Mobile Platform Wars: AMD vs. Intel

X1250 vs. GMA X3100

For the Graphics comparison, we decided to focus on the best case scenario for AMD and compare performance using the TL-66 to the Intel GMA X3100. We will include charts later that show that the difference between using a TL-60 and a TL-66 isn't particularly great when talking about gaming performance, but we just wanted to make this point clear.

The Radeon X1250 is based off of the Radeon X700 hardware, with a few changes. First, half the pipelines have been removed (which actually makes the hardware more like an X300/X600), so it comes with four pixel pipelines. All of the vertex shader pipelines have also been removed, letting the CPU handle that part of the graphics equation. Note also that the origins of this IGP mean that it lacks support for Shader Model 3.0, but unlike the X300/X600 it does include SM2.0b support. The X1250 also includes some additional functionality related to video processing, although we won't be testing that area of performance in this article.

For the Intel camp, the GM965 Northbridge includes the GMA X3100 graphics processor. Figuring out exactly what is and isn't supported by this chip can be a little complex, in part because the drivers have been so bad (at least in terms of gaming support), particularly under Windows Vista. We can say for sure that the GMA X3100 supports at least a subset of SM3.0, because it is able to complete that section of Futuremark's 3DMark06, and it appears to be capable of running certain SM3.0 games. In terms of features, that theoretically moves the X3100 ahead of the X1250, and it should also be better than the GMA 3000/3100 that's found in the Q33/Q35/G33 chipsets.

In the past, the assumption has always been that NVIDIA and ATI/AMD integrated graphics solutions were superior to the stuff from Intel (as well as smaller chipset companies like VIA). We want to determine a couple of things in this article: first, does that still hold true (at least in the mobile market)? Second, even if AMD Radeon Mobility X1250 (in this case) is faster than GMA X3100, does it even matter? In other words, is the performance provided enough to actually run certain applications (games) that fail to run on competing hardware?

To help answer this second question, we will also be including gaming performance results from a Gateway laptop (E-475M) that includes a Radeon Mobility HD 2300 discrete graphics chip. We're not yet ready to complete our review of the Gateway laptop, but we should have that ready within the next couple of weeks. The discrete graphics chip adds about $80 to the price of the laptop, which isn't too bad provided the performance increase is substantial. The Gateway E-475M was also equipped with a T7300 and 2GB of memory, so it ends up acting as the discrete GPU version of the dv6500t (which is also an option from HP).

Rather than starting with tons of graphs, we thought it would be easiest to just use a table to summarize the performance differences. All games tested were run at either low (or in some cases very low) and medium detail settings indicated by LQ/VLQ and MQ in the following table. Generally speaking, low-quality means that we turned everything off, although in some games that provide a VLQ/Minimum detail setting we may still run at low-quality if performance is acceptable. Medium quality sets everything to a middle value where possible. Here are the results.

GPU Performance Comparison
Game	X1250 vs. GMA X3100	HD2300 vs. GMA X3100	HD2300 vs. X1250
Battlefield 2 LQ	-10.25%	200.33%	238.19%
Battlefield 2 MQ	3.01%	161.00%	155.28%
Bioshock	N/A	N/A	N/A
Company of Heroes LQ	15.01%	146.80%	113.97%
Far Cry LQ	55.97%	245.60%	124.73%
Far Cry MQ	62.78%	270.93%	128.66%
FEAR LQ	29.70%	108.74%	60.88%
HL2: Episode One LQ	52.33%	234.92%	119.75%
HL2: Episode One MQ	31.16%	115.90%	64.72%
HL2: Lost Coast LQ	50.58%	219.22%	111.81%
HL2: Lost Coast MQ	26.96%	160.30%	104.86%
Quake 4 VLQ	9.74%	269.64%	235.14%
Quake 4 MQ	-25.38%	188.24%	286.27%
Oblivion LQ	N/A	N/A	136.37%
STALKER LQ	8.41%	180.88%	158.89%
Supreme Commander LQ	-11.64%	46.73%	66.04%
.
Average Performance Change (LQ)	22.21%	183.65%	136.60%
Average Performance Change (MQ)	19.71%	179.27%	147.96%
Average Performance Change (Total)	21.31%	182.09%	140.37%

Starting with the IGP comparison, we find that AMD does indeed continue to place ahead of Intel in overall performance. Somewhat interesting to note, however, is that Intel does manage to run a couple of games faster. Supreme Commander is extremely CPU intensive, which may help to explain that particular result, but most of the remaining games should be pretty much GPU bottlenecked. Battlefield 2 was at one point completely unable to run on the GMA X3100, as were many of the other games. Over the past several months, Intel has continued to improve the drivers and we're now at the point where nearly all of the games ran without issue. Battlefield 2 at medium quality still had some graphical artifacts, so that result should be disqualified, but performance at medium quality is too slow regardless.

Given the improvements we've seen with updated drivers, we would actually go so far as to say that Intel could probably be equal to or slightly faster than the X1250 with the GMA X3100 if they could only optimize their drivers further. That may be surprising to hear, but in reality the Intel GPU has as many pipelines as the Radeon X1250, and current results in Battlefield 2, 3DMark, and a few other titles indicate that there's still untapped potential. Performance under Windows XP tends to be even better, as those drivers are more fully developed. (We will include results from a laptop running Windows XP using the GMA X3100 in a forthcoming article.) As it stands, however, AMD still has about a 20% performance advantage in the IGP sector. That really isn't much, especially considering the relatively low frame rates we're already talking about.

The 20% performance lead looks even less impressive in light of the performance of the Radeon Mobility HD 2300. Frankly, the HD 2300 still isn't particularly fast, and most games need to be run at medium or low detail levels in order to achieve acceptable frame rates at resolutions up to 1280x800. However, while the performance of the HD 2300 might pale in comparison to faster desktop offerings, it generally turns in performance figures that are two or three times faster than either of the IGPs we're looking at today. It also offers complete SM3.0 support along with DirectX 10 capability, though not surprisingly the DX10 support is more of a feature checklist item than anything truly useful right now - of the few DX10 enabled games currently available, most cause pretty severe performance drops even on top-end hardware like the GeForce 8800 and Radeon HD 2900.

Our conclusion as usual is that for $80 more, anyone that actually intends to play any 3D games on a laptop should at least invest in an entry level discrete GPU. Even better would be a midrange HD 2600 or GeForce 8600M/8700M, though those tend to be found in laptops that cost closer to $1500 (barring sales and other special offers - as usual, shop around). You can look at the detailed performance charts to see exactly how slow the IGP solutions run, but there are several titles that are completely unplayable even at minimum detail settings. There are also games like Bioshock and Ghost Recon: Advanced Warfighter that require SM3.0 and are incapable of running properly on either of these two IGPs. (We did try the SM2.0 hack to get Bioshock to run, but the results weren't pretty to say the least.)