Slashdot Mirror
Haswell Integrated Graphics Promise 2-3X Performance Boost
crookedvulture writes "Intel has revealed fresh details about the integrated graphics in upcoming Haswell processors. The fastest variants of the built-in GPU will be known as Iris and Iris Pro graphics, with the latter boasting embedded DRAM. Unlike Ivy Bridge, which reserves its fastest GPU implementations for mobile parts, the Haswell family will include R-series desktop chips with the full-fat GPU. These processors are likely bound for all-in-one systems, and they'll purportedly offer close to three times the graphics performance of their predecessors. Intel says notebook users can look forward to a smaller 2X boost, while 15-17W ultrabook CPUs benefit from an increase closer to 1.5X. Haswell's integrated graphics has other perks aside from better performance, including faster Quick Sync video transcoding, MJPEG acceleration, and support for 4K resolutions. The new IGP will support DirectX 11.1, OpenGL 4.0, and OpenCL 1.2, as well." Note: Same story, different words, at Extreme Tech and Hot Hardware.
is that Intel provides very nice open source drivers for their integrated GPUs
the Hot Hardware link confirms DisplayPort 1.2, which is the only thing I /really/ care about. The others are nice, but 4K out of the laptop means my next mid-range laptop can be my primary desk machine as well. This should push along the display manufacturers after their decade of stalling (perhaps soon we'll see screens in the 20-24" range with more resolution than our 5" displays have).
yep
i love buying games the day of release with all the DRM, bugs, always connected to internet issues where they can't support all the players, etc
i'd rather buy a game a year or two after release after its on sale at least 50% off
Think of these chips with integrated graphics like hybrid cars. You're not gonna go down to the drag strip with them, or haul a camper, or pick up the 10 kid carpool group. But for the vast majority of trips you'll get to the same destination is basically the same amount of time, with less noise and higher efficiency.
New Intel GPUs are surprisingly competent. No, they don't stand up to higher end discrete solutions but you can game on them no problem. You have to turn down the details and rez a bit in some of the more intense ones, but you can play pretty much any game on it. (http://www.notebookcheck.net/Intel-HD-Graphics-4000-Benchmarked.73567.0.html). For desktops I always recommend dropping $100ish to get a reasonable dedicated card but for laptops, gaming on an integrated chip is realistic if your expectations are likewise realistic.
Gone are the days of the GMA 950 that sucked at even simple GDI+ rendering. The integrated GPUs now are competent, though lower end (in keeping with their power profile).
Wouldn't these kinds of things be more accurately described as GPUs with integrated CPUs?
It's been 10 years since Intel started panicking when they realized a Pentium core could be tucked into a tiny corner of a GPU, as far as transistor count went.
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
So unexpected that you can't even name one!
Intel GPUs are fine for 99% of use. Heck, most games run fine on them. Sure the latest Call of Honor: Medal of Duty will not run on UItra, but most games will be fine on medium and low settings.
Performance improves in ways you might not even expect.
I guess it improves it in such an unexpected way I didn't even notice. In the two desktop computers I have in my household, I originally built them using integrated video cards and then upgraded to discrete cards couple months later when I had some more time to pick out something and some spare cash. Neither my wife nor I noticed any difference in normal desktop performance for office related software or web browsing. The only difference seen was in video games. My work issued laptop has integrated GPU, and I'm not sure what performance could be improved on it other than maybe a faster hard drive (there is only so fast I can type, click through documents, and my code runs on a cluster instead of locally...).
Or to put it in automotive terms: even a family car needs the ability to get onto the highway without being a threat to public safety.
Yes, while this is true, it really isn't relevant. You look foolish if you try to claim hybrids are not usable because they can't drive at highway speeds as they are already capable of that. Likewise, integrated GPUs are quite fast enough for a the vast majority of basic computer use and only a hindrance to specific uses that many people may or may not need (i.e. not every one does high end gaming, video editing, etc.).
Except your analogy is total nonsense. Modern operating systems do in fact benefit from having a decent discrete GPU. Performance improves in ways you might not even expect.
Nope. They benefit from having 3D acceleration, yes, but the integrated graphics on modern Intel chips is a discrete core. It just happens to be on the same die as the CPU. My laptop's CPU is clocked at 1.2GHz dual core, with two extra cores for the video clocked at 300-500MHz. Those cores are dedicated to the video only, and it's *plenty* fast enough for normal use on the operating system, with all of the blingy effects. Switching to a discrete graphics card won't make any difference at all, because the video cores are physically separated from the CPU cores, and have a completely different execution pipeline so wouldn't be able to run OS calls anyway.
