If you followed our coverage of AMD’s Trinity launch last month, you may recall that our benchmarking efforts were thwarted by an unexpected motherboard fire. Such events are problematic as far as launch-day coverage is concerned, but it also gave us time to consider some of the different use cases for AMD’s mainstream desktop processor.
The A10-5800K’s low price point, plus its integrated HD 7660D GPU, opens up some intriguing options compared to the current crop of Intel processors. Intel’s Core i5-3550, the CPU we tested against, currently retails for $209. It’s a 3.3GH part with a 3.7GH Turbo Boost and no Hyper-Threading. We used the Core i5-3550 because it’s the lowest-end Intel chip we had available. A Core i3-3220 would have been the most appropriate chip, price-wise. We’ll address the theoretical value proposition of the Intel chip after we examine how the A10-5800K performs against a substantially more expensive CPU.
What we’ve done is pair the A10-5800K with a low-end Radeon 6570 GPU (combined with the APU’s integrated 7660D), and compared it against the Core i5-3550 + Radeon 6570. At $55-$70, the 6570 is exactly the sort of low-end card a regular user might opt for to juice up a low-cost desktop. The Radeon 6570 is based on the older “Turks” GPU; it offers 480 shader processors and 1-2GB of RAM (our card had 1GB). Both systems used the same 8GB of DDR3-1333 RAM. Total cost of this baseline configuration was $304 for the Intel system, $224 for the AMD rig. We used an MSI FM2-A85XA-G65 motherboard. The board offers a full set of tweakable features for overclockers and tuners; we were quite pleased with its overall performance.
Then, to make things interesting, we swapped out the RAM on the AMD system as well. AMD’s APUs are capable of benefiting significantly from additional RAM bandwidth; we wanted to see if that trend would hold true in a combined GPU configuration. With 8GB of DDR3-2133, the total AMD cost rises to $244.
We ran play tests in four games: Orcs Must Die 2 (I’m a huge fan of the OMD series), Batman: Arkham City, The Elder Scrolls: Skyrim, and Left 4 Dead 2. Does the A10-5800K’s ability to leverage a dual-GPU actually translate to a real-world performance advantage?
Vsync was turned off in all cases, all games were benchmarked in 1920×1080, and AMD’s Catalyst Control Panel was configured for “Quality” texture filtering. Anisotropic filtering was enabled in-driver rather than forced on via the Control Panel.
Orcs Must Die 2 is an enjoyable simple first-person tower-defense title. We benchmarked the game’s “Twisted Halls” finale map at 1900×1080, with graphics details set to High and 4xAA. Average frame rates are in the first graph, minimum frame rates in the second.
We chose “Twisted Halls” because it’s a map that can noticeably chug on lower-end graphics cards once a full array of traps are set. This is a game where AMD’s higher minimum frame rates make more of a difference than the increased average. Both systems hit low points at various times, but the 5800K’s 23 FPS was 36% faster than what we saw with the Intel system.
Next up, Batman: Arkham City. The game was set for High detail (DX11 was disabled). PhysX and Ambient Occlusion were both disabled, FXAA was enabled in the game’s settings. We benchmarked Arkham City in two modes. First, a single-room challenge map (Prison Riot), and second while soaring over the streets and alleys of Gotham City.
In Arkham City, the AMD system is 14% and 18% faster in our two scenarios. The minimum Prison Riot frame rate is 9% higher; the minimum City Flight frame rate is 24% higher.
In fairness to Intel, though this isn’t easily captured in a frame rate test, the transitions from map to map — when moving between rooms in the same building, for example, were smoother with the Core i5.
The Elder Scrolls: Skyrim
Skyrim was configured for Medium details, with no AA and 8x AF enabled. Standard textures were installed (the game offers a high-resolution texture pack as well). Of all the games we tested, Skyrim saw the greatest performance boost from DDR3-2133 as compared to DDR3-1333. The A10′s average frame rate was a full 40% faster than the Core-i5 3550′s; its minimum frame rate was 28% higher.
Left 4 Dead 2
Finally, there’s Left 4 Dead 2. We tested the game with Very High shaders and High Detail, with multi-core rendering enabled, 4x MSAA, and 16x AF.
Again, AMD outperformed Intel here — L4D picked up a nice boost to its average frame rate from the faster RAM, even if the minimum FPS didn’t change much.
What about other tweaks?
One of the theoretical advantages of buying an A10-5800K is that the chip is fully unlocked out of the box. We experimented with various settings and overclocking options, but saw mixed results. Increasing the integrated memory controller’s speed to 2200MH , up from 1800MH , had virtually no impact on our gaming tests.
Interestingly, overclocking the integrated GPU had virtually no impact on Combined performance. We confirmed that the GPU was running faster with various GPU compute tests, but in games, there was no difference. Evidently AMD’s internal profiles keep most of the GPU workloads on the discrete card.
Setting the DDR3-2133 to 1T operation boosted performance by roughly 5% in some games, but Skyrim would crash almost instantly. Disabling Turbo Boost and running the entire CPU at 4.2GH gave a 3-5% FPS boost in some tests. Results from around the internet suggest that the A10-5800K can hit 4.6GH with a solid third-party cooler — that extra 10% won’t do your power bill any favors, but you can squee e a bit more performance out of the chip if you’ve got the hardware.
A modest proposal
Our results indicate that AMD’s Trinity-based desktop APUs may be a better option for gamers on a budget, but there are some significant caveats to consider. AMD’s performance advantage over Intel is limited to GPU configurations that combine the on-die GPU with a discrete solution. Paired with a high-end GPU, Intel CPUs are significantly faster than their AMD counterparts.
Swapping the Core i5-3550 with a hypothetical Core i3-3220 would eliminate the price discrepancy between Intel and AMD, but it wouldn’t make the Intel system faster in the tests above. Even if we assume both Intel CPUs performed identically, the AMD chips would have a 10-15% edge when paired with a low-end GPU that could make use of Trinity’s Combined Graphics mode.
The flip side to this is that an Intel system with a Core i3-3220 is a more certain upgrade path. Enthusiasts can buy a low-end Intel system today, knowing they’ll be able to upgrade to a CPU that can push a high-end modern GPU to its limit. For AMD buyers, that’s not a given. Sunnyvale has promised that the FM2 socket will be supported by Kaveri and the Steamroller-based APUs that are currently expected to debut in 2014, but Steamroller’s performance isn’t known.
Prospective buyers should be aware that while Steamroller is expected to significantly improve on Trinity’s single-thread performance, it is unlikely to close the gap with the Haswell/Broadwell parts Intel will be shipping by that date. Best case CPU improvement, based on what we currently know, would be approximately 15%. That’s not enough to unseat Intel; Santa Clara remains the better option if you want serious upgrade potential.
Budget-constrained gamers who want to maximi e current performance and aren’t concerned about future upgrades should give the A10-5800K a look. It’s not the best option for everyone, but there’s a valid argument for adopting AMD over Intel for gaming in this market segment. We’re the first to admit that it’s a pretty specific use scenario, but the performance gaps are large enough to support it.
AMD’s Trinity analy ed: A surprisingly good, inexpensive gaming solution
0 comments:
Post a Comment