Summary metrics and deep voodoo
Let's see how the move to Windows 7 affects the 7950's placement on our famous value scatter plots. As ever, we've converted our 99th percentile frame time results into FPS, so both of the plots below can be read the same, with the best values being closer to the top left corner of the plot area.
Predictably, there's not much movement, because the 7950's performance doesn't change much from Win8 to Win7. Also, since Windows 7 doesn't alleviate the latency spikes that plague the 7950 in many of these newer games, the Radeon continues to drop into a much less desirable position on the 99th-percentile plot than the GTX 660 Ti.
We should note that the overall performance number above is a geometric mean of the results from our seven game test sequences. (We've excluded Whiterun since we didn't use it last time around.) We started using the geomean back in August in order to reduce the impact of outliers on our overall performance scores. In the past, games with vendor-friendly optimizations like HAWX 2 and DiRT Showdown tended to push the average a long way in one direction, so we excluded them from our calculations, prompting controversy. We were hopeful the switch to the geometric mean would curb outliers without manual intervention.
That said, some folks still objected to our inclusion of Assassin's Creed III in our overall index, since it performs so much better on the GeForce than on the Radeon. I figured we might as well oblige them by taking a look at the overall scores with AC3 excluded, to see whether it moves the needle.
Well, things don't change drastically, but excluding AC3 allows the 7950 to pull ahead of the GTX 660 Ti in our overall FPS average. In fact, that change produces virtually the same sort of outcome in the overall FPS numbers as we saw back in August, when the Radeon HD 7950 reference card with Boost edged out this same Zotac GTX 660 Ti AMP! card by a margin of several frames per second.
Trouble is, that doesn't really matter. A moral victory in the borderline-meaningless FPS sweeps doesn't overcome the fact that the Radeon HD 7950 has a persistent problem with high-latency frames across a range of test scenarios based on the latest games. The 99th-percentile frame times reflect that reality. Our latest round of tests shows that Windows 8 is not the problem. On the contrary, Windows 8 generally improves the latency picture somewhat.
When we first published our rematch between the 7950 and the GTX 660 Ti, we pinged AMD to ask if they could explain the Radeon's struggles in recent games. AMD spokesman Antal Tungler told us that our article had "raised some alarms" internally at the company, and he said they hoped to have some answers for us "before the holiday." He also noted that AMD is continually working to improve its drivers and that the 7950 does perform well in FPS-based benchmarks.
We're hopeful that we may have a more detailed answer from AMD before too long, but in the interim, we have an even firmer grasp of the reality that caused us to recommend the GeForce GTX 660 Ti over the Radeon HD 7950 in our last article. Again, the outcome of our testing may be counter the expectations of many folks; they certainly weren't what we expected when we set out to stage this rematch.
The tragedy here is one of wasted potential. The Radeon HD 7950 is, on paper, clearly a more powerful GPU than the GTX 660 Ti, with double the memory interface width and a theoretical edge in peak ROP rate and shader flops. AMD is giving you more hardware for your money when you buy a 7950. For whatever reason—and we suspect the main culprit is graphics driver software—the 7950 can't convert that advantage into consistently smoother in-game animation. As one of my fellow TR editors pointed out to me the other day, this wouldn't be the first time Radeon owners were let down by driver issues during the holiday rush. AMD was plagued by a painful series of driver issues last year, too.
To those who would claim that other "professional" review sites haven't seen results like ours in their comparisons of the 7950 and GTX 660 Ti, I would simply respond: of course not. Virtually nobody else tests like we do. We're working on persuading folks to focus on latency, to use a timer with finer granularity, but such changes are hard. They take time and effort. Heck, we may be nearly alone in using this approach for a long time yet.
The question you have to ask is what matters to you. Do you want the graphics card that scores best in the FPS beauty pageant, or do you want that one that gives you the smoothest gaming experience when you fire up a freshly downloaded game this Christmas? If you just want bragging rights, by all means, choose the Radeon HD 7950. If you're looking for the friction-free fluidity that only comes from consistently quick frame delivery, though, our recommendation remains the GeForce GTX 660 Ti.
213 comments — Last by Rigel84 at 8:00 AM on 01/02/13
|Radeon Software Crimson ReLive Edition 17.7.2 boasts refinements galoreTidying up ahead of RX Vega||9|
|Corsair's Hydro GFX GeForce GTX 1080 Ti graphics card reviewedNo assembly required||28|
|Computex 2017: Corsair goes high-conceptClothe your hardware in carbon and silica||20|
|G.Skill's Ripjaws KM570 RGB gaming keyboard reviewedSimplify, simplify||6|
|The Tech Report System Guide: May 2017 editionRyzen 5 takes the stage||111|
|Aorus' GeForce GTX 1080 Ti Xtreme Edition 11G graphics card reviewedThe eagle has landed||36|
|Corsair's K95 RGB Platinum gaming keyboard reviewedA lean, mean macro machine||15|
|HyperX's Pulsefire gaming mouse reviewedKeeping it simple the first time out||9|
|Silverstone shines RGB LEDs on the Mini-ITX RVZ03 chassis||3|
|Radeon Software Crimson ReLive Edition 17.7.2 boasts refinements galore||9|
|Cooler Master gives the MasterBox Lite 5 case an RGB makeover||2|
|USB 3.2 spec pushes bandwidth up to 20 Gbps||48|
|Razer Tiamat 7.1 V2 headset packs ten drivers for immersive audio||13|
|EVGA unleashes the GTX 1080 Ti K|ngp|n graphics card||23|
|Corsair sells a majority stake to private equity for $525 million||67|
|AMD turned a $25 million operating profit in Q2 2017||100|
|Rumor: Radeon RX Vega benched in 3DMark Fire Strike||67|
|edit: i'm not funny||+35|