XenTegra - Nutanix Weekly

Nutanix Weekly: Full Frame Rate VDI Video - Nutanix and NVIDIA A16 GPUs eliminate the jaggies!

September 13, 2022 XenTegra / Andy Whiteside Season 1 Episode 59
XenTegra - Nutanix Weekly
Nutanix Weekly: Full Frame Rate VDI Video - Nutanix and NVIDIA A16 GPUs eliminate the jaggies!
Show Notes Transcript

The results are in! With NVIDIA’s newest GPUs designed for virtual desktops and applications, you can address a wider variety of use cases than you could ever before. In our independent testing using the NVIDIA® nVector benchmarking tool, we saw that a single A16 GPU could support 64 virtual desktops, each with 2vCPUs and 1GB framebuffer profiles, per node. With two A16 cards installed we were able to scale to 128, albeit we would recommend a high-performance CPU when scaling to 128 desktops. In a minimal Nutanix® configuration of a three node cluster that equates to 384 users per cluster, all having access to 1GB of framebuffer profile!

Host: Harvey Green
Co-host: Jirah Cox
Co-host: Ben Rogers

WEBVTT

1
00:00:03.800 --> 00:00:17.889
Jirah Cox: Welcome to episode fifty, nine of the newutenant's weekly podcast. I got two of my co-hosts here, this to Jarra Kat, who ran away as soon as I hit the record. But I Gyra, Hey, man, I like to keep it interesting.

2
00:00:17.900 --> 00:00:20.120
Jirah Cox: You were wondering if I was gonna come back.

3
00:00:20.540 --> 00:00:21.490
He's

4
00:00:21.500 --> 00:00:24.489
Harvey Green: and Mr. Ben Rogers. How are you today, sir?

5
00:00:24.500 --> 00:00:34.090
Jirah Cox: I will have to say, Harvey, my jaw drop My,

6
00:00:34.100 --> 00:00:35.090
Ben Rogers: that's right.

7
00:00:35.100 --> 00:00:51.890
Ben Rogers: You You did a good job. We were. We were talking really quick before this about episode numbers, and ah, I wanted to make sure I didn't forget the episode number, and Gyra made the comment that he was going to distract me in the first five science, and unknowingly did it. But I still got the number off,

8
00:00:51.900 --> 00:01:02.990
Jirah Cox: and some people also need to hear things multiple times. I mean, someone is wondering what episode number this was. What would that answer? Be? Fifty Nine.

9
00:01:03.000 --> 00:01:05.700
Jirah Cox: There we go nailed it

10
00:01:06.300 --> 00:01:21.419
Harvey Green: So all right. Um. Today's topic is a full frame rate. Vdi Video: Newtonics and Nvidia a sixteen Gpus eliminate the jaggies.

11
00:01:21.430 --> 00:01:30.460
Harvey Green: So that's That's fun jarless. Let's start with. Actually, we'll leave this with the Ben first. Then he'll take the first question of the day, sir.

12
00:01:30.610 --> 00:01:39.890
Ben Rogers: What is a jaggy? I think, back in my day it would have been considered the robot. David.

13
00:01:39.900 --> 00:01:48.269
Ben Rogers: You know It's just not a good video experience, and and sometimes it feels like the old cell phone days where you feel like you're missing part of the conversation.

14
00:01:48.280 --> 00:02:17.429
Ben Rogers: So getting that eliminated It's going to be pretty important,

15
00:02:17.440 --> 00:02:19.980
Ben Rogers: and, as we, you know, for business reasons also

16
00:02:20.700 --> 00:02:28.520
Jirah Cox: Very good. Not a not a British sports car that the Germans, looking to, you know, take over the market from not that kind of jaggy.

17
00:02:31.580 --> 00:02:38.499
Ben Rogers: I wonder if Jack? He was even a term man. I was embarrassed to even ass that. But I will. Is Jack even a technical term?

