WE GOT INTEL'S PROTOTYPE GRAPHICS CARD!!

Linus Tech Tips ·Linus Tech Tips ·2019-05-06 · 2,152 words · ~10 min read
Floatplane YouTube

Transcript

JSON SRT VTT 262
0:00 Intel keeps very careful track of its internal engineering samples going to great lengths to ensure that if
0:08 They leave the lab it is in pieces so small that they could never be reassembled again
0:15 So the first question we need to answer is
0:19 How did we get our hands on?
0:22 this thing
0:25 eBay obviously what can't you buy on eBay a
0:28 Sponsorship online is tech tips for that you have to talk to Colton like Corsair did Corsair's dark core se RGB wireless mouse
0:36 Features one millisecond 2.4 gigahertz and low latency Bluetooth connectivity check it out at the link below
0:43 So the seller turned out to have been a contractor at Intel a few years ago who in the middle of a sixth floor
1:00 renovation went dumpster diving through the boxes full of
1:04 Junk that was destined for the e-waste pile
1:07 apparently
1:07 It was a
1:07 treasure trove of press samples laptops and
1:12 this
1:13 GPU looking thing that was noteworthy for being blue instead of green or red
1:20 You see by that point
1:22 project Larrabee this
1:25 Sort of had been cancelled for years and how many years depends on which cancellation you're going by
1:32 So talking to Tom Forsythe who was one of the key team members he figures
1:37 They got cancelled anywhere from four to five times and remembers getting these weird memos
1:44 Yeah
1:46 You guys are gonna see some headlines
1:48 It's just a thing
1:50 None of you are laid off. Just keep on working
1:54 BT dubs you've been rebranded Zeon fight. Thanks. Bye. So
1:59 What is this thing?
2:01 that's actually a somewhat complicated question, but
2:05 Because it's got a DVI port not to mention DisplayPort and HDMI soldered onto it
2:12 It is technically an engineering sample board for Intel's first and to date only
2:20 dedicated graphics card now
2:23 Most people who follow the mainstream tech press believe that project Larrabee was an abject failure
2:30 But as is so often the case the truth is actually
2:34 Stranger than fiction not only was it a success
2:39 But it powered th2, which was the world's most powerful
2:45 supercomputer for over two years and ten years later
2:48 You can actually still buy its descendants either in socketed form as we reviewed just last year or on
2:57 Amazon for a cool
2:59 1500 greenbacks
3:01 So as it turns out the goal of the project
3:05 Never was actually to create a gaming GPU
3:10 That was just a workload that was already fairly well understood at the time because you gotta remember back in the mid
3:17 2000s the idea of using a GPU as a
3:21 General purpose computing unit was just emerging
3:25 so this this idea of using it for gaming was actually just a small part of a business case to build a
3:34 processor that had many highly efficient
3:38 x86 cores that could be easily just like
3:42 slotted into these powerful supercomputers
3:45 but
3:47 That doesn't mean that it couldn't have been used for gaming in fact by the time they wound down the units that were working on
3:55 graphics they had about
3:57 300 of the top-selling games on Steam
4:01 running on the thing with a card just
4:04 like this one as the only GPU in the system and the way this whole thing worked is incredible now a
4:12 Normal graphics card or GPU rather uses a lot of fixed function hardware
4:18 So if you told it okay, look, I don't need shaders
4:23 Just draw a ton of tiny lines with really nice anti-aliasing
4:28 So that's pretty much CAD in a nutshell
4:31 It would use only a fraction of its hardware
4:34 But with Larrabee
4:36 Everything is software. So the whole chip is lit up doing that
4:41 So that actually helped to offset the x86 overhead a fair bit
4:46 This was the fastest CAD card at the time and it had other benefits with regular GPUs
4:53 You might run into a situation where?
4:55 Enabling a particular feature in a game might hit the AMD users a lot harder than the NVIDIA users or vice versa
5:04 So during development
5:06 AMD and NVIDIA they both have to actually guess as best
5:10 They can what the next couple of years of games will demand and then try to look into their crystal ball and build their hardware
5:18 Around that
5:19 Larrabee, no such limitation. This thing is a full-blown
5:25 Computer with up to 61 quad threaded cores running a normal operating system like FreeBSD
5:32 like you could
5:33 you could actually telnet into the thing
5:36 and run a top command and see a list
5:39 of all the processes that were running on it.
