Intel Iris Pro 5200 Graphics Review: Core i7-4950HQ Tested
by Anand Lal Shimpi on June 1, 2013 10:01 AM ESTCompute Performance
With Haswell, Intel enables full OpenCL 1.2 support in addition to DirectX 11.1 and OpenGL 4.0. Given the ALU-heavy GPU architecture, I was eager to find out how well Iris Pro did in our compute suite.
As always we'll start with our DirectCompute game example, Civilization V, which uses DirectCompute to decompress textures on the fly. Civ V includes a sub-benchmark that exclusively tests the speed of their texture decompression algorithm by repeatedly decompressing the textures required for one of the game’s leader scenes. While DirectCompute is used in many games, this is one of the only games with a benchmark that can isolate the use of DirectCompute and its resulting performance.
Iris Pro does very well here, tying the GT 640 but losing to the 650M. The latter holds a 16% performance advantage, which I can only assume has to do with memory bandwidth given near identical core/clock configurations between the 650M and GT 640. Crystalwell is clearly doing something though because Intel's HD 4600 is less than 1/3 the performance of Iris Pro 5200 despite having half the execution resources.
Our next benchmark is LuxMark2.0, the official benchmark of SmallLuxGPU 2.0. SmallLuxGPU is an OpenCL accelerated ray tracer that is part of the larger LuxRender suite. Ray tracing has become a stronghold for GPUs in recent years as ray tracing maps well to GPU pipelines, allowing artists to render scenes much more quickly than with CPUs alone.
Moving to OpenCL, we see huge gains from Intel. Kepler wasn't NVIDIA's best compute part, but Iris Pro really puts everything else to shame here. We see near perfect scaling from Haswell GT2 to GT3. Crystalwell doesn't appear to be doing much here, it's all in the additional ALUs.
Our 3rd benchmark set comes from CLBenchmark 1.1. CLBenchmark contains a number of subtests; we’re focusing on the most practical of them, the computer vision test and the fluid simulation test. The former being a useful proxy for computer imaging tasks where systems are required to parse images and identify features (e.g. humans), while fluid simulations are common in professional graphics work and games alike.
Once again, Iris Pro does a great job here, outpacing everything else by roughly 70% in the Fluid Simulation test.
Our final compute benchmark is Sony Vegas Pro 12, an OpenGL and OpenCL video editing and authoring package. Vegas can use GPUs in a few different ways, the primary uses being to accelerate the video effects and compositing process itself, and in the video encoding step. With video encoding being increasingly offloaded to dedicated DSPs these days we’re focusing on the editing and compositing process, rendering to a low CPU overhead format (XDCAM EX). This specific test comes from Sony, and measures how long it takes to render a video.
Iris Pro rounds out our compute comparison with another win. In fact, all of the Intel GPU solutions do a good job here.
Facebook
Twitter
RSS
177 Comments
View All Comments
8steve8 - Saturday, June 1, 2013 - link
Great work intel, and great review anand.As a fan of low power and small form factor high performance pcs, I'm excited about the 4770R.
my question is how do we get a system with 4770R ?
will it be in an NUC, if so, when/info?
will there be mini-itx motherboards with it soldered on?
bill5 - Saturday, June 1, 2013 - link
Anand, would you say the lack of major performance improvement due to crystalwell bodes ill for Xbox one?The idea is ESRAM could make the 1.2 TF Xbox One GPU "punch above it's weight" with more efficiency due to the 32MB of low latency cache (ALU's will stall less waiting on data). However these results dont really show that for Haswell (the compute results that scale perfectly with ALU's for example).
Here note I'm distinguishing between the cache as bandwidth saver, I think we can all agree it will serve that purpose- and as actual performance enhancer. I'm interested in the latter for Xbox One.
Kevin G - Saturday, June 1, 2013 - link
A couple of quotes and comments from the article:"If Crystalwell demand is lower than expected, Intel still has a lot of quad-core GT3 Haswell die that it can sell and vice versa."
Intel is handicapping demand for GT3e parts by not shipping them in socketed form. I'd love to upgrade my i7-2600k system to a 4770K+GT3e+TSX setup. Seriously Intel, ship that part and take my money.
"The Crystalwell enabled graphics driver can choose to keep certain things out of the eDRAM. The frame buffer isn’t stored in eDRAM for example."
