willtheorangeguy/2018-WAN-Show-Transcripts

text stringlengths 14 502
[1019.54 --> 1029.98] so the instinct mi60 and the mi50 so these are aimed at the enterprise accelerator market with amd
[1029.98 --> 1035.68] looking to significantly improve their performance competitiveness in everything from hpc to machine
[1035.68 --> 1041.52] learning so these would be the kinds of things that they would do at sfu's um super computers so
[1041.52 --> 1047.94] researchers who are working on um oh man i forget what some of the examples of things that they were
[1047.94 --> 1055.40] doing there were but things like simulation um like processing gigantic data sets um those can be
[1055.40 --> 1061.32] accelerated by gpu compute i what's the biology stuff that they do um sequencing
[1061.32 --> 1067.82] yep that kind of thing uh i mean we're we're just kind of guessing that's just what that space does
[1067.82 --> 1073.98] i don't know how much market share amd actually gained with their initial instinct processors but
[1073.98 --> 1081.20] these look like uh an improvement so they've got support for half rate double precision up from 1 16th rate
[1081.20 --> 1087.18] so they've got uh so this is especially useful for machine learning inferencing where high precision isn't necessary
[1087.18 --> 1092.66] okay oh no they have new low precision data types that's useful for machine learning inferencing where
[1092.66 --> 1096.90] high precision isn't necessary and they're also able to get up to four times the performance
[1096.90 --> 1104.50] of an fp16 data type when using the smallest in four days okay okay so basically they've got some new
[1104.50 --> 1114.96] some new modes that they operate in pretty much peak throughput of 7.4 teraflops fp64 14.7 fp32
[1114.96 --> 1122.72] so okay neat yeah i mean this is the kind of thing that won't have a lot of relevance to you and me
[1122.72 --> 1130.58] necessarily at least if the rumors on the street are to be believed with respect to amd's consumer gpus
[1130.58 --> 1138.30] so the original article is from wcc f tech so we'll have to take it with the grain of salt that
[1138.30 --> 1143.38] people seem to think we should take it with from my experience their rumor stuff isn't really any worse
[1143.38 --> 1151.94] than anyone else's rumor stuff like is it i don't know um but apparently their third polaris gpu revision
[1151.94 --> 1160.44] is coming and it's going to be called the rx 590 so wow it looks like they've even gotten
[1160.44 --> 1168.24] their hands on the amd confidential nda required under embargo until oh there's this thing there
[1168.24 --> 1173.96] that says comments i can't see what's under embargo till but uh this looks like an official amd slide deck
[1173.96 --> 1179.90] um i don't even know why they bother having embargoes at this point
[1179.90 --> 1187.20] well i don't even know why they bother calling it something new at this point making this gpu
[1187.20 --> 1200.04] if if the rumors are true um it has um be based on 2304 cores compared to the radeon rx 580s 200
[1200.04 --> 1208.94] or 2304 cores yep so that's same number of course how much ram does it have eight gigabytes compared to
[1208.94 --> 1219.40] the eight gigabytes okay the bus width is 556 bit compared to 556 okay and the name is 590 though
[1219.40 --> 1224.42] that's that's a change that's a change that's an improvement and it's slightly higher clocked
[1224.42 --> 1232.36] okay so so what we don't know right now because we can see on that main title slide um latest
[1232.36 --> 1242.24] generation finfet 12 that it could be going from 14 nanometer to 12 nanometer sort of because if it
[1242.24 --> 1248.34] uses tsmc's 12 nanometer process that might actually be an improvement to power characteristics
[1248.34 --> 1256.22] and performance if it uses global foundry's 12 nanometer that's actually based very closely
[1256.22 --> 1263.50] like i it might even be a little disingenuous to call it 12 nanometer compared to the 14 nanometer
[1263.50 --> 1273.58] process that amd was already using on the rx 580 so um do they have actually do they have leaked
[1273.58 --> 1282.30] die sizes here at all uh no because that might tell us something interesting no they don't so what we
[1282.30 --> 1287.28] don't know is if this 12 nanometer shrink is just giving us some better power efficiency actually do
[1287.28 --> 1294.54] they have rated power power draw no no nothing leaked in power draw either so it's possible that
[1294.54 --> 1301.30] these are just overclocked but i mean really this was all amd could come up with uh to say to sell this
[1301.30 --> 1308.44] thing latest generation finfet 12 which quite frankly in my personal opinion should not be a selling
[1308.44 --> 1317.36] feature that you put in front of a consumer does that seem fair yeah why does my cousin care what
[1317.36 --> 1325.46] finfet process you're using he's not going to what matters is how many fps how many dollars i pay
[1325.46 --> 1333.80] what cool features it has like you know um automatic you know gameplay recording or special anti-aliasing
[1333.80 --> 1339.32] modes or whatever the case may be or drivers that work yeah how many finfets it has doesn't matter
[1339.32 --> 1346.52] okay selling point number two aggressive tuning for higher clocks so we already have that graphics card
[1346.52 --> 1357.42] that's called the rx 580 oc okay enhanced idle and multi-monitor efficiency that is a driver improvement
[1357.42 --> 1366.96] and finally expanded options and robust partner designs so basically they're saying they're not
[1366.96 --> 1374.40] making a reference card is that what that sounds like to you yeah because that's what that sounds
[1374.40 --> 1382.32] like to me all right so i'm not impressed expected pricing is 250 to 300 us what was the 580 right around
[1382.32 --> 1388.06] there i think they should just call it the 580 x and then we would be like awesome they made a faster
