willtheorangeguy/2018-WAN-Show-Transcripts

text stringlengths 14 502
[649.48 --> 658.74] I don't think I've been a computer enthusiast long enough for AMD to have even been on the same process node as Intel.
[658.74 --> 673.50] And I don't mean where they, like, Intel went ahead and then AMD caught up and then Intel went, like, I mean for AMD to have, like, launched at around a similar time with using the same manufacturing process as Intel.
[673.50 --> 686.92] So just for, for those of you who aren't familiar with what this means, there are a number of things that make a CPU or a GPU or any kind of processor inherently better.
[686.92 --> 696.12] So adding more cache, for example, adding smarter cache, that is, as long as it's not suffering from Spectre or meltdown related issues.
[696.86 --> 698.44] Minor problems.
[698.72 --> 700.32] Increasing the clock speed.
[700.62 --> 701.72] All of these things help.
[701.80 --> 706.36] So architectural changes to the design of the chip and how its logic works.
[706.78 --> 709.86] These help performance and can potentially help efficiency.
[709.86 --> 719.32] But the kind of freebie that for decades now we've pretty much taken for granted is shrinking the manufacturing process.
[719.32 --> 720.80] So the size of the transistors.
[721.04 --> 731.20] With a couple of notable exceptions, this inherently makes the chips more power efficient and smaller, allowing you to increase the complexity.
[731.62 --> 736.44] I mean, that's why a CPU today can have in excess of, like, 20 megabytes of cache.
[736.44 --> 754.70] Whereas back when I was getting into this stuff, I mean, having 256 kilobytes or 512 kilobytes was considered pretty darn good.
[754.70 --> 772.72] So AMD, by manufacturing on a smaller process, inherently has an advantage in terms of power consumption and chip complexity that Intel has just taken for granted for the last, what, 20 years?
[773.58 --> 775.12] Yeah, pretty much.
[775.64 --> 777.28] It's given them their huge advantage.
[779.18 --> 780.88] That's incredible.
[780.88 --> 786.14] So, I mean, there's a couple other just sort of housekeeping items here.
[786.32 --> 788.46] They're shipping Canon Lake in low volume.
[789.68 --> 792.42] Haven't pointed to specific customers or products.
[794.58 --> 801.88] And basically, their multi-patterning process is generating too many yield-reducing defects to produce 10 nanometer cost effectively.
[802.28 --> 806.46] So that's another thing that the layperson would not necessarily be aware of.
[806.46 --> 811.40] When you're manufacturing, it's not just about can you make a CPU.
[812.02 --> 818.14] Intel could probably make a CPU at 10 nanometer or even a node or even two nodes below.
[818.38 --> 823.82] If all they had to make was one CPU and it could cost them $6 million to produce or whatever the case may be.
[823.82 --> 839.50] But for them to make it cost effective enough that they can bring these 10 nanometer products to market and actually sell them at a price that would be acceptable to consumers and competitive with their own 14 nanometer products,
[839.50 --> 845.62] well, for that, you have to have a significant amount of the wafer actually be usable.
[846.18 --> 847.76] And they can accept small defects.
[847.90 --> 854.64] You'll often see that in the form of something like a Core i5 or a Core i3 where they'll turn off some cash or turn off a couple of cores.
[854.96 --> 856.08] That's perfectly acceptable.
[856.54 --> 863.38] But for them to not ship at all means they are not getting enough even usable dyes to productize the thing.
[863.38 --> 874.82] Well, we may actually see AMD pull ahead in performance then for a change because Ryzen's awesome, but it's still per Core not quite as fast.
[874.90 --> 876.34] But what if that changes by next year?
[876.86 --> 886.48] So one thing that's sort of a potential ace up AMD's sleeve is that we've seen second generation Ryzen.
[886.48 --> 892.36] And I mean, whatever the hype train might have said, you know, our thoughts on it were, okay.
[893.44 --> 896.04] I mean, yeah, it's more better for the same price.
[896.12 --> 897.56] I'm certainly not going to say no.
[897.70 --> 897.90] Yeah.
[898.26 --> 899.90] No, send it back.
[900.40 --> 901.50] Make another one.
[901.52 --> 902.42] I don't want the improvements.
[903.34 --> 907.12] But there was nothing revolutionary about it.
[907.34 --> 913.82] What we haven't seen from AMD yet is something that they have clearly indicated is on their roadmap, and that is Zen 2.