5:41 And if you were running a game,
5:43 you'd see, I don't know, 128 or 200 processes
5:47 called DirectX graphics.
5:50 And you could do that while the thing was working.
5:55 So if you wanted, you could cordon off some of the cores
5:59 and use them for something else,
6:01 or you could just YOLO it
6:03 and throw another workload into the mix
6:06 and then just let the processor manage itself.
6:10 The only non-programmable hardware on this puppy
6:13 is the texture unit, which takes very simple commands.
6:18 I mean, wrap your brain around this.
6:21 The thing that I'm looking at right here
6:23 is Intel's first ever DirectX 11 GPU.
6:29 Even though it was built,
6:31 it was built before DirectX 11.
6:34 So this was possible because all of those graphics card
6:38 features that are normally running in hardware
6:42 are just running in software.
6:44 So you could actually update it to DirectX 11
6:47 or DirectX 12 with a driver update.
6:52 Now, there are some caveats here.
6:54 I mean, there's a reason that the thing
6:56 never made it into a computer near you.
6:59 It wasn't as efficient as a dedicated graphics card,
7:01 for a lot of things.
7:05 So it only got about a quarter of the performance in games
7:08 as a comparably power consuming card
7:14 from AMD or NVIDIA at the time.
7:16 But it was really good at certain graphics workloads
7:20 for a number of reasons.
7:22 And I mean, if you think about it
7:25 and you look at how far off they were,
7:27 considering that they were effectively emulating,
7:32 dedicated hardware, it's damn impressive.
7:37 So what happened?
7:41 Well, management happened.
7:44 Intel at its core, haha, is a hardware company.
7:48 So they wanted all the features completed
7:51 so they could either ship this thing or can it.
7:55 Because in the hardware world,
7:57 making up a four times difference in performance
8:00 is impossible and you might as well just pull the plug.
8:03 But the team wanted to work on performance optimization
8:07 instead because in the software world,
8:10 it's not unheard of to go from like two pixels
8:13 showing up on a screen and dog slow
8:16 to a hundred times faster in a week
8:19 if you have a breakthrough.
8:21 And it got to the point where they had to have
8:23 separate teams for performance and for features
8:28 to get management off their backs.
8:30 So the performance team actually got Quake
8:33 running like really fast,
8:35 but then they found out that Quake was this weird edge case
8:38 and the architecture would have to be completely redone.
8:41 I mean, to give you guys some idea of the dysfunction,
8:43 at one point there were three to four software teams
8:47 with different ideas and working on
8:49 different rendering architectures.
8:51 But depending who you ask,
8:52 the continued development would have been worth it.
8:55 I mean, imagine this.
8:56 Instead of turning anti-aliasing on for an entire scene,
9:01 imagine if a game developer
9:03 could say, well, you know what?
9:04 This sky is not important to be anti-aliased.
9:08 Why don't we just focus all of our AA
9:11 on these characters here or this foliage there?
9:15 Or how about this?
9:16 Like, oh crap, that texture wasn't loaded.
9:20 You know what?
9:21 Let's just procedurally generate a placeholder.
9:24 Boom.
9:26 Arguably the stupidest decision that was made
9:29 was to make the Larrabee graphics team
9:32 and the Gen graphics team,
9:35 which is what Intel calls
9:36 its integrated graphics internally,
9:38 compete together for the same budget
9:42 and then like make internal presentations
9:45 arguing about why their approach was good in the future
9:48 and the other groups was bad and not the future
9:50 because they were both perfectly suitable
9:53 for what they were doing.
9:55 Larrabee was never going to be a five watt part
9:59 that you could bake right into a CPU.
10:02 A 200 watt PCI express part
10:05 was nowhere on the roadmap for Gen.
10:09 So what I've got here, come on, come on, come on,
10:12 is not Knight's Ferry.
10:15 That was the first Larrabee revision
10:17 that had some deal breaking bugs.
