Intel Discusses Real World Performance Focus, Teases Gen11 Graphics & i9-9900KS
In advance of its official Computex keynote, Intel on Sunday held a preview event detailing how the company is developing its software and hardware to target real-world performance improvements in upcoming products. The event, “Performance for the Real World,” was hosted by Jon Carvill, Intel’s VP of Tech Market Leadership, and Ryan Shrout*, Intel’s Chief Performance Strategist.
The event provided initial performance data for both Intel’s upcoming Sunny Cove-based Ice Lake processors overall as well as Intel Gen11 graphics specifically, which will make their debut in Ice Lake. Intel also used the event to expand its argument for making performance measurements based on more “real world” workloads and tasks.
“Real World” Benchmarks
Intel’s initial claim is that the traditional methods of benchmarking and performance analysis of PC hardware hasn’t yet adapted to the way that consumers now use their devices (i.e., we’re benchmarking it wrong). The gist is that hardware reviewers, analysts, and even hardware manufacturers themselves often measure and make important decisions or recommendations based on testing methods that don’t actually reflect modern workloads or needs.
To illustrate, Intel provided data comparing the prevalence of certain benchmarks among hardware reviewers compared to actual usage share of the apps those benchmarks are based on. For example, Intel claimed that 80% of the tech press use the Cinebench benchmark in their reviews (PC Perspective included). Yet, according to Intel’s statistics, only 0.54% of users have ever run Cinema4D, the 3D modeling and graphics rendering application on which Cinebench is based. Another example is POV-Ray, used by about 53 percent of the tech press but only 0.043% of users.
Instead, Intel argues, the company and its counterparts should make development and optimization decisions that maximize performance in the applications that are most frequently used by consumers, such as Chrome, Microsoft Office, Adobe Creative Cloud, and popular games. To be fair, Intel clarified that this argument is meant to explain how and why the company focuses on certain areas of hardware and software development and not a mandate to the tech press on how to test hardware, but since the tech press has an influence in the perception of new products and consumer buying decisions, a change to real-world testing by reviewers is clearly something Intel would like to see.
Intel’s argument is valid to a degree; from a consumer’s standpoint, the more that benchmarks reflect their intended usage the better. However, a valid counter-argument is that the class of consumers most interested in detailed PC hardware reviews would have more demanding needs — 3D rendering, ray tracing, etc. — than the “average” consumer presumably captured by Intel’s data.
It’s also the same fundamental idea Intel offered during its Cascade Lake launch earlier this year. A key reason for the messaging, of course, is increased competition in recent years. In some “traditional” workloads that are benchmarking favorites, Intel is falling behind (or about to), if not in performance outright then certainly in a performance-per-dollar comparison. Therefore, Intel’s attempt to shift the focus to fights it can win is essential for the company. That doesn’t make Intel’s efforts invalid, however, and we agree with the idea of benchmarks that reflect real world workloads, but we feel that there is still value in some of the benchmarks Intel singled out. There’s also the constant challenge of balancing the number of types of benchmarking tests with limits on resources and time. In short, a theoretical shift to benchmarks that more accurately reflect common workloads is a good thing, but we’re still going to use Cinebench.
With this real-world focus in mind, Intel touted several examples in terms of both current and upcoming products.
Ice Lake & Gen11 Graphics
Intel provided additional info on its upcoming 10nm Ice Lake platform. Although more details will follow shortly, Intel specified that Ice Lake will offer integrated Thunderbolt 3 and Wi-Fi 6 support.
In terms of Gen11 graphics, making their debut in Ice Lake, Intel showed a significant performance improvement over the current Gen9 design. In a comparison of 15-watt parts, Intel claims that Gen11 graphics are between 40 and 100 percent faster than Gen9 in a variety of popular games at 1080p resolution and low or medium quality settings.
While this represents a huge single-generation improvement (“Gen10” graphics were never productized), perhaps more impressive is that Gen11 graphics close, and in some cases exceed, the gap with AMD’s Ryzen APU performance. In a comparison between a 25-watt Ryzen 7 3700U and an unnamed 25-watt Ice Lake-U part, Intel’s performance ranges from as little as a 6 percent deficit to as much as a 16 percent advantage. AMD has long touted the higher performance graphics of its APUs as a key advantage. Intel’s ability to seemingly close the gap so quickly is huge.
It’s important to note, of course, that we thus far only have data provided by Intel on the games and benchmarks it chose to test, and we don’t yet know what lies in store for AMD’s continued development of Vega and Navi for mobile, but it’s clear that APU benchmarking and comparisons will be of major interest this year.
Intel also reiterated that Gen11 graphics will support variable rate shading (VRS), a capability that can control shading rate independent of the final image resolution, allowing for potentially large performance improvements when enabled. Intel claimed that VRS, which is not available in AMD APUs, can offer up to a 1.3X performance improvement in Unreal Engine, up to 1.2X in Civilization VI, and up to 1.4X in 3DMark.
