text stringlengths 14 502 |
|---|
[260.84 --> 262.34] Up to you. |
[264.66 --> 266.08] Uh, uh, uh, uh, sorry. |
[266.12 --> 268.00] I wasn't listening because I was checking my notes. |
[268.10 --> 275.66] I have some notes from Dr. Cutris, my favorite writer over at Anant Tech, about this topic. |
[275.82 --> 276.98] So, I'm just going to jump into it. |
[277.14 --> 278.80] I didn't hear what you said, and I'm sorry. |
[278.94 --> 279.40] It's cool. |
[279.46 --> 279.68] Go for it. |
[279.68 --> 282.98] This was posted by NumLock21 on the phone. |
[285.64 --> 286.34] Thanks, NumLock. |
[286.38 --> 287.08] What are you going to do, quit? |
[289.04 --> 290.86] Um, so the original... |
[290.86 --> 291.10] Done. |
[292.18 --> 294.78] The original article here is from Tech Power Up. |
[295.42 --> 300.38] Intel unveils discrete GPU prototype development. |
[300.72 --> 300.98] Yeah. |
[300.98 --> 307.76] So, it wasn't that long ago that AMD made a high profile hire. |
[307.76 --> 311.82] Uh, Raja Kuduri, AMD's former head of Radeon Technologies Group. |
[312.50 --> 314.76] Um, but clearly... |
[314.76 --> 317.12] So, there's a bunch of stuff here. |
[317.20 --> 318.76] So, there's like this. |
[319.84 --> 322.64] There's got like a block diagram, which is really interesting. |
[323.64 --> 328.30] There's what looks like sort of a mock-up of what a die shot might look like. |
[328.34 --> 332.70] Like, there's a bunch of what looks like very internal information. |
[332.70 --> 342.14] Um, and a lot of people are extrapolating a bunch of information about this supposed 14 nanometer dedicated GPU. |
[342.14 --> 347.14] And, um, so Dr. Kutris actually proactively because... |
[348.70 --> 363.30] Ian, do you really not, you know, believe in our fact-checking to the point where we would, you know, dig through all these diagrams and figure out the misinterpretation that other sites, you know, might have accidentally made here? |
[363.42 --> 365.56] Do you really not believe that we're capable of that? |
[365.74 --> 366.40] I appreciate it. |
[366.40 --> 370.94] Because, you know, thank you for covering our butts because we were not capable of that. |
[371.08 --> 372.00] Thank you, Dr. Kutris. |
[372.24 --> 372.32] Yeah. |
[372.68 --> 377.36] So, he proactively reached out because he was sure that we were going to talk about this. |
[377.82 --> 378.76] And I'm just going to read it. |
[378.78 --> 380.80] I'm just going to read it because he put it better than I could have. |
[381.64 --> 383.68] It's almost like he's a professional writer or something. |
[384.20 --> 384.72] I don't know. |
[384.86 --> 385.20] Maybe. |
[385.68 --> 385.82] Weird. |
[385.82 --> 385.86] Weird. |
[386.32 --> 392.94] Just wanted to poke about the Intel discrete GPU hullabaloo that's been going around this week that might end up in WAN show. |
[392.94 --> 399.84] So, everywhere is quoting it as a discrete GPU prototype, but that is because of a slight misreading of the Japanese article source. |
[400.00 --> 400.32] Okay. |
[400.50 --> 405.38] Leave it to Dr. Kutris to pick out the slight misreading of the Japanese article. |
[406.76 --> 416.28] While it uses Intel graphics, the chip is actually a test design for a better dynamic voltage frequency system implementation applicable to a wide range of chips. |
[416.58 --> 416.74] Oh. |
[416.74 --> 418.30] GPU, GPU, and FPGA. |
[418.76 --> 423.36] It was done on a GPU core because it was an easy design on 14 nanometer to start with. |
[423.76 --> 431.60] The Japanese article from which the translation comes, when auto-translated, does state that it's a discrete GPU prototype. |
[431.60 --> 440.66] But the author, who is very well-respected and knowledgeable, Gotoh-san, clarifies this point in the second paragraph by stating it was only a vehicle for it was actually being announced and tested. |
[441.04 --> 443.58] And the fact that it was built on a GPU was inconsequential. |
[444.12 --> 444.28] Okay. |
[444.28 --> 450.08] So, basically, no, in the, I don't know, has it been two months since Raja got hired at Intel? |
[450.40 --> 456.30] No, in the last however many weeks it's been, he has not built a GPU, necessarily. |
[456.56 --> 456.66] Yeah. |
[457.52 --> 458.36] He might have. |
[458.38 --> 458.66] I don't know. |
[458.68 --> 459.32] He's kind of amazing. |
[459.58 --> 460.78] But he has not. |
[461.88 --> 470.58] And this is like some power and clock management stuff that might end up in that GPU, but might not. |
[471.32 --> 471.98] But probably would. |
[472.20 --> 472.78] But might not. |
[472.78 --> 475.98] It sounds like Raja and whatever team he has is probably just playing. |
[476.96 --> 478.24] So, yeah, there you go. |
[478.40 --> 479.86] This is, yeah. |
[480.62 --> 483.08] Like it says, test vehicle, not a future product. |
[483.20 --> 484.46] It sounds like they're just playing around. |
[484.68 --> 486.06] So, let's go through our notes. |
[486.12 --> 491.86] With all of that in mind, let's go through the notes that we had based on other tech press' reporting. |
[491.86 --> 497.48] The first 14 nanometer dedicated GPU prototype is a two-chip solution. |
[497.48 --> 500.44] The first contains the GPU itself and the system agent. |
[500.60 --> 503.20] The second is an FPGA that interfaces with the system bus. |
[503.54 --> 505.54] It's 1.5 billion transistors. |
[505.54 --> 511.04] So, NVIDIA's GP102 is built on 60 nanometer and has 12 billion. |
[511.04 --> 514.34] But keep in mind, none of this really matters. |
[514.44 --> 514.66] Yes. |
[515.00 --> 516.56] Because that's not what it is at all. |
[516.66 --> 516.92] Yeah. |
[516.92 --> 525.48] And following media speculation about the slides, Intel has stated that they presented a research paper exploring new circuit techniques optimized for power management. |
[525.98 --> 527.72] And actually, you know what? |
[528.12 --> 529.06] Kudos to James. |
[529.52 --> 530.46] He had it right. |
[530.74 --> 532.44] He had all the clarifications in here. |
[532.68 --> 534.92] It is a test vehicle only, not a future product. |
[534.92 --> 535.26] It does. |
[535.26 --> 536.32] Much ado about nothing. |
[536.68 --> 537.10] In bold. |
[537.30 --> 537.42] Here. |
[537.58 --> 538.74] You guys can see James' notes. |
[539.02 --> 539.70] Credit to James. |
[540.02 --> 540.64] Good job, James. |
[541.26 --> 542.00] Good job, James. |
[542.22 --> 542.80] Good job, James. |
[542.92 --> 547.12] You'll, you'll, you know, whether you like it or not, you'll have to work here another day. |
[548.34 --> 549.48] Because you're not allowed to leave. |
[550.30 --> 550.88] Actually, wait. |
[550.98 --> 551.18] No. |
[551.36 --> 551.54] No. |
[551.66 --> 551.94] James. |
[552.06 --> 554.20] I have not added James to my soul collection yet. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.