WTF?!? The eDRAM would be the ideal place to store various frequently used buffers. Having 128 MB of memory leaves plenty of room for streaming in textures as need be. The only reason not to hold the full frame buffer is if Intel has an aggressive tile based rendering design and only a tile is stored there. I suspect that Intel's driver team will change this in the future.
"An Ultrabook SKU with Crystalwell would make a ton of sense, but given where Ultrabooks are headed (price-wise) I’m not sure Intel could get any takers."
I bet Apple would ship a GT3e based part in the MacBook Air form factor. They'd do something like lower the GPU clocks to prevent it from melting but they want it. It wouldn't surprise me if Apple managed to negotiate a custom part from Intel again.
Ultimatley I'm pleased with GT3e. On the desktop I can see the GPU being used for OpenCL tasks like physics while my Radeon 7970 handles the rest of the graphics load. Or for anything else, I'd like GT3e for the massive L4 cache.
tipoo - Saturday, June 1, 2013 - link
"Ultimatley I'm pleased with GT3e. On the desktop I can see the GPU being used for OpenCL tasks like physics while my Radeon 7970 handles the rest of the graphics load. Or for anything else, I'd like GT3e for the massive L4 cache."I'd love that to work, but what developer would include that functionality for that niche setup?
Kevin G - Saturday, June 1, 2013 - link
OpenCL is supposed to be flexible enough that you can mix execution targets. This also includes the possibility of OpenCL drivers for CPU's in addition to those that use GPU's. At the very least, it'd be nice for a game or application to manually select the OpenCL target in some config file.Egg - Saturday, June 1, 2013 - link
I'm only a noob high school junior, but aren't frame buffers tossed after display? What would be the point of storing a frame buffer? You don't reuse the data in it at all. As far as I know, frame buffer != unpacked textures.Also, aren't most modern fully programmable GPUs not tile based at all?
Also, wasn't it mentioned that K-series parts don't have TSX?
Kevin G - Saturday, June 1, 2013 - link
The z-buffer in particular is written and often read. Deferred rendering also blends multiple buffers together and at 128 MB in size, a deferred render can keep several in that memory. AA algorithms also perform read/writes on the buffer. At some point, I do see Intel moving the various buffers into the 128 MB of eDRAM as drivers mature. In fairness, this change may not be universal to all games and dependent on things like resolution.Then again, it could be a true cache for the GPU. This would mean that the drivers do not explicitly store the frame buffers there but can could be stored there based upon prefetching of data. Intel's caching hierarchy is a bit weird as the CPU's L3 cache can also be used as a L4 cache for the GPU under HD2000/2500/3000/4000 parts. Presumably the eDRAM would be a L5 cache under the Sandy Bridge/Ivy Bridge schema. The eDRAM has been described as a victim cache though for GPU operations it would make sense to prefetch large amounts of data (textures, buffers). It'd be nice to get some clarification on this with Haswell.
PowerVR is still tile based. Previous Intel integrated solutions were also tile base though they dropped that with the HD line (and I can't remember if the GMA line was tile based as well).
And you are correct that the K series don't have TSX, hence why I'd like a 4770K with GT3e and TSX. Also I forgot to throw in VT-d since that too is arbitrarily disabled in the K series.
IntelUser2000 - Sunday, June 2, 2013 - link
Kevin G: Intel dropped the Tile-based rendering in the GMA 3 series generation back in 2006. Although, their Tile rendering was different from PowerVR's.Egg - Sunday, June 2, 2013 - link
Fair points - I was being a bit myopic and only thought about buffers persisting across frames, neglecting the fact that buffers often need to be reused within the process of rendering a single frame! Can you explain how the CPU's L3 cache is an L4 cache for the GPU? Does the GPU have its own L3 cache already?Also I don't know whether PowerVR's architecture is considered fully programmable yet. I know they have OpenCL capabilities, but reading http://www.anandtech.com/show/6112/qualcomms-quadc... I'm getting a vague feeling that it isn't as complete as GCN or Kepler, feature wise.
IntelUser2000 - Tuesday, June 4, 2013 - link
Gen 7, the Ivy Bridge generation, has its own L3 cache. So you have the LLC(which is L3 for the CPU), and its own L3. Haswell is Gen 7.5.