer another 5% or so above what the EE 955 offers based on the tests we've done, just in case you're curious.

CPU: AMD Athlon 64 and Athlon 64 X2s
Intel Pentium Extreme Edition, Pentium D and Pentium 4
Motherboard: ASUS A8R32-MVP
Intel X975XBX
Chipset: ATI CrossFire 3200
Intel 975X
Chipset Drivers: ATI Catalyst 6.4
Intel 7.2.2.1007
Hard Disk: Seagate 7200.9 300GB SATA
Memory: 2 x 1GB OCZ PC3500 DDR 2-3-2-7 1T
2 x 1GB OCZ PC8000 DDR2 4-4-4-12
Video Card(s): ATI Radeon X1900 XT CrossFire
Video Drivers: ATI Catalyst 6.4 w/ Chuck Patch
Desktop Resolution: 1280 x 1024 - 32-bit @ 60Hz
OS: Windows XP Professional SP2

Armed with a pair of X1900 XTs running in CrossFire mode - the clear GPU performance leader in our first Oblivion article - we set out to run some additional tests. Pay attention to the rest of the system as well: we've installed 2GB of high quality (i.e. low latency) RAM, which also helps performance. 1GB is sufficient, but Oblivion appears to do a good job of making use of additional memory; load times and area transitions are noticeably quicker with 2GB of RAM. We used the same "High Quality" settings we introduced in the last review:

Oblivion Performance Settings High Quality
Resolution 1280x1024
Texture Size Large
Tree Fade 50%
Actor Fade 65%
Item Fade 65%
Object Fade 65%
Grass Distance 50%
View Distance 100%
Distant Land On
Distant Buildings On
Distant Trees On
Interior Shadows 50%
Exterior Shadows 50%
Self Shadows On
Shadows on Grass On
Tree Canopy Shadows On
Shadow Filtering High
Specular Distance 50%
HDR Lighting On
Bloom Lighting Off
Water Detail High
Water Reflections On
Water Ripples On
Window Reflections On
Blood Decals High
Anti-aliasing Off


Oblivion CPU Performance

Here's one thing we really didn't expect, for our most GPU intensive test to be extremely CPU dependent as well. In its natural state with no out-of-game tweaks, Oblivion will give dual-core CPUs about a 10% increase in performance over their single core counterparts at the top of our charts. Moving down the graphs, the X2 3800+ has a 15% performance advantage over the 3200+, while the Pentium 930 has a 20% advantage over a higher clocked Pentium 4 641.

While 10% may not sound like a lot, especially given that our FRAPS benchmarks can vary by up to 5% between runs, keep in mind that this is an extremely GPU intensive benchmark. A 10% difference with the fastest clockspeeds available is pretty significant. Intel clearly has more need of help, and the larger 15-20% boost they get from dual-core processors is nice to see. Unfortunately, it's not nearly enough to catch up to AMD's competing offerings.

As we've seen in other games, AMD's Athlon 64 X2 and vanilla Athlon 64 are the way to go, taking up the overwhelming majority of the top spots in this graph. The Pentium Extreme Edition 955 is about the only Intel CPU that is significantly competitive here; the Pentium D 930 is the next time that Intel makes another appearance and it offers performance lower than the single core Athlon 64 3500+.

Later in this review we will look at another way of characterizing CPU performance in Oblivion, but rest assured that this graph is far more important than just stating for the millionth time that the Athlon 64 FX-60 is faster than the Athlon 64 3500+....

Once again, we see a 10-15% boost from dual core CPUs in our Town benchmark. The rest of the standings and performance echo what we saw in the Oblivion Gate test above. The usefulness of the Radeon X1900 XT CrossFire setup diminishes significantly as you begin to move down the list of contenders; you'll actually lose over 20% of your real world frame rate if you've got an Athlon 64 3500+ vs. if you had an Athlon 64 FX-60. These aren't low resolution tests designed to isolate the impact of your CPU, these are tests at reasonable display settings for the GPU setup and these are real world results.

Oblivion also clearly benefits from larger CPU caches. The abysmal performance of the Celeron D 351 exemplifies this statement, but you can also look at higher performing parts. The difference between the Athlon 64 3700+ and the 3500+ is around 5-7%, which is more than many other titles show.

Our final Oblivion benchmark delivers the same message we've seen in the first two: CPU performance matters. But the real question is, how fast of a CPU do you really need to make your GPU investment worthwhile? That's the question we're answering next...



GPU Performance vs. CPU Clock Speed

For these tests, we took a single core AMD Athlon 64 (1MB L2) and increased its clock speed from 1.8GHz all the way up to 2.6GHz, measuring performance at each step of the way. The image quality settings haven't changed; what we're looking for here is if there's a pattern in the CPU/GPU relationship.

We picked four GPUs to look at their CPU dependency: the Radeon X1900 XT CrossFire, X1900 XT, X1800 XT and X1800 XL. We chose these four configurations because they represent the best ultra high end, high end, upper mid-range and mid-range GPUs for Oblivion. Our main interest is in finding out if there is a point at which the best mid-range GPU ends up being faster than the best high end GPU because of being paired with a faster CPU, or if having a faster GPU is really all that matters in Oblivion.

What this graph proves is that our Oblivion Gate benchmark is really only CPU bound if you've got a pair of X1900 XTs in CrossFire. What this does mean is that if you've got a low end Athlon 64, you won't see much of a performance difference between a single X1900 XT and a pair of them running in CrossFire mode. But for the most part this benchmark is no different than what we've seen from other games, with the X1900 XT, X1800 XT and X1800 XL being basically GPU bound - let's see if our other two tests show the same picture.