Intel's integrated video is enough to run most games these days. I game on that laptop and while the graphics aren't as fast as my desktop's 6970, they're plenty adequate for gaming on the go... and I'm not just talking about ancient games here, either: it's good enough for Civ5 in WINE, and also for stuff like Torchlight II, which it'll run at max on the laptop's 1366x768 screen. You won't be playing the latest Call of Duty at maximum settings on a dual 1920x1080 display with Intel's integrated graphics, but then again, the people who want to do that won't be bothering with integrated graphics to begin with, will they? A significant portion of Steam's userbase are using Intel integrated graphics these days....
http://www.engadget.com/2013/04/12/seiki-50-inch-4k-1300/
$1300 for a 4k display. Granted, it's locked to 30Hz, but for most of us 60Hz will be as fast as we need to go (though we'll get more for that god-awful 3D crap they keep trying to push). 4k @ 50 is very close the 2560x1600 30" monitor I have for pixel size, which is fine enough for me at my working distance.
We stalled at 1920x1080 because every moved to TV production. Now that 4k/8k has broken free, we can get over that hump. Not saying there aren't hurdles, but the consumer-limit has been breached and I expect the next to years to result in a shift. Note: there is no material broadcast at 60 frames at 1080, but people are all bonkers over 240Hz displays anyway. They'll be the same ones who wanted 1080p devices to watch their (upscaled) DVDs.
Is it just my observation, or are there way too many stupid people in the world?
Haswell parts are expected to be 10-15% faster than Ivy Bridge, which was itself barely any faster than Sandy Bridge.
Anyone remember the days when computing performance doubled or even tripled between generations?
I have a desktop PC running a Sandy Bridge i5-2500K running at a consistent 4.5GHz (on air). At this rate, it could be another couple generations before Intel has anything worthwhile as an upgrade... I suspect that discrete-GPU-buying home PC enthusiasts are going to continue to be completely ignored going forward while Intel continues to focus on chips for tablets and ultrabooks.
IIRC transcoding / decoding with Intel's quicksync is actually VERY competitive with a discrete GPU. And all of those CUDA / OpenCL tasks are hardly representative of the average user.
As parent said, for the 99% use case, Intel integrated are sufficient.
Of course, when it comes down to raw performance comparison for high performance applications integrated solutions won't be able to keep up with current discrete solutions.
But when you take other criteria into account like the volume that is used by discrete hardware in a casing, the energy consumption and subsequently the need of a cooling systems, then integrated solutions become a lot more attractive.
thank you CUDA and OpenCL
OpenCL-heavy tasks can be done on a compute server at home or in a data center, and you can SSH (or VNC or RDP or whatever) to use an application on a compute server from your laptop. The only real use case I see for carrying an OpenCL powerhouse with you, apart from running shaders in a high-detail 3D game, is for editing huge images or high-definition video in a vehicle or some other place with no Wi-Fi. One workaround is to downscale the video to low definition (e.g. 320x180), edit the low-definition video while away from the net, and then export the edit decision list (EDL) back to the compute server to render the result in high definition. I used to do that with AviSynth.
Running games at resolutions and detail levels that look better than doom
Games are the other reason for carrying a beefy GPU with you. But Skyrim looks better than Doom, Doom II, and Doom 3, and Skyrim runs playably on the HD 4000 at 720p medium.
actually, video editing doesn't really benefit from a discrete GPU since the damn encoding support is still crap. Most of the various software I've looked at still get more bang from a better CPU then GPU encoding and if you're in the industry like ILM/Pixar, then you aint using GPU encoding anyhow - its mainly dedicated ASICS and such. For someone doing it as a hobby, they're buying a video card specifically supported by their software so it makes no god damn difference to 99.99 percent of the folks out there that onboard graphics suck.
In my case, small business owner; I've been planning a new build for 4th quarter (part of my 4yr replacement cycle) based on a Xeon E3 1275 with onboard graphics because the system purpose doesn't need much in the way of a GPU. It's a development system (builds and Database work) so why waste money. Hell all of my employees systems are onboard graphics just to save a few bucks that's better spent on more ram or a slightly better cpu. As with anyone, trade-offs are required when building/specing our systems and as a business, we tend to go with the cheapest configurations we can get. Keep in mind that the cheapest configuration does not mean the cheapest parts. We learned a long time ago that spending a bit more for quality hardware resulted in less downtime.
Mod me up/Mod me down: I wont frown as I've no crown
And all of those CUDA / OpenCL tasks are hardly representative of the average user.
