AMD hosted its financial & analyst day today, revealing information on Vega, Threadripper, notebook deployments of its CPUs & GPUs, and data center products. Some timelines were loosely laid-out with initial benchmark previews, provided an outline for what to expect from AMD in the remainder of 2017.
Most of our time today will be spent detailing Vega, as it’s been the topic of most interest lately, with some preliminary information on the CPU products.
After pointing out that Intel’s budget-option Pentium G4560 CPU somewhat invalidates the Intel i3 lineup, particularly when that lineup is flanked by i5s and R5s, the next question was how good of a GPU can be paired with the G4560. Someone buying a $70 CPU won’t likely be buying a GTX 1080 – and probably not a 1070 – but we wanted to see how far up the scale we could go before encountering a CPU bottleneck. This kind of test has all manner of variables, naturally, so we’ve done our best to constraint them; the biggest is that of the games tested. Depending on graphics settings, GPU constraints could be imposed all over the place. We decided to opt for what we thought to be a somewhat realistic test: We took the G4560, paired it with GPUs ranging from ~$115 to ~$600, and then configured graphics to high/ultra with a 1080p resolution. We then included titles that are known to CPU choke, titles known to be more GPU constrained, and titles balanced in the middle. This gives a wide berth of tested content (FPS, RTS, and popular titles) from which we can draw some conclusions.
We are using the Pentium G4560 for this test, naturally. Included in our Intel Pentium G4560 GPU bottleneck test are the following GPUs (listed in order of price):
Intel’s i3/i5/i7 and AMD’s R5/R7 CPUs are the big competitors in the PC gaming world, but they aren’t the only options out there: AMD released cheap but capable Athlon X4s in 2016, and in January of this year Intel released the 2C/4T Pentium G4560 ($70), a 14nm Kaby Lake processor for ~$64~$70. We didn’t fully review the older and (briefly) popular Pentium G3258, but it has showed up in Ask GN and individual benchmarks, so we were excited to do comprehensive testing on this modern iteration.
The G4560 lacks the feature that made the G3258 so popular: the ability to overclock. Buying a dirt-cheap dual-core processor and cranking the frequency up was enough for decent performance in limited-thread games, although the G3258 often suffers from extreme stuttering in more modern titles. The limitations lead us to believe that Intel doesn’t want to compete with its own more expensive 2C/4T unlocked i3 and locked i3-7100 ($120) & 7300 ($150).
We came away from our revisit of the once-king Sandy Bridge 2600K and 2500K CPUs impressed by the staying power of products that came out in Q1 2011, considering Intel’s unimpressive gains since that time.
At the time of Sandy Bridge’s release, AMD’s flagship CPUs were 45nm K10-based Phenom IIs, designed to compete in price/performance with the 45nm Lynnfield (Nehalem i5) quad cores. Later that year, AMD’s underwhelming Bulldozer architecture would launch and inevitably replace the Phenom line. Given that we’ve already looked at Intel’s 1Q11 offerings, we decided to revisit AMD’s Phenom II CPUs in 2017, including the Phenom II X6 1090T (Black Edition) and Phenom II X6 1055T. These benchmarks look at AMD Phenom II performance in gaming and production workloads for the modern era, including comparisons to the equal-aged Sandy Bridge CPUs, modern Ryzen 5 & 7 CPUs, and modern Intel CPUs.
Prior to the Ryzen launch, we discovered an issue with GTA V testing that would cause high-speed CPUs of a particular variety to stutter when achieving high framerates. Our first video didn’t conclude with a root cause, but we now believe the game is running into engine constraints – present on other RAGE games – that trigger choppy behavior on those CPUs. Originally, we only saw this on the best i5s – older gen i5 CPUs were not affected, as they were not fast enough to exceed the framerate limiter in GTA V (~187FPS, or thereabouts), and so never encountered the stutters. The newest i5 CPUs, like the 7600K and 6600K, would post high framerates, but lose consistency in frametimes. As an end user, the solution would be (interestingly) to increase your graphics quality, resolution, or otherwise bring FPS to around the 120-165 mark.
Then Ryzen came out, and then Ryzen 5 came out. With R5, we encountered a few stutters in GTA V when SMT was enabled and when the CPU was operating under conditions permitting the CPU to achieve the same high framerates as Intel Core i5-7600K CPUs. To better illustrate, we can actually turn down graphics settings to a point of forcing framerates to the max on 4C/8T R5 CPUs, relinquishing some of the performance constraint, and then encounter hard stuttering. In short: A higher framerate overall would result in a much worse experience for the player, both on i5 and R5 CPUs. The 4C/8T R5 CPUs exhibited this same stutter performance (as i5 CPUs) most heavily when SMT was disabled, at which point we spit out a graph like this:
Following our in-depth Ryzen VR benchmark (R7 1700 vs. i7-7700K with the Rift + Vive), we immediately began compiling results for the concurrent R5 test efforts by GN Sr. Editor Patrick Lathan. Working together, we were able to knock-out the VR benchmarks (check those out here – some cool data), Ryzen Revisit piece, and today’s R5 reviews.
Both the R5 1600X ($250) and R5 1500X ($190) CPUs are in for review today, primarily matched against the Intel i5-7500 and i5-7600K. For comparison reasons, we have still included other CPUs on the bench – notably the i7-7700K and R7 1700, just to give an understanding of what the extra ~$70-$130 gets.