The Town benchmark is extremely CPU bound as you can see by this graph, and in Oblivion you do spend quite a bit of time walking around in towns. Being able to isolate the individual lines in this graph isn't very important because they basically all show the same thing, but what is important is to be able to look at the graph two dimensionally. What this graph shows us is that a single X1800 XT paired with a 2.4GHz CPU offers much better performance than an X1900 XT with a 1.8GHz Athlon 64, thus stressing the need to have a balanced CPU and GPU setup in order to avoid wasting money on a fast GPU. We already saw in our GPU performance article that CrossFire (and SLI) do nothing for performance in our Town/Dungeon benchmarks so the behavior here is not surprising.

Much to our surprise, the Dungeon benchmark ended up being a lot more GPU bound than the Town test but the conclusions we can draw are very similar. The Radeon X1800 XT does extremely well when paired with a high end CPU and will actually offer the same performance as a Radeon X1900 XT with a lower end or mid range CPU; it isn't until you give the X1900 XT a faster CPU as well that it can really stretch its legs and offer the performance advantage we know it has over its predecessor.



Hyper-Threading and SMP Tweaks

Another topic some people will find interesting is Hyper-Threading. We've seen a 10 to 20% performance increase with dual core processors, but HyperThreading is often regarded as a "poor man's SMP". We took the Pentium 4 641 chip and ran test both with and without HyperThreading. Here are the results:

Pentium 4 631 Hyper-Threading
HT Off HT On % Improvement
Oblivion Gate
Min 18 17 -5.56%
Avg 25.1 28.4 13.15%
Max 30 35 16.67%
In Town
Min 16 18 12.50%
Avg 29.1 31.2 7.22%
Max 45 48 6.67%
In Dungeon
Min 22 23 4.55%
Avg 44.9 48.6 8.24%
Max 119 130 9.24%

We were surprised to find that HyperThreading actually brought some pretty substantial performance improvements. The outdoor environment gets the biggest improvement, increasing average frame rates by 13%, but even the town and dungeon tests show a 7-8% increase. Certainly, this is not the difference between playable and unplayable frame rates, but even a small increase is nice to get.

So far, we've been looking at Oblivion performance without performing any special tweaks. While there are certainly many other tweaks that we could make to try to improve performance, we wanted to focus on changes that wouldn't affect the graphics quality but might still improve performance. We turn to TweakGuides.com and used their SMP-enhancing performance tweaks. For this test, we used an Athlon X2 4400+, once again with and without the tweaks enabled.

Athlon X2 4400+ SMP Tweaks
Tweaks Off Tweaks On % Improvement
Oblivion Gate
Min 32 34 6.25%
Avg 44.1 45.7 3.63%
Max 53 55 3.77%
In Town
Min 28 30 7.14%
Avg 48.5 48.9 0.82%
Max 75 74 -1.33%
In Dungeon
Min 36 37 2.78%
Avg 74 74.1 0.14%
Max 178 180 1.12%

As with HyperThreading, the SMP performance tweaks bring slight improvements in performance. However, none of the improvements are higher than 5%, which is roughly the margin of error for our benchmarks. The outdoor benchmark does see the largest increases, however, so some people might find these tweaks to be useful. Unfortunately, we also noticed some graphical glitches with all of the SMP tweaks enabled, so we can't honestly recommend using these tweaks for the small increase in performance. Your mileage may vary, so feel free to try them out; if you need more performance, though, you're better off looking at some of the other modifications you can make.



Final Words

In many ways, the message this article sends is something we've already known: a fast GPU needs a fast CPU. The difference with Oblivion is that the impact of a slow CPU on a fast GPU is far more noticeable than on many games we've tested in the past. While we can't realistically provide a CPU scaling chart for every GPU we tested in the last review, you can draw some general conclusions based on the four GPUs we focused on in this review. If you're using an older Athlon 64 3000+ or 3200+ with anything around the speed of a Radeon X1800 XT or faster, you'll want to look at upgrading (or overclocking, naturally) your CPU; otherwise, you'll be leaving a decent amount of GPU performance on the table.

As we've continually seen during the past couple of years, AMD's Athlon 64 and X2 processors are the best of the best, but even within the family you'll still want to opt for something faster than an Athlon 64 3500+ in order to make good use of any high end GPU.

It's also important to look at the multi-core optimizations that Oblivion provides. The benefit of a dual core processor is definitely visible in Oblivion, and we welcome more games where there's a tangible real world performance improvement to multi-core processors. The difference isn't quite as large as what we've seen with Quake 4, but we're heading in the right direction.

Those lucky enough to have a high end CrossFire setup, for example with two X1900 XTs, will definitely want to invest in a high end Athlon 64 X2. Oblivion is quite possibly the first game we've tested where we can actually justify (and this is a stretch) an FX-60 and a pair of X1900 XTs, as they enable you to have much more than you get out of them in most games. As we stated in the beginning, you can also try hacking your configuration files and downloading some mods, improving performance in other ways. If you just want to set the detail sliders on Maximum and play the game at high resolutions, though, X1900 XT CF and a fast dual-core CPU will get the job done nicely. (Good luck convincing yourself or your significant other of that "need", though!)

Log in

Don't have an account? Sign up now