Intel’s Integrated Graphics Mini-Review: Is Rocket Lake Core 11th Gen Competitive?by Dr. Ian Cutress on May 7, 2021 10:20 AM EST
In the last few months we have tested the latest x86 integrated graphics options on the desktop from AMD, with some surprising results about how performant a platform with integrated graphics can be. In this review, we’re doing a similar test but with Intel’s latest Rocket Lake Core 11th Gen processors. These processors feature Intel’s Xe-LP graphics, which were touted as ‘next-generation’ when they launched with Intel’s mobile-focused Tiger Lake platform. However, the version implemented on Rocket Lake has fewer graphics units, slower memory, but a nice healthy power budget to maximize. Lo, Intel set forth for battle.
When a CPU meets GPU
Intel initially started integrating graphics onto its systems in 1999, by pairing the chipset with some form of video output. In 2010, the company moved from chipset graphics to on-board processor graphics, enabling the graphics hardware to take advantage of a much faster bandwidth to main memory as well as a much lower latency. Intel’s consumer processors now feature integrated graphics as the default configuration, with Intel at times dedicating more of the processor design to graphics than to actual cores.
|Intel CPUs: IGP as a % of Die Area|
Die Area %
|Sandy Bridge||i7-2600K||Jan 2011||4||Gen6||GT2||11%|
|Ivy Bridge||i7-3770K||April 2012||4||Gen7||GT2||29%|
|Kaby Lake||i7-7700K||Jan 2017||4||Gen9||GT2||36%|
|Coffee Lake||i7-8700K||Sept 2017||6||Gen9||GT2||30%|
|Coffee Lake||i9-9900K||Oct 2018||8||Gen9||GT2||26%|
|Comet Lake||i9-10900K||April 2020||10||Gen9||24 EUs||22%|
|Rocket Lake||i9-11900K||March 2021||8||Xe-LP||32 EUs||21%|
|Ice Lake-U||i7-1065G7||Aug 2019||4||Gen11||64 EUs||36%|
|Tiger Lake-U||i7-1185G7||Sept 2020||4||Xe-LP||96 EUs||32%|
All the way from Intel’s first integrated graphics to its 2020 product line, Intel was reliant on its ‘Gen’ design. We saw a number of iterations over the years, with updates to the function and processing ratios, with Gen11 featuring heavily in Intel’s first production 10nm processor, Ice Lake.
The latest graphics design however is different. No longer called ‘Gen’, Intel upcycled its design with additional compute, more features, and an extended effort for the design to scale from mobile compute all the way up to supercomputers. This new graphics family, known as Xe, is now the foundation of Intel’s graphics portfolio. It comes in four main flavors:
- Xe-HPC for High Performance Computing in Supercomputers
- Xe-HP for High Performance and Optimized FP64
- Xe-HPG for High Performance Gaming with Ray Tracing
- Xe-LP for Low Power for Integrated and Entry Level
Intel has initially rolled out its LP designs into the market place, first with its Tiger Lake mobile processors, then with its Xe MAX entry level notebook graphics card, and now with Rocket Lake.
Xe-LP, A Quick Refresher
Intel’s LP improves on the previous Gen11 graphics by reorganizing the base structure of the design. Rather than 7 logic units per execution unit, we now have 8, and LP’s front-end can dispatch up two triangles per clock rather than one. The default design of LP involves 96 execution units, split into a centralized ‘slice’ that has all the geometry features and fixed function hardware, and up to 6 ‘sub-slices’ each with 16 logic units and 64 KiB of L1 cache. Each variant of LP can then have up to 96 execution units in a 6x16 configuration.
Execution units now work in pairs, rather than on their own, with a thread scheduler shared between each pair. Even with this change, each individual execution unit has moved to an 8+2 wide design, with the first 8 working on FP/INT and the final two on complex math. Previously we saw something more akin to a 4+4 design, so Intel has rebalanced the math engine while also making in larger per unit. This new 8+2 design actually decreases the potential of some arithmetic directly blocking the FP pipes, improving throughput particularly in graphics and compute workloads.
The full Tiger Lake LP solution has all 96 execution units, with six sub-slices each of 16 execution units (6x16), Rocket Lake is neutered by comparison. Rocket Lake has 4 sub-slices, which would suggest a 64 execution unit design, but actually half of those EUs are disabled per sub-slice, and the final result is a 32 EU implementation (4x8). The two lowest Rocket Lake processors have only a 3x8 design. By having only half of each sub-slide active, this should in theory give more cache per thread during operation, and provides less cache pressure. Intel has enabled this flexibility presumably to provide a lift in edge-case graphics workloads for the parts that have fractional sub-slices enabled.
Xe-LP also comes with a revamped media engine. Along with a 12-bit end-to-end video pipeline enabling HDR, there is also HEVC coding support and AV1 decode, the latter of which is a royalty-free codec providing reported similar or better quality than HEVC. Intel is the first desktop IGP solution to provide AV1 accelerated decode support.