For anyone who hasn’t checked in on our content since the initial Ryzen reviews, we’d strongly encourage checking the Ryzen Revisit piece for a better understanding of how the scene has changed since launch. That revisit looks at Windows updates (and debunks some myths), EFI updates, and memory overclocking impact on Ryzen performance.
Although we have rerun the R7 gaming benchmarks with higher memory frequency (thanks to GSkill and Geil for providing B-die kits), we have not yet rerun them in synthetic tests. The 2933MHz frequency, as a reminder, was a hard limitation on our test platforms in the initial round of R7 reviews.
We will be including that data (albeit truncated) in our new tests, alongside Intel retests for the same games. For now, though, we’re reviewing the R5 1600X and R5 1500X CPUs in the Ryzen family, priced at $250 and $190, respectively.
This content marks the beginning of our in-depth VR testing efforts, part of an ongoing test pattern that hopes to determine distinct advantages and disadvantages on today’s hardware. VR hasn’t been a high-performance content topic for us, but we believe it’s an important one for this release of Kaby Lake & Ryzen CPUs: Both brands have boasted high VR performance, “VR Ready” tags, and other marketing that hasn’t been validated – mostly because it’s hard to do so. We’re leveraging a hardware capture rig to intercept frames to the headsets, FCAT VR, and a suite of five games across the Oculus Rift & HTC Vive to benchmark the R7 1700 vs. i7-7700K. This testing includes benchmarks at stock and overclocked configurations, totaling four devices under test (DUT) across two headsets and five games. Although this is “just” 20 total tests (with multiple passes), the process takes significantly longer than testing our entire suite of GPUs. Executing 20 of these VR benchmarks, ignoring parity tests, takes several days. We could do the same count for a GPU suite and have it done in a day.
VR benchmarking is hard, as it turns out, and there are a number of imperfections in any existing test methodology for VR. We’ve got a good solution to testing that has proven reliable, but in no way do we claim that perfect. Fortunately, by combining hardware and software capture, we’re able to validate numbers for each test pass. Using multiple test passes over the past five months of working with FCAT VR, we’ve also been able to build-up a database that gives us a clear margin of error; to this end, we’ve added error bars to the bar graphs to help illustrate when results are within usual variance.
AMD today made available a power plan update which should change how the Balanced plan impacts Ryzen performance.
Problems with Windows preset power modes have been one of the biggest annoyances with Ryzen, and AMD has officially recommended the High Performance preset in the past in order to avoid subpar performance in benchmarks. This wasn’t a big deal from a testing point of view since High Performance mode effectively avoids all of these issues, but for everyday use, it was: High Performance mode doesn’t allow CPU frequency to drop when idle, and the additional power consumption can really hurt the long-term value of the system (it’s also just wasteful). Balanced mode does drop frequency, but it’s also been overly aggressive with core parking on Ryzen chips specifically, making it sub-optimal for use. We discussed what this looks like from a user’s point of view in our “Just Research” article, where frequency plots offer visualization for the impact of Performance vs. Balanced mode. The same article contains some FPS benchmarks between the two power modes.
AMD has made two major changes in this update. Quoting their statement:
Maintain residency in CPU p0 or p1 to give Zen full control over clocks and volts.
Disable core parking.
They specifically noted that Intel also fully disables core parking in the Balanced power plan. Our tests have always used High Performance mode for Ryzen platforms (except power tests), and our results will not be affected by this update.
We’ve received a ton of positive feedback on our i5-2500K revisit, and we’ve received a similar amount of questions about including overclocked i7-2600K numbers in our benchmark charts. The solution is obvious: a full 2600K revisit using our modern benchmark course. As demonstrated with the 2500K, old K-SKU Sandy Bridge CPUs had impressive overclocking capacity--partly thanks to a better thermal solution than what Intel offers today--but the stock i7-2600K regularly outperformed our 4.5GHz 2500K in some tests. Synthetic benchmarks and games like Watch Dogs 2, both of which take advantage of high thread counts, are included in those tests showing favor to the 2600K.1
Although we ended the 2500K review with the conclusion that now is a good time to start thinking about an upgrade, i7 CPUs are considered as more future-proof. Today, we’re testing that conception to see how it holds up to 2017’s test suite. With Ryzen 7 now fully released, considering 2600K owners are likely looking (price-wise) at a 7700K ($345) or 1700 ($330), it makes sense to revisit SNB one more time.
Note: For anyone who saw our recent Ryzen Revisit coverage, you know that there are some fairly important changes to Total War: Warhammer and Battlefield 1 that impacted Ryzen, and could also impact Intel. We have not fully retested our suite with these changes yet, and this content was written prior to the Ryzen revisit. Still, we’re including some updated numbers in here – but it’s not really the focus of the content, we’re more interested now in seeing how the i7-2600K performs in today’s games, especially with an overclock.
This first revisit to Ryzen’s performance comes earlier than most, given the tempestuous environment surrounding AMD’s latest uarch. In the past weeks, we’ve seen claims that Windows updates promise a significant boon to Ryzen performance, as has also been said of memory overclocking, and we were previously instructed that EFI updates alone should bolster performance. Perhaps not unrelated, game updates to major titles could have potentially impacted performance, amounting to a significant number of variables for a revisit.
Today’s content piece aims to isolate each of these items as much as reasonable – not all can be isolated, like game updates – to better determine the performance impact from the individual changes and updates. We’ll then progress cumulatively through charts as updates are applied. Our final set of charts will contain Windows version bxxx.970, version 1002 EFI on the CH6, and memory overclocking efforts.