[1388.06 --> 1395.86] 580 good job guys instead of being like it's a 580 but now we're mad you know that is a really good
[1395.86 --> 1402.18] point and it's actually not a video about an amd product but we do have a video coming up about a
[1402.18 --> 1406.60] product that is currently embargoed so i can't say what it is unfortunately which is really frustrating
[1406.60 --> 1412.80] because it would make this story more interesting but basically we've got a product that we've got our
[1412.80 --> 1422.12] hands on already and we've been testing and the manufacturer refused to disclose to us certain
[1422.12 --> 1429.36] aspects of the specifications of the device and here's the stupid thing about it
[1429.36 --> 1444.32] it makes me feel kind of disrespected yeah like we have it we have it in our hands and we can measure
[1444.32 --> 1453.80] these characteristics it's not particularly complicated to do so so they're either disrespecting
[1453.80 --> 1459.68] our competence because they don't think that if they don't tell us we'll be able to put two and two
[1459.68 --> 1468.26] together here or they are disrespecting our time because now we have to invest time to measure something
[1468.26 --> 1476.60] that they could have simply provided on a spreadsheet or they are disrespecting our thoroughness because
[1476.60 --> 1483.40] they don't think that we're gonna bother to talk about this like there's no way that i can slice
[1483.40 --> 1489.88] this and feel respected maybe they're disrespecting our viewers that's another way because they don't
[1489.88 --> 1496.02] seem to think that they need the full story so why make it any easier to get there like can you come up
[1496.02 --> 1503.04] with a way that not providing some spec of your product is respectful to your partners because that's what
[1503.04 --> 1512.60] the media are no i can't and it's just it's just baffling and that's what i honestly that's what i feel
[1512.60 --> 1519.60] like a lot of these kinds of skews um whether the 590 turns out to be what it looks like it might be
[1519.60 --> 1527.16] which is just a rehash of the 580 but like maybe from a different supplier depending on what manufacturer
[1527.16 --> 1531.92] they're using for the chips um and this other thing that i'm talking about it just like these these
[1531.92 --> 1538.82] rebrands it just feels like why even bother then why even seed us a unit why not just say outright
[1538.82 --> 1544.00] we are trying to mislead the general consumer into thinking this is something new when it isn't
[1544.00 --> 1549.62] actually do you think they're doing the 590 because they did the 580 whatever sp the
[1549.62 --> 1557.02] they rebranded the 570 to be a 580 i really don't like what amd is doing on their gpu side like i
[1557.02 --> 1563.18] understand why they're doing it they're doing it because they got nothing it's like it's very obvious
[1563.18 --> 1569.04] they just they got nothing and nvidia has new stuff that they are obviously planning to roll out
[1569.04 --> 1574.70] and amd's like well that's the way it works like when your competitor launches something you got to
[1574.70 --> 1578.40] launch something so if we can't launch something new i guess we'll launch what we already got
[1578.40 --> 1585.02] um and they already did that with the with the 580 but at least they called it something different
[1585.02 --> 1589.18] no that's what they're doing again they're just calling it something different
[1589.18 --> 1594.80] no the 580 did i okay my understanding is the 580 was a different die even though it was only
[1594.80 --> 1604.64] slightly slightly optimized okay so actually okay 480 all right all right 2304 590 same thing
[1604.64 --> 1612.26] oh wait when is the last time amd launched a new graphics card well they did vega but
[1612.26 --> 1620.34] yeah they don't exist really do they not like can you buy a vega 56 on new egg like you couldn't
[1620.34 --> 1626.18] during the whole mining thing like they were impossible to get but i don't really know why
[1626.18 --> 1635.04] you would like you can get them in an imac pro yeah okay there's an azrock there's an azrock
[1635.04 --> 1643.42] reference card on new egg uh there's a power color mini that's pretty cool or nano rather oh cool
[1643.42 --> 1650.22] um okay yeah there's a handful of vega 56 is in stock but then okay let's take a pretty well-known
[1650.22 --> 1660.94] brand like xfx you could make the argument xfx is kind of too amd what evga is to nvidia at least you
[1660.94 --> 1664.72] used to be able to they used to have like a lifetime warranty and stuff actually i don't think
[1664.72 --> 1670.98] they do anymore anyway let's take some brand xfx and let's try to buy 10 of these and let's update
[1670.98 --> 1676.90] our cart um how do you update your cart update quantities and it's looking like that you know
[1676.90 --> 1682.42] what 70s are considerably cheaper they've got uh no they've got 10 in stock though so apparently
[1682.42 --> 1689.24] no they've got okay new egg used to correct you like like here let's see because there's no way
[1689.24 --> 1695.42] they have like 50 of these things yeah okay they have 49 of these in stock so apparently they're
[1695.42 --> 1701.54] actually selling all right maybe we didn't give rx vega enough credit or maybe they have 49 in stock
[1701.54 --> 1706.82] because they can't figure out how to get rid of them i don't know the point is i don't like what amd
[1706.82 --> 1711.92] is doing on the graphic side of things but uh at the end of the day as long as they're kicking ass
[1711.92 --> 1718.06] on one side of the business then it seems like they can keep surviving because for a while there it was
[1718.06 --> 1725.00] graphics that was keeping them alive and they're like okay you guys just just please just carry us
[1725.00 --> 1730.30] deadweight cpu guys until we can until we can get zen out and then and then you can rest
[1730.30 --> 1741.18] um all right okay so apple blocking linux installs why would they do this yeah i just don't get it