10:20 Apparently the saying in the hardware industry is
10:22 always plan to make a prototype
10:25 since you'll end up making one anyway.
10:27 So this is Knight's Corner
10:30 and probably has anywhere from six to 16
10:33 16 gigs of RAM and up to 62 cores,
10:36 depending on how many of them had some manufacturing flaws.
10:40 Should we fire it up?
10:41 I mean, come on.
10:43 I wasn't not gonna do that at this point.
10:46 I spent like $400 on this thing off of eBay.
10:51 I've got no drivers for it.
10:53 So it's actually, this is the first time I've turned it on.
10:56 So it is very possible that it won't manage
10:58 to display anything even in 2D,
11:01 but I definitely, definitely, definitely, definitely,
11:03 definitely have to try.
11:07 By the way, if anyone out there
11:09 has the secret sauce drivers
11:12 or has access to the secret sauce drivers
11:14 that would make this run games,
11:17 please hit me up.
11:21 I mean, assuming that it even works,
11:23 which we don't know yet.
11:26 I actually haven't tried this.
11:28 I wanted to save the suspense for the video.
11:31 This is like far more post codes
11:33 than I'm accustomed to seeing,
11:36 but it hasn't stopped.
11:38 And it hasn't rebooted.
11:39 We've got some kind of LED on here.
11:47 It looks like it stalled on D6,
11:49 but I don't know what that is.
11:51 Now, when I talked to Tom,
11:52 he did specifically mention it's got DVI soldered to it.
11:56 Now, I don't know if that's because DVI
11:58 was the most relevant output at the time.
12:01 So that's like what they used internally,
12:03 or if the DisplayPort and HDMI were just dummies
12:06 and DVI was the only thing that actually worked.
12:08 So, take two, I'm gonna run and grab a DVI monitor
12:13 and gonna try this again.
12:14 Like, I kind of wonder about,
12:17 you know, what it's PCI-e.
12:19 I mean, would that be even Gen 2 at that point?
12:25 Like 2007, 2009.
12:33 I wonder about compatibility
12:34 with the new board and stuff like that.
12:38 You know what?
12:39 I don't think it's gonna boot.
12:41 Well, that's pretty disappointing.
12:44 I thought I might be onto something
12:45 with the whole DVI thing.
12:46 I'm just gonna try, I'm gonna try one other slot
12:51 just to, I think there's only one other one
12:56 out in the wild and some like Russian collector
12:59 of like weird hardware has it.
13:00 Yeah, not you, a different one.
13:04 Okay, sometimes it hangs on 79 for a bit
13:07 and then this thing boots.
13:09 So that might've been a good sign.
13:12 No, that's D6 again.
13:13 I think it's not going anywhere.
13:16 Well, that was disappointing,
13:18 but I'm gonna let it keep trying
13:21 while I tell you guys about Mastrop.
13:26 I'm like sad, it's like hard to have any energy.
13:28 Okay, I'm gonna try that again.
13:31 Mastrop, Mastrop is featuring
13:33 the Sennheiser PC37X gaming headset.
13:37 They've got angled drivers and an open back design
13:39 and the drivers actually come from the same family
13:41 as the HD598 and HD600 headphones.
13:45 They offer superior stereo imaging and locational accuracy
13:49 and come with a noise canceling microphone.
13:50 They're available on Mastrop at the link below
13:53 for a limited time for just 120 US dollars.
13:57 So go check them out.
13:59 So thanks for watching guys.
14:00 If this video sucked, you know what to do,
14:03 but if it was awesome, get subscribed, hit that like button.
14:07 You can especially hit that like button
14:08 if you wanna make me feel better
14:10 about how disappointed I feel right now,
14:12 or you can check out the link to where to buy the stuff
14:14 we featured in the video description.
14:17 Also linked in the description is our merch store,
14:19 which has cool shirts like this one
14:20 and our community forum, which you should totally join.
14:26 Oh, I really, I was really hoping
14:29 I was just gonna get the screen to light up.
14:31 That was all I was really, that was all I really wanted.
14:39 Good night, sweet prince.
14:42 You were too good for this world.