[913.82 --> 929.70] So that's an actual architectural improvement to Zen that's more than just, okay, we shrunk the node a little bit, like a fraction of a node down, and we've reduced power consumption a little bit, even though it ended up not actually amounting to much, and we've increased clock speed.
[931.28 --> 936.04] So can you imagine what the world looks like where AMD outperforms Intel?
[936.04 --> 944.68] It's kind of exciting, because I think it'll light a fire under Intel again.
[945.22 --> 946.72] And we've been saying this for a long time.
[946.80 --> 960.82] It's going to be weird consumer-wise, because even if AMD is ahead of Intel in that way, it'll be interesting to see how fast or even if the market really shifts that much, because people have been ingrained to just buy Intel stuff for so long now.
[960.82 --> 962.92] I'm going to go ahead.
[963.00 --> 967.52] I'm going to say something that might be a little unpopular with the AMD fanboys out there.
[968.60 --> 971.98] I don't think it's going to affect consumer choice very much at all.
[972.10 --> 975.64] That's where I was kind of going with that, is I think people are still going to buy it.
[976.32 --> 982.70] I think AMD is going to raise prices, and the people who are going to buy AMD are going to buy AMD, because it's a good choice.
[982.92 --> 984.98] And the people who buy Intel are going to buy Intel.
[984.98 --> 990.18] Well, that was kind of true back in the Pentium 4 days when you had Athlon XP.
[990.68 --> 990.82] Yep.
[991.20 --> 999.16] And in particular, I mean, the Pentium 4 wasn't even the worst, because the Pentium 4 legitimately outperformed the Athlon XP.
[999.82 --> 1012.22] I mean, us AMD people, you know, I had a 2500 plus, and I overclocked it to 2.2 gigahertz, which would have made it a 3200 plus by bumping the frontside bus from 333 to 400.
[1012.58 --> 1012.80] Okay?
[1012.80 --> 1014.96] No, no, crap.
[1015.06 --> 1016.50] No, it was from 266 to 330.
[1016.60 --> 1017.52] Whatever, whatever it was.
[1017.80 --> 1027.00] It was very confusing, because AMD back in those days would actually have multiple models of the same processor at different frontside bus speeds, and then just with, like, different multipliers.
[1027.64 --> 1032.06] And because of board compatibility with, because remember, that was back when third-party chipsets existed.
[1032.30 --> 1032.48] Yeah.
[1032.66 --> 1032.88] Right?
[1033.00 --> 1034.12] Remember via chipsets?
[1034.28 --> 1034.68] Whoa.
[1034.92 --> 1035.32] Enforce?
[1035.66 --> 1036.08] Yeah, baby.
[1036.94 --> 1037.98] Blaster in the past.
[1038.16 --> 1038.84] I know, right?
[1038.84 --> 1049.48] So, anyway, I had a 2500 plus, and I had it overclocked to a 3200 plus, which AMD very optimistically would compare to Intel's 3.2 gigahertz.
[1049.48 --> 1055.44] But if we're being honest with ourselves, I was one of those people who was like, you know what?
[1055.52 --> 1057.68] Hyper threading is not real processor cores.
[1057.92 --> 1058.90] So, you know what?
[1059.00 --> 1062.84] My 32 hybrid plus is, like, as good as your 3.2 gigahertz P4.
[1062.84 --> 1067.04] But in actuality, hyper threading was a good thing.
[1067.24 --> 1074.16] And this wasn't something that I actually discovered until dual cores were sort of the norm in the enthusiast circles.
[1074.36 --> 1080.14] And I got my hands on a P4C 2.4 gigahertz.
[1080.34 --> 1083.78] So this was a Northwood C, and this was an M0 stepping.
[1084.44 --> 1086.14] This was a cool chip.
[1086.36 --> 1090.16] This is a 2.4 gigahertz chip that was capable of, like, 4 plus.
[1090.28 --> 1092.08] Do people hunt that stuff as much as they used to?
[1092.08 --> 1100.12] A P4C 800-E Deluxe.
[1100.98 --> 1102.90] That's an Asus board from back in the day.
[1103.00 --> 1106.46] That was actually the board that I had that I was playing around with this M0 stepping on.
[1106.88 --> 1111.82] And that M0 stepping was out of, like, a salvaged system someone had given to me when I built them a new one.
[1111.88 --> 1112.50] I was like, whoa!
[1112.74 --> 1113.98] You had no idea what you had here.
[1114.46 --> 1116.64] I mean, dual cores were obviously still better.
[1116.70 --> 1117.96] But I was like, oh, hyper threading worked.
[1118.02 --> 1118.50] This is cool.
[1118.58 --> 1119.22] I'm overclocked.
[1119.24 --> 1120.20] Man, that thing was fast.
[1120.20 --> 1122.54] Anyway, what am I talking about again?
[1123.28 --> 1123.50] Right.
[1123.68 --> 1125.16] A P4C 800-E Deluxe.
[1125.28 --> 1127.44] Those still go for $100 on eBay.