[1019.54 --> 1029.98] so the instinct mi60 and the mi50 so these are aimed at the enterprise accelerator market with amd

[1029.98 --> 1035.68] looking to significantly improve their performance competitiveness in everything from hpc to machine

[1035.68 --> 1041.52] learning so these would be the kinds of things that they would do at sfu's um super computers so

[1041.52 --> 1047.94] researchers who are working on um oh man i forget what some of the examples of things that they were

[1047.94 --> 1055.40] doing there were but things like simulation um like processing gigantic data sets um those can be

[1055.40 --> 1061.32] accelerated by gpu compute i what's the biology stuff that they do um sequencing

[1061.32 --> 1067.82] yep that kind of thing uh i mean we're we're just kind of guessing that's just what that space does

[1067.82 --> 1073.98] i don't know how much market share amd actually gained with their initial instinct processors but

[1073.98 --> 1081.20] these look like uh an improvement so they've got support for half rate double precision up from 1 16th rate

[1081.20 --> 1087.18] so they've got uh so this is especially useful for machine learning inferencing where high precision isn't necessary

[1087.18 --> 1092.66] okay oh no they have new low precision data types that's useful for machine learning inferencing where

[1092.66 --> 1096.90] high precision isn't necessary and they're also able to get up to four times the performance

[1096.90 --> 1104.50] of an fp16 data type when using the smallest in four days okay okay so basically they've got some new

[1104.50 --> 1114.96] some new modes that they operate in pretty much peak throughput of 7.4 teraflops fp64 14.7 fp32

[1114.96 --> 1122.72] so okay neat yeah i mean this is the kind of thing that won't have a lot of relevance to you and me

[1122.72 --> 1130.58] necessarily at least if the rumors on the street are to be believed with respect to amd's consumer gpus

[1130.58 --> 1138.30] so the original article is from wcc f tech so we'll have to take it with the grain of salt that

[1138.30 --> 1143.38] people seem to think we should take it with from my experience their rumor stuff isn't really any worse

[1143.38 --> 1151.94] than anyone else's rumor stuff like is it i don't know um but apparently their third polaris gpu revision