18
00:02:38.950 --> 00:02:46.010
Jirah Cox: It's I see it here in a technical blog,

19
00:02:52.360 --> 00:03:06.599
Ben Rogers: but as we were, you know, we'd like this to be thrashing pile on the fly. But as we discussed very briefly before this, you and I. This is kind of close to our heart. But we actually had to do this to the facility,

20
00:03:06.610 --> 00:03:25.169
Ben Rogers: and it was killer. It was cool, but it was with some caveats that I think are warranted to talk about today. So as we get into this, you know, Ah, being a great technology works. Ah, that you know in that, in my opinion, there are some things you have to pay attention to as your influence

21
00:03:25.180 --> 00:03:39.669
Harvey Green: absolutely. So let's let's get into it, and i'm sure that will definitely come out as we discuss it. Um, the the the blog post here is about Nvidia's newest gpus and what they can do for you.

22
00:03:39.680 --> 00:03:45.470
Harvey Green: The topic today is around the A sixteen gpu,

23
00:03:45.480 --> 00:04:03.930
Harvey Green: which on and videos, benchmarking tool can run and support sixty, four virtual desktops, which is like, Wow! We don't normally see that many out of a single gpu so definitely be definitely

24
00:04:04.300 --> 00:04:11.119
Harvey Green: want to look more into this particular one, but that one particular is

25
00:04:11.150 --> 00:04:23.740
Harvey Green: two B cpus and one gig for your frame buffer profiles per node, and they were able to use two, a sixteen cards to scale up to one hundred and twenty eight.

26
00:04:23.750 --> 00:04:35.530
Harvey Green: But they make the disclaimer there that you might want to make sure that you've got a high-performance Cpu before scaling to one hundred and twenty eight desktops.

27
00:04:35.540 --> 00:04:44.690
Jirah Cox: So i'll pause there for a second, because I said a whole lot of words, and I'm. Going to bring in our resident translator, Mr. Gyra Cox, to help translate that.

28
00:04:44.700 --> 00:04:47.740
Jirah Cox: Let's let's start with something easy.

29
00:04:47.810 --> 00:04:52.510
Harvey Green: What is a Gpu versus a Cpu? And why do I need both?

30
00:04:52.750 --> 00:05:12.050
Jirah Cox: Totally so? The the Cpu does all the general purpose uh thinking that runs all of the compute like in this case, like the host itself, but also the virtual desktop uh all the calculations in excel um powers a lot of your input out output, right? So like keyboard input network bits and so forth.

31
00:05:12.060 --> 00:05:30.250
Jirah Cox: The Gpu is the graphics card specifically so it makes things look prettier. Renders for graphics test video decoding. If you're playing games. It does lots of work for threed rendering. And of course, even in today's day and age, there's also Ai Ml. Work as well for machine learning and

32
00:05:30.260 --> 00:05:34.929
Jirah Cox: computational parallel processing that can happen on the on the Graphics card.

33
00:05:35.260 --> 00:05:51.919
Jirah Cox: Um, but yes, that you can, added the Gpu to really any desktop right laptop desktop virtual desktop to make things prettier. Do better. Graphics make certain kinds of operations run faster, run better, run more slowly. So that's It's not always been that way, since almost like the dawn of

34
00:05:51.930 --> 00:06:07.900
Jirah Cox: of enterprise. Vdi. Right like, you know, the last ten, fifteen years right is is, of course, you can run regular to Vms. But you'll get a better end user experience when you also can bolster the the graphics portion of them right with a gpu in the in the have a server.

35
00:06:08.600 --> 00:06:18.849
Harvey Green: So that's good that. But you said a lot of words, too. Why do I need both? Can I just have just Cpu? Can. I just have just Gpu:

36
00:06:19.280 --> 00:06:33.970
Jirah Cox: I can't have just Gpu for a end. User. Compute. Use case.

37
00:06:33.980 --> 00:06:47.819
Jirah Cox: Um, not not. Enterprise is great, not not terribly functional, But yeah, Cpu only would be valid if you don't have very serious requirements, for you know, and user experience aesthetics.

38
00:06:47.830 --> 00:07:01.060
Jirah Cox: The fancier apps like in the vent touch on right, like the video conference, saying they have decoding, You know, if you want to basically have a virtual desktop that that is indistinguishable from a but a physical desktop, you know, a real laptop.

39
00:07:01.070 --> 00:07:08.350
Jirah Cox: But you want the benefits of Vdi right? Patching back ups