Heterogeneous Computing, Software, and “The AI PC”
Software is a part of Intel’s new “Six Pillars” development strategy and the company spent time demonstrating how its internal software teams have worked with third party developers to improve performance of both current and upcoming products, especially with a focus on the new era of heterogenous computing.
An example of this type of collaboration is Epic’s Chaos physics demo revealed earlier this year at GDC 2019. Intel worked with Epic to optimize low level solvers, data structures, and thread parallelism. As a result, Intel’s 18-core/36-thread i9-9980XE is more than 40 percent faster in terms of frames per second than AMD’s 32-core/64-thread Threadripper 2990WX.
Also of note is Intel’s focus on AI capabilities. Features such as Deep Learning Boost (DL Boost), first introduced in enterprise SKUs, are making their way to consumer parts alongside the introduction of consumer-facing applications that can take advantage of them. With DL Boost and other architectural improvements, Ice Lake will offer significant performance advantages over its Whiskey Lake-based predecessors as well as Picasso-based AMD Ryzen parts.
Although the list of supported software is currently limited, applications such as Cyberlink PowerDirector and PhotoDirector are significantly faster on Ice Lake for AI-based features such as automatically applying a visual style to a video or de-blurring an out-of-focus image.
Finally, while the company is waiting until its official Computex keynote to provide more detailed specifications, Intel did tease a new flagship desktop processor, the Core i9-9900KS. Sharing the same architecture as last year’s 14nm Coffee Lake-based i9-9900K, the upcoming 9900KS is an 8-core, 16-thread part with a base clock of 4.0GHz and an out-of-the-box all-core boost frequency of 5.0GHz.
Intel declined to provide additional details on TDP, price, or availability, but did specify that, unlike last year’s i7-8086K, the i9-9900KS will not be a limited release SKU and will join the company’s 9th Gen CPU lineup as a regularly available part.
More details on these and other Intel announcements are expected during the company’s Computex keynote on Tuesday.
*Ryan Shrout is founder and former editor-in-chief of PC Perspective. He no longer has any editorial or financial relationship with the site.
AMD System: RAM frequency DDR4 2400
Intel System: RAM Frequency DDR4 3733
I see Ryan is already honoring his salary at Intel.
Most PC users don’t need any more performance since they essentially only run web browsers and such. At this point, people often only build or buy a desktop pc for specific uses. For general use, like web browsing, they just use a laptop or their phone. Those buying desktops are probably going to be gamers or people who want it for some specific application, like video encoding. You can run web browsing benchmarks, but they are almost completely irrelevant for desktop PCs. I am all for benchmarking more real applications, but general desktop applications are not what is important. It is actually better to bench the “niche” applications that people are buying PCs to run.
In my experience, the actual usability of a system for general use applications depends more on whether there is hardware support for the video codecs in use than just about anything else. I have tried playing YouTube on my older MacBook Pro and it struggles with high resolution / high frame rate VP9 video since it doesn’t have hardware support for it. If I switch to mp4 version, the fans spin down and the battery drain is minimal due to hardware decoding. A MacBook Pro from 2009 is still perfectly usable for general desktop applications except for the VP9 video. I added an SSD and I run a script blocker normally though, so your mileage may vary.
With the 3D performance, Intel should be improving rapidly, but they aren’t going to be there yet. They may be able to get close to the current AMD 12nm+ IGP chips, but those will make their way into laptops right before AMD comes out with a 7 nm or 7 nm+ IGP with updated CPU and GPU cores. Those will probably leave intel far behind again. Cadences often don’t match, so comparing at the time they are out of sync or an unreleased product against a current product is very misleading.
I wish they tested with a real world corporate business load out. My work machine is chugging away all the time on Cisco AMP security mitigation, Trend Micro AV, Symantec Management/Altiris, Dell encryption sync, some rapid 7 data collector, was lucky to get Aegis data collection client recently gone. Throw in Skype for Biz along with Teams and Onedrive trying to sync. My fan is blasting and CPU at 90 to 100% sitting at a blank desktop some days for 10 minutes after boot depending on whats getting updated.
What is Intel’s point exactly? That very few people can actually make use of i7s, so they should just buy i3s instead?
Intel lost the manufacturing lead and can not cost efficiently compete with AMD on number of cores. So they need to downplay that by trying to make the single core performance more important as they still have a small edge there.
But saying that single core is more important is not going to work so they instead try to say the the popularity of the program should impact the benchmark knowing that most popular programs is not multithreaded. they simply do not need to be, word is fast enough as it is.
I like that Cinema4D got mention at the end. No wonder popularity is so low considering the pricing of all major 3DFX/Animation whatever software. Only serious business can afford that, folks who just print money with their work… and CPU cores don’t really matter much when diving into 3D.
Myself I know I use e.g. DAZ, ZBrush, Keyshot all the time, but NEVER touched Adobe Scam Suite or Cinebench which is only for benchmark nerds. So that’s the relevance of that survey in a nutshell.
BTW: It is so weird seeing Ryan doing Intel Podcast instead of PCPer! 😀