[1151.94 --> 1160.44] is coming and it's going to be called the rx 590 so wow it looks like they've even gotten

[1160.44 --> 1168.24] their hands on the amd confidential nda required under embargo until oh there's this thing there

[1168.24 --> 1173.96] that says comments i can't see what's under embargo till but uh this looks like an official amd slide deck

[1173.96 --> 1179.90] um i don't even know why they bother having embargoes at this point

[1179.90 --> 1187.20] well i don't even know why they bother calling it something new at this point making this gpu

[1187.20 --> 1200.04] if if the rumors are true um it has um be based on 2304 cores compared to the radeon rx 580s 200

[1200.04 --> 1208.94] or 2304 cores yep so that's same number of course how much ram does it have eight gigabytes compared to

[1208.94 --> 1219.40] the eight gigabytes okay the bus width is 556 bit compared to 556 okay and the name is 590 though

[1219.40 --> 1224.42] that's that's a change that's a change that's an improvement and it's slightly higher clocked

[1224.42 --> 1232.36] okay so so what we don't know right now because we can see on that main title slide um latest

[1232.36 --> 1242.24] generation finfet 12 that it could be going from 14 nanometer to 12 nanometer sort of because if it

[1242.24 --> 1248.34] uses tsmc's 12 nanometer process that might actually be an improvement to power characteristics

[1248.34 --> 1256.22] and performance if it uses global foundry's 12 nanometer that's actually based very closely

[1256.22 --> 1263.50] like i it might even be a little disingenuous to call it 12 nanometer compared to the 14 nanometer

[1263.50 --> 1273.58] process that amd was already using on the rx 580 so um do they have actually do they have leaked

[1273.58 --> 1282.30] die sizes here at all uh no because that might tell us something interesting no they don't so what we

[1282.30 --> 1287.28] don't know is if this 12 nanometer shrink is just giving us some better power efficiency actually do

[1287.28 --> 1294.54] they have rated power power draw no no nothing leaked in power draw either so it's possible that

[1294.54 --> 1301.30] these are just overclocked but i mean really this was all amd could come up with uh to say to sell this

[1301.30 --> 1308.44] thing latest generation finfet 12 which quite frankly in my personal opinion should not be a selling

[1308.44 --> 1317.36] feature that you put in front of a consumer does that seem fair yeah why does my cousin care what

[1317.36 --> 1325.46] finfet process you're using he's not going to what matters is how many fps how many dollars i pay

[1325.46 --> 1333.80] what cool features it has like you know um automatic you know gameplay recording or special anti-aliasing

[1333.80 --> 1339.32] modes or whatever the case may be or drivers that work yeah how many finfets it has doesn't matter

[1339.32 --> 1346.52] okay selling point number two aggressive tuning for higher clocks so we already have that graphics card

[1346.52 --> 1357.42] that's called the rx 580 oc okay enhanced idle and multi-monitor efficiency that is a driver improvement

[1357.42 --> 1366.96] and finally expanded options and robust partner designs so basically they're saying they're not

[1366.96 --> 1374.40] making a reference card is that what that sounds like to you yeah because that's what that sounds

[1374.40 --> 1382.32] like to me all right so i'm not impressed expected pricing is 250 to 300 us what was the 580 right around

[1382.32 --> 1388.06] there i think they should just call it the 580 x and then we would be like awesome they made a faster

[1388.06 --> 1395.86] 580 good job guys instead of being like it's a 580 but now we're mad you know that is a really good

[1395.86 --> 1402.18] point and it's actually not a video about an amd product but we do have a video coming up about a

[1402.18 --> 1406.60] product that is currently embargoed so i can't say what it is unfortunately which is really frustrating

[1406.60 --> 1412.80] because it would make this story more interesting but basically we've got a product that we've got our

[1412.80 --> 1422.12] hands on already and we've been testing and the manufacturer refused to disclose to us certain

[1422.12 --> 1429.36] aspects of the specifications of the device and here's the stupid thing about it

[1429.36 --> 1444.32] it makes me feel kind of disrespected yeah like we have it we have it in our hands and we can measure

[1444.32 --> 1453.80] these characteristics it's not particularly complicated to do so so they're either disrespecting

[1453.80 --> 1459.68] our competence because they don't think that if they don't tell us we'll be able to put two and two

[1459.68 --> 1468.26] together here or they are disrespecting our time because now we have to invest time to measure something