[649.48 --> 658.74] I don't think I've been a computer enthusiast long enough for AMD to have even been on the same process node as Intel.

[658.74 --> 673.50] And I don't mean where they, like, Intel went ahead and then AMD caught up and then Intel went, like, I mean for AMD to have, like, launched at around a similar time with using the same manufacturing process as Intel.

[673.50 --> 686.92] So just for, for those of you who aren't familiar with what this means, there are a number of things that make a CPU or a GPU or any kind of processor inherently better.

[686.92 --> 696.12] So adding more cache, for example, adding smarter cache, that is, as long as it's not suffering from Spectre or meltdown related issues.

[696.86 --> 698.44] Minor problems.

[698.72 --> 700.32] Increasing the clock speed.

[700.62 --> 701.72] All of these things help.

[701.80 --> 706.36] So architectural changes to the design of the chip and how its logic works.

[706.78 --> 709.86] These help performance and can potentially help efficiency.

[709.86 --> 719.32] But the kind of freebie that for decades now we've pretty much taken for granted is shrinking the manufacturing process.

[719.32 --> 720.80] So the size of the transistors.

[721.04 --> 731.20] With a couple of notable exceptions, this inherently makes the chips more power efficient and smaller, allowing you to increase the complexity.

[731.62 --> 736.44] I mean, that's why a CPU today can have in excess of, like, 20 megabytes of cache.

[736.44 --> 754.70] Whereas back when I was getting into this stuff, I mean, having 256 kilobytes or 512 kilobytes was considered pretty darn good.

[754.70 --> 772.72] So AMD, by manufacturing on a smaller process, inherently has an advantage in terms of power consumption and chip complexity that Intel has just taken for granted for the last, what, 20 years?

[773.58 --> 775.12] Yeah, pretty much.

[775.64 --> 777.28] It's given them their huge advantage.

[779.18 --> 780.88] That's incredible.

[780.88 --> 786.14] So, I mean, there's a couple other just sort of housekeeping items here.

[786.32 --> 788.46] They're shipping Canon Lake in low volume.

[789.68 --> 792.42] Haven't pointed to specific customers or products.

[794.58 --> 801.88] And basically, their multi-patterning process is generating too many yield-reducing defects to produce 10 nanometer cost effectively.

[802.28 --> 806.46] So that's another thing that the layperson would not necessarily be aware of.

[806.46 --> 811.40] When you're manufacturing, it's not just about can you make a CPU.

[812.02 --> 818.14] Intel could probably make a CPU at 10 nanometer or even a node or even two nodes below.

[818.38 --> 823.82] If all they had to make was one CPU and it could cost them $6 million to produce or whatever the case may be.

[823.82 --> 839.50] But for them to make it cost effective enough that they can bring these 10 nanometer products to market and actually sell them at a price that would be acceptable to consumers and competitive with their own 14 nanometer products,

[839.50 --> 845.62] well, for that, you have to have a significant amount of the wafer actually be usable.

[846.18 --> 847.76] And they can accept small defects.

[847.90 --> 854.64] You'll often see that in the form of something like a Core i5 or a Core i3 where they'll turn off some cash or turn off a couple of cores.

[854.96 --> 856.08] That's perfectly acceptable.

[856.54 --> 863.38] But for them to not ship at all means they are not getting enough even usable dyes to productize the thing.

[863.38 --> 874.82] Well, we may actually see AMD pull ahead in performance then for a change because Ryzen's awesome, but it's still per Core not quite as fast.

[874.90 --> 876.34] But what if that changes by next year?

[876.86 --> 886.48] So one thing that's sort of a potential ace up AMD's sleeve is that we've seen second generation Ryzen.

[886.48 --> 892.36] And I mean, whatever the hype train might have said, you know, our thoughts on it were, okay.

[893.44 --> 896.04] I mean, yeah, it's more better for the same price.

[896.12 --> 897.56] I'm certainly not going to say no.

[897.70 --> 897.90] Yeah.

[898.26 --> 899.90] No, send it back.

[900.40 --> 901.50] Make another one.

[901.52 --> 902.42] I don't want the improvements.

[903.34 --> 907.12] But there was nothing revolutionary about it.

[907.34 --> 913.82] What we haven't seen from AMD yet is something that they have clearly indicated is on their roadmap, and that is Zen 2.

[913.82 --> 929.70] So that's an actual architectural improvement to Zen that's more than just, okay, we shrunk the node a little bit, like a fraction of a node down, and we've reduced power consumption a little bit, even though it ended up not actually amounting to much, and we've increased clock speed.

[931.28 --> 936.04] So can you imagine what the world looks like where AMD outperforms Intel?

