Can AMD make magic? Check out our in-depth Vishera benchmarks.
On paper, AMD’s Bulldozer microarchitecture always sounded like a mean, green machine. When it landed last year, though, in the form of the Zambezi processor (aka FX-8150), it actually went about as fast as a bulldozer.
AMD didn’t just give up and curl into a ball. The company went back to work polishing the FX chip into the new AMD FX-8350 “Vishera.” The chip might look like a Zambezi, but it features an improved branch predictor, improved scheduler, larger L1 translate lookaside buffer, new FMA3 and F16C instructions, L2 improvements, among many other changes.
Vishera looks the same externally and the good news: it’ll use the same AM3+ socket too.
The best feature of Zambezi though is its backwards compatibility. Zambezi should work with most if not all AM3+ boards with a BIOS update and AMD is promising future chips will work with it as well.
To see how Zambezi stacks up, we used the same Asus Crosshair V board that we tested Bulldozer with more than a year ago and outfitted it with a GeForce GTX 580, 8GB of dual-channel DDR3/1600, OCZ Vertex 3 SSD and Windows 8. Why Windows 8? One issue that cropped up with the original Bulldozer chip was Windows 7’s scheduler didn’t know how to deal with the Bulldozer’s shared multi-cores.
For a comparison chip, we dusted off our old FX-8150 and set up a near identical Intel system using Intel’s Core i5-3570K on an Asus P8Z77-V Premium motherboard and Windows 8 as well. Why the 3570K? It’s the chip AMD uses as the benchmark for the FX-8350 and, frankly, our recommendation for the sweet spot of computing today.
The result? First, FX-8150 is still slow. It could barely compete with the Core i5-2500K last year and takes it on the chin in just about every test here from the Ivy Bridge-based Core i5-3570K. In fact, it was so slow in our Premiere Pro CS6 encode that we had to rerun the tests on both the Intel and FX-8150 because we couldn’t believe it.
Vishera at a Glance
You can’t deny the power of the individual Ivy Bridge cores though. We ran Cinebench 10 on a single core and the Ivy Bridge slaughtered both FX CPUs. In several of the gaming benchmarks, Intel’s more efficient cores also put it on top although we did see the FX parts unexpectedly pull ahead in the graphics department which are usually all the same when the same GPU is used.How about the FX-8350? Far better. In fact, we’d dare say the FX-8350 is very competitive with the 3570K in some heavily multi-threaded tasks and offers significant performance improvements over the FX-8150. Take for example, our Premiere Pro CS6 benchmark. It doesn’t trounce the 3570K, but it cuts the encode time in half over the FX-8150 part. In other tests it aces the Ivy Bridge part.
What do we recommend? If your chores are mostly limited to gaming and tasks that can’t exploit all eight cores, the Intel part has the advantage. If, however, you are rendering 3D, transcoding or rendering video (except in Adobe’s Premiere Pro CS6), the new FX-8350 should be your pick. It offers a longer socket roadmap and gives you better performance in multi-threaded apps. We do have to add though, that the performance gap probably isn’t as good as AMD fans would have hoped for considering the clock and core count difference between the FX-8350 and Core i5-3570K. Still, these days for AMD, a tie is probably good news considering its up against Intel’s best cores to date.
For a deeper dive on how well Vishera performs, click on for our individual benchmark highlights.
The older version of Cinebench lets you run the 3D rendering engine across all available threads in the CPU or just one. We opted here to run one to see how efficient each “core” in the new FX-8350 is. Vishera has a small advantage over Bulldozer which AMD says half comes from the higher clocks, and the other half from the under-the-hood improvements. The bad news though is pretty clear from this graph. Despite slightly lower clocks, the Ivy Bridge cores in the Intel part is all over Vishera and Bulldozer.
Winner: Ivy Bridge
To gauge the performance of the CPUs when all threads are tapped, we ran Cinebench 11.5 using Maxon’s rendering engine across the chips. The eight-core FX-8150 Bulldozer gets stomped on by the four-core Core i5-3570K but Vishera gets vengeance by acing the Intel chip by a healthy clip.
The POV Ray benchmark is a freeware ray tracer that has its roots on the Amiga. The benchmark exploits all available threads on a CPU and the advantage should go to the 8-core chips. And no surprise it does. Vishera wins this one by a wide margin although, with but four-cores at its back, Ivy Bridge doesn’t too poorly.
Click the next page for 3DMark 11 and gaming benchmarks.
This one actually surprised us a bit but both Bulldozer and Vishera aced the Ivy Bridge chip. Since we used the same GeForce 580 card with the same drivers for all of the tests, we didn’t expect much of a difference.
Winner: Vishera and Bulldozer
When we saw the result of the physics test in 3DMark 11, we were even more confused. The physics test in 3DMark11 is mostly about the CPU and more cores generally win. But here we see the Intel CPU ahead, albeit slightly, over both AMD cpus.
Winner: Ivy Bridge
Our head scratcher got even weirder once we saw the results of the 3DMark 11 graphics test. Again, we used a matching set of GeForce GTX 580 cards which we verified was running at the same speed on each platform and the same drivers. In the last five or six CPU showdowns, this usually leads to the same graphics score since it’s all about the GPU. In fact, even though we have recorded those scores, we generally don’t report the 3DMark11 graphics score since they’re almost virtually the same. Yet both AMD chips have a clear advantage. If we had used an ATI card, we could attribute it to secret sauce performance optimizations from the CPU and graphics teams but we used an Nvidia graphics card. And we know Nvidia and ATI like each other as much as much as Tupac and Biggie. Frankly, we have no idea why there’s such a difference but it’s possible better drivers and/or better PCIe performance in the 990x chipset are the reason.