[1468.26 --> 1476.60] that they could have simply provided on a spreadsheet or they are disrespecting our thoroughness because

[1476.60 --> 1483.40] they don't think that we're gonna bother to talk about this like there's no way that i can slice

[1483.40 --> 1489.88] this and feel respected maybe they're disrespecting our viewers that's another way because they don't

[1489.88 --> 1496.02] seem to think that they need the full story so why make it any easier to get there like can you come up

[1496.02 --> 1503.04] with a way that not providing some spec of your product is respectful to your partners because that's what

[1503.04 --> 1512.60] the media are no i can't and it's just it's just baffling and that's what i honestly that's what i feel

[1512.60 --> 1519.60] like a lot of these kinds of skews um whether the 590 turns out to be what it looks like it might be

[1519.60 --> 1527.16] which is just a rehash of the 580 but like maybe from a different supplier depending on what manufacturer

[1527.16 --> 1531.92] they're using for the chips um and this other thing that i'm talking about it just like these these

[1531.92 --> 1538.82] rebrands it just feels like why even bother then why even seed us a unit why not just say outright

[1538.82 --> 1544.00] we are trying to mislead the general consumer into thinking this is something new when it isn't

[1544.00 --> 1549.62] actually do you think they're doing the 590 because they did the 580 whatever sp the

[1549.62 --> 1557.02] they rebranded the 570 to be a 580 i really don't like what amd is doing on their gpu side like i

[1557.02 --> 1563.18] understand why they're doing it they're doing it because they got nothing it's like it's very obvious

[1563.18 --> 1569.04] they just they got nothing and nvidia has new stuff that they are obviously planning to roll out

[1569.04 --> 1574.70] and amd's like well that's the way it works like when your competitor launches something you got to

[1574.70 --> 1578.40] launch something so if we can't launch something new i guess we'll launch what we already got

[1578.40 --> 1585.02] um and they already did that with the with the 580 but at least they called it something different

[1585.02 --> 1589.18] no that's what they're doing again they're just calling it something different

[1589.18 --> 1594.80] no the 580 did i okay my understanding is the 580 was a different die even though it was only

[1594.80 --> 1604.64] slightly slightly optimized okay so actually okay 480 all right all right 2304 590 same thing

[1604.64 --> 1612.26] oh wait when is the last time amd launched a new graphics card well they did vega but

[1612.26 --> 1620.34] yeah they don't exist really do they not like can you buy a vega 56 on new egg like you couldn't

[1620.34 --> 1626.18] during the whole mining thing like they were impossible to get but i don't really know why

[1626.18 --> 1635.04] you would like you can get them in an imac pro yeah okay there's an azrock there's an azrock

[1635.04 --> 1643.42] reference card on new egg uh there's a power color mini that's pretty cool or nano rather oh cool

[1643.42 --> 1650.22] um okay yeah there's a handful of vega 56 is in stock but then okay let's take a pretty well-known

[1650.22 --> 1660.94] brand like xfx you could make the argument xfx is kind of too amd what evga is to nvidia at least you

[1660.94 --> 1664.72] used to be able to they used to have like a lifetime warranty and stuff actually i don't think

[1664.72 --> 1670.98] they do anymore anyway let's take some brand xfx and let's try to buy 10 of these and let's update

[1670.98 --> 1676.90] our cart um how do you update your cart update quantities and it's looking like that you know

[1676.90 --> 1682.42] what 70s are considerably cheaper they've got uh no they've got 10 in stock though so apparently

[1682.42 --> 1689.24] no they've got okay new egg used to correct you like like here let's see because there's no way

[1689.24 --> 1695.42] they have like 50 of these things yeah okay they have 49 of these in stock so apparently they're

[1695.42 --> 1701.54] actually selling all right maybe we didn't give rx vega enough credit or maybe they have 49 in stock

[1701.54 --> 1706.82] because they can't figure out how to get rid of them i don't know the point is i don't like what amd

[1706.82 --> 1711.92] is doing on the graphic side of things but uh at the end of the day as long as they're kicking ass

[1711.92 --> 1718.06] on one side of the business then it seems like they can keep surviving because for a while there it was

[1718.06 --> 1725.00] graphics that was keeping them alive and they're like okay you guys just just please just carry us

[1725.00 --> 1730.30] deadweight cpu guys until we can until we can get zen out and then and then you can rest

[1730.30 --> 1741.18] um all right okay so apple blocking linux installs why would they do this yeah i just don't get it