[936.04 --> 944.68] It's kind of exciting, because I think it'll light a fire under Intel again.

[945.22 --> 946.72] And we've been saying this for a long time.

[946.80 --> 960.82] It's going to be weird consumer-wise, because even if AMD is ahead of Intel in that way, it'll be interesting to see how fast or even if the market really shifts that much, because people have been ingrained to just buy Intel stuff for so long now.

[960.82 --> 962.92] I'm going to go ahead.

[963.00 --> 967.52] I'm going to say something that might be a little unpopular with the AMD fanboys out there.

[968.60 --> 971.98] I don't think it's going to affect consumer choice very much at all.

[972.10 --> 975.64] That's where I was kind of going with that, is I think people are still going to buy it.

[976.32 --> 982.70] I think AMD is going to raise prices, and the people who are going to buy AMD are going to buy AMD, because it's a good choice.

[982.92 --> 984.98] And the people who buy Intel are going to buy Intel.

[984.98 --> 990.18] Well, that was kind of true back in the Pentium 4 days when you had Athlon XP.

[990.68 --> 990.82] Yep.

[991.20 --> 999.16] And in particular, I mean, the Pentium 4 wasn't even the worst, because the Pentium 4 legitimately outperformed the Athlon XP.

[999.82 --> 1012.22] I mean, us AMD people, you know, I had a 2500 plus, and I overclocked it to 2.2 gigahertz, which would have made it a 3200 plus by bumping the frontside bus from 333 to 400.

[1012.58 --> 1012.80] Okay?

[1012.80 --> 1014.96] No, no, crap.

[1015.06 --> 1016.50] No, it was from 266 to 330.

[1016.60 --> 1017.52] Whatever, whatever it was.

[1017.80 --> 1027.00] It was very confusing, because AMD back in those days would actually have multiple models of the same processor at different frontside bus speeds, and then just with, like, different multipliers.

[1027.64 --> 1032.06] And because of board compatibility with, because remember, that was back when third-party chipsets existed.

[1032.30 --> 1032.48] Yeah.

[1032.66 --> 1032.88] Right?

[1033.00 --> 1034.12] Remember via chipsets?

[1034.28 --> 1034.68] Whoa.

[1034.92 --> 1035.32] Enforce?

[1035.66 --> 1036.08] Yeah, baby.

[1036.94 --> 1037.98] Blaster in the past.

[1038.16 --> 1038.84] I know, right?

[1038.84 --> 1049.48] So, anyway, I had a 2500 plus, and I had it overclocked to a 3200 plus, which AMD very optimistically would compare to Intel's 3.2 gigahertz.

[1049.48 --> 1055.44] But if we're being honest with ourselves, I was one of those people who was like, you know what?

[1055.52 --> 1057.68] Hyper threading is not real processor cores.

[1057.92 --> 1058.90] So, you know what?

[1059.00 --> 1062.84] My 32 hybrid plus is, like, as good as your 3.2 gigahertz P4.

[1062.84 --> 1067.04] But in actuality, hyper threading was a good thing.

[1067.24 --> 1074.16] And this wasn't something that I actually discovered until dual cores were sort of the norm in the enthusiast circles.

[1074.36 --> 1080.14] And I got my hands on a P4C 2.4 gigahertz.

[1080.34 --> 1083.78] So this was a Northwood C, and this was an M0 stepping.

[1084.44 --> 1086.14] This was a cool chip.

[1086.36 --> 1090.16] This is a 2.4 gigahertz chip that was capable of, like, 4 plus.

[1090.28 --> 1092.08] Do people hunt that stuff as much as they used to?

[1092.08 --> 1100.12] A P4C 800-E Deluxe.

[1100.98 --> 1102.90] That's an Asus board from back in the day.

[1103.00 --> 1106.46] That was actually the board that I had that I was playing around with this M0 stepping on.

[1106.88 --> 1111.82] And that M0 stepping was out of, like, a salvaged system someone had given to me when I built them a new one.

[1111.88 --> 1112.50] I was like, whoa!

[1112.74 --> 1113.98] You had no idea what you had here.

[1114.46 --> 1116.64] I mean, dual cores were obviously still better.

[1116.70 --> 1117.96] But I was like, oh, hyper threading worked.

[1118.02 --> 1118.50] This is cool.

[1118.58 --> 1119.22] I'm overclocked.

[1119.24 --> 1120.20] Man, that thing was fast.

[1120.20 --> 1122.54] Anyway, what am I talking about again?

[1123.28 --> 1123.50] Right.

[1123.68 --> 1125.16] A P4C 800-E Deluxe.

[1125.28 --> 1127.44] Those still go for $100 on eBay.