Winner: Vishera and Bulldozer
To see if that graphics performance translated into more real-world gaming tasks, we ran all three chips using the STALKER: CoP benchmark at 1366x768 resolution to intentionally not make the graphics card the bottleneck. We are running with DX11 mode on, but no tessellation and contact hardening shadows off. The other graphics settings are also at the default settings and we report the SunShafts result only.
Winner: Ivy Bridge
Valve created its Particle Test benchmark using the Source Engine almost six years ago when the first quad-core CPUs were introduced. We used to think the test topped out with quad-cores but lately we’ve been seeing this benchmark run faster the more threads you give it. For example, the six-core Sandy Bridge-E chips with Hyper-Threading slaughter all others in this benchmark. So would the eight-core FX procs take it? No. Not at all. The Ivy Bridge eats both AMD parts despite the higher core counts.
Winner: Ivy Bridge
Our final game benchmark involved using the popular Heaven 3.0 benchmark. We run this at 1366x768 with tessellation set to medium. Even so, this is a pretty hefty GPU-centric benchmark. Overall, if we had to assign a winner, it would be the Ivy Bridge chip, but not by much of a margin. It’s pretty close between it and the Vishera CPU. Even the Bulldozer pulls up some respectable numbers. This probably represents the best lesson here when it comes to CPUs and gaming: Gaming today is 80 percent GPU. If we were gamers, we’d put more money in the GPU than the CPU for most games.
Click the next page for video/media-intensive benchmarks.
AMD has been pushing an OpenCL-based version of Handbrake as a benchmark but the beta was available to us at presstime. Instead, we stuck to the publically available 0.9.8 version which is heavily multi-threaded. We took the decrypted contents of the Blu-ray Super Troopers and timed how long it would take to convert the file to ah H.264 file using the high-profile setting. This test shows us that whatever silicon changes AMD made to Vishera, they paid off. There’s a pretty sizeable performance difference between Bulldozer and Vishera that can’t be explained just by clock speeds. Ivy Bridge also has a good showing but Vishera easily out paces it. Bulldozer, however, is pulling up the rear.
We’ve been using TechArp.com’s X264 HD 5.0.1 benchmark to measure performance on new PCs. The test does two passes using the freeware x264 encoding library. The first pass is seemingly a little more sensitive to clock speeds and memory bandwidth rather than just pure core count. A higher frame rate is better and the Vishera pulls out ahead of the other two chips. Again, Ivy Bridge doesn’t do bad at all but it is faster than the older Bulldozer chip.
The second pass in TechARP’s X264 HD 5.0.1 benchmark is more sensitive to core count and the hips stack up the way we’d expect them to given the number of cores in each. Although, we are seeing two things here: the first is how efficient those Ivy Bridge cores are even up against the eight-core Bulldozer part. The second: The changes AMD made to the Piledriver cores pay off well in encoding tests.
Vishera has thus far won all of the encoding tests where core count matters. Would that translate into our punishing Premiere Pro CS6 benchmark? Flat out no. For this test, we use Adobe’s awesomely fast Premiere Pro CS6, feed it several 1080p videos shot with an EOS 5D MkII and encode it as a Blu-ray video. The video features 1080p video with moving picture-in-picture streams of 1080p moving across the frame. It’s a severe nut buster for even the heftiest of CPUs. We use the Maximum Render option and set it to encode using the host CPU rather than the CUDA-based GPU that we have installed. Interestingly, the Ivy Bridge part kicked the holy hell out of both AMD parts. The Vishera part did markedly better but still was a distant second. The performance of the Bulldozer was so bad, we ran several additional times to see if we had somehow misconfigured our encode. In fact, the performance of the Ivy Bridge part was so fast, we also retested it multiple times to ensure that we hadn’t accidentally rendered the project on the GeForce GTX 580 (which, by the way, smokes the CPUs.)
Winner: Ivy Bridge
We’ve long been using Proshow Producer to gauge CPU and system performance. The latest iteration supports GPU accelerated previews, but for rendering, it’s all about the CPU. The current engine seems to top out at four threads though and adding additional core or Hyper-Threading doesn’t help much. Why do we use it? The truth is that very few applications will soak up the number of cores and threads offered by a Vishera or six-core Sandy Bridge-E chip so we consider this a good way to illustrate performance that you’d get on a typical application. Hey, it could be worse—the vast majority of apps still don’t use more than a single-thread. The efficiency that we’ve been seeing from the Ivy Bridge cores pay off again as it moves past both AMD chips. It’s not a killing, but enough that there is a clear winner.
Winner: Ivy Bridge.
Stitch.Efx 2.0 is another of our new benchmarks. We picked it because it’s a real-world application and actually offers an interesting take on threading. First, we take around 200 images shot with a Canon EOS 7D on a GigaPan motorized head. That’s roughly 1.9GB of image files. The images are then fed into Stich.Efx 2.0 where they are stitched into a giant gigapixel image. The first third of the render where the images are stitched together is single-threaded. The final two-thirds where the image is blended together is multi-threaded. We’ve found that this test generally favors clock speed and a quad-core with Hyper-Threading seems to offer the optimal foot print. The efficiency of the Ivy Bridge chips, which we’ve seen drub the FX chips all day, again shows its head here. There is a silver lining though: again, Whatever changes AMD did to the chip are paying off as the advantage Vishera has over Bulldozer can’t be explained by pure clock speeds.
Winner: Ivy Bridge.
Click the next page for the final test and complete benchmark chart.
We’ll end this on a good note for AMD. Our final test highlight is the Vishera’s performance in 7-Zip’s built-in benchmark. This measures how theoretically faster each chip would be at compressing files using the 7Zip engine and it scales nicely with core and thread counts. The Vishera is first, Bulldozer second and the Ivy Bridge part a clearly third place.