Rocket Lake Comparisons
For this review, we are using the Core i9-11900K, Core i7-11700K, and Core i5-11600K. These three are the highest power processors in Intel’s Rocket Lake lineup, and as a result they support the highest configuration of LP graphics that Intel provides on Rocket Lake. All three processors have a 4x8 configuration, and a turbo frequency up to 1300 MHz.
|Intel Integrated Graphics|
|Cores||8 / 16||8 / 16||6 / 12||10 / 20|
|Base Freq||3500 MHz||3600 MHz||3900 MHz||3700 MHz|
|1T Turbo||5300 MHz||5000 MHz||4900 MHz||5300 MHz|
|GPU uArch||Xe-LP||Xe-LP||Xe-LP||Gen 11|
|GPU EUs||32 EUs||32 EUs||32 EUs||24 EUs|
|GPU Base||350 MHz||350 MHz||350 MHz||350 MHz|
|GPU Turbo||1300 MHz||1300 MHz||1300 MHz||1200 MHz|
Our comparison points are going to be Intel’s previous generation Gen11 graphics, as tested on the Core i9-10900K which has a 24 Execution Unit design, AMD’s latest desktop processors, a number of Intel’s mobile processors, and a discrete graphics option with the GT1030.
In all situations, we will be testing with JEDEC memory. Graphics loves memory bandwidth, and CPU memory controllers are slow by comparison to mobile processors or discrete cards; while a GPU might love 300 GB/s from some GDDR memory, a CPU with two channels of DDR4-3200 will only have 51.2 GB/s. Also, that memory bank needs to be shared between CPU and GPU, making it all the more complex. The use case for most of these processors on integrated graphics will often be in prebuilt systems designed to a price. That being said, if the price of Ethereum keeps increasing, integrated graphics might be the only thing we have left.
The goal for our testing comes in two flavors: Best Case and Best Experience. This means for most benchmarks we will be testing at 720p Low and 1080p Max, as this is the area in which integrated graphics is used. If a design can’t perform at 720p Low, then it won’t be going anywhere soon, however if we can achieve good results at 1080p Max in certain games, then integrated graphics lends itself as a competitive option against the basic discrete graphics solutions.
If you would like to see the full CPU review of these Rocket Lake processors, please read our review:
Intel Rocket Lake (14nm) Review: Core i9-11900K, Core i7-11700K, and Core i5-11600K
Pages In This Review
- Analysis and Competition
- Integrated Graphics Gaming
- Conclusions and Final Words
Post Your CommentPlease log in or sign up to comment.
View All Comments
GeoffreyA - Saturday, May 8, 2021 - linkOn Twitter, he posts a "Happy New Year" and retweets some Apple things, but other than that, nothing much. I do recall one message saying, he was happy and couldn't complain.
Hulk - Friday, May 7, 2021 - linkBeyond hardware video decode I was referring to the OpenCL accelerated video filters Vegas Pro uses to make previewing smooth. Rendering isn't a big deal as I can frameserve a project to Handbrake and transcode overnight. But smooth previews during the editing process are essential for creative on-the-fly work.
brucethemoose - Saturday, May 8, 2021 - linkAhhh, yeah, that's very different. Some sites like Pudget bench the GPU filters, but I'm not sure how well that translates to preview performance, and they don't do IGPs.
mode_13h - Friday, May 7, 2021 - linkIt'd have been nice if you'd bothered to look up the EU-counts for the earlier CPUs. Here they are:
Sandy Bridge i7-2600K Jan 2011 4 Gen6 12 EU 11%
Ivy Bridge i7-3770K April 2012 4 Gen7 16 EU 29%
Haswell i7-4770K June 2013 4 Gen7.5 20 EU 29%
Broadwell i7-5775C June 2015 4 Gen8 48 EU 48%
Skylake i7-6700K Aug 2015 4 Gen9 24 EU 36%
Kaby Lake i7-7700K Jan 2017 4 Gen9 24 EU 36%
Coffee Lake i7-8700K Sept 2017 6 Gen9 24 EU 30%
Coffee Lake i9-9900K Oct 2018 8 Gen9 24 EU 26%
mode_13h - Friday, May 7, 2021 - linkI knew the formatting wouldn't look great, but that's a mess. It kept only one space between words!
Here's a simplified table, with the full list:
i7-2600K 12 EUs
i7-3770K 16 EUs
i7-4770K 20 EUs
i7-5775C 48 EUs
i7-6700K 24 EUs
i7-7700K 24 EUs
i7-8700K 24 EUs
i9-9900K 24 EUs
i9-10900K 24 EUs
i9-11900K 32 EUs
i7-1065G7 64 EUs
i7-1185G7 96 EUs
mode_13h - Friday, May 7, 2021 - linkNote that Broadwell's GT2's mostly had 24 EUs, but that's one of the models with GT3e graphics.
mode_13h - Friday, May 7, 2021 - linkOf course, the EU-count doesn't tell the full story. One of the more notable changes was in Gen 7, when they went from 1x 128-bit SIMD to 2x.
WaltC - Friday, May 7, 2021 - linkI'm sort of perplexed--I didn't see that the title question was answered in the article...unless Dr. Potato Head is asking whether Intel's current IGPs are "competitive" with older Intel IGPs...which would seem to be the case. I mean, we should hope that Intel's latest should best its previous efforts. But is that really being "competitive"...? Intel seems very confused of late as to who its chief competitor is, imo--at least publicly...;)
Oxford Guy - Friday, May 7, 2021 - linkI’m confused by your reference to a child’s game.
29a - Friday, May 7, 2021 - linkI'm guessing it's because the Dr has a YouTube channel called Tech Potato or something close to that and he calls low end computers potatoes.