There's a meme lately in Slashdot comment sections that everything must be made for "the average user" without any room to grow. I see it popping up whenever anybody mentions limits of a product sold to the public, especially artificial market-segmenting limits. Where did it come from?
Wow, that will bring it up to almost the same speed as the 2 year old AMD APUs if I'm not mistaken, lol. Talk about being behind! I was amazed when the first couple P-series graphics adapters came out built into the first Sandy Bridge chips. You could actually play Skyrim on an i3-2100 at low settings and it killed at HD video playback. Now for $110, my demo unit at my shop is an A10 APU with a 6.9 graphics rating. It can play Starcraft II at almost maxed settings at 60FPS at 1280x1024 and the CPU is just a hair slower than an i5-2400 for a crap-ton less money. At least $60 if I remember correctly. The 1866 native memory controller helps too since even Ivy Bridge only hits 1600. So then it's better at video encoding and gaming than any i5. Intel is really playing catch-up at this point. I don't know why everyone's all doom and gloom over AMD getting crushed by Intel. I would have thought that was over by now, especially after releasing Trinity, Zambezi, and Vishera. Those 3 really are better than Intel in almost every way.
I do HPC engineering for a living, and I really don't see the point in private discrete GPUs anymore. We've added 8000 Teslas to our cluster, and I've come to prefer using them over CPUs simply for performance reasons. But likewise I've come to prefer IGPs to discrete cards for private use in the last 2-3 years. There is no game that isn't playable on an Ivy Bridge IGP (last I've run is Skyrim on 1920p and settings in the middle between average and max, 40-50 FPS), the power usage is lower (in general but also when watching movies), complexity is tremendously lower (which translates into more stable drivers), it's cheaper and fwiw lighter.
Sure, if you're a CAD guy or 3D artist, you'll be able to to bigger stuff faster, but apart from those highly specific needs by a tiny part of the population (who can and should buy "specialised hardware", that means discrete GPUs), there is little advantage over IGPs.
You mention transcoding media, hardly a common task, but what's the problem with one core chugging away at it in the background while you use your other system resources? A GPU may do it quicker, but either you can't use it to it's full capacity for transcoding while at the same time playing that AAA title, or you're not needing the resources at the moment anyway, so "quicker" becomes nothing but a luxury.
VIdeo decompression is a moot point, as 500 MHz ARM chips can do full HD via DSP decoders, and you've got plenty of the latter on modern x86 chips.
The only somewhat "valid" point is e-peen: you'll get higher benchmark scores and feel better for having high-end hardware, which is up to everyone personally. I like the feeling of my tiny little "thingie" reaching the same goals more efficiently, with lower power usage, thermal emission and complexity.
captcha: tearful
Haven't looked at the latest cards, have you? One of the reasons that AMD went with GCN over their previous VLIW design is that GCN allows for core parking, with core parking when you are doing non GPU intensive jobs the GPU can put to sleep all the cores you no longer need thus bringing the wattage and heat WAY down, down almost to IGP levels without having to be stuck with only IGP levels of performance.
And as far as noise goes most of the cards have a passively cooled variant, can't get more quiet than that and again without giving up the huge performance advantage the discrete cards give you.
ACs don't waste your time replying, your posts are never seen by me.
Seiki has a 50" TV with a 3840x2160 resolution, available right now for $1499. So I don't buy the argument that it's somehow technologically prohibitive. Why can this crappy company no one has ever heard of bring out a 4K TV under $1500, but no one else can make a 4K monitor in an even smaller size (32" or so) without charging over $5500? (and that's for Sharp's offering, which is the next least expensive – most 4K monitors cost as much as a new car). As far as I can tell, it's not technological barriers but a desire to segment the market and charge professional and medical users out the ass for 4K displays.
Best quality encoding is always done in software because it requires multiple passes, however, when you're talking about actual editing, effects and such (even a simple blur) benefit from GPU support by an order of magnitude. Image processing is massively parallel. Ever use Apple's motion, for example? It's GPU support allows you to do in hardware in real-time what used to take hours in After Effects, where the GPU support sucks.
When you're doing any sort of composting or image processing, you need a GPU if you want any sort of WYSIWYG capability. Oddly enough, though, when it comes to production quality 3d rendering, many of the best renderers, such as Pixar's Renderman, are entirely software (some others do use GPU assist, but almost none are pure GPU). This will likely change in the future as brute force path tracing and MLT becomes more reasonable to do on GPU, but for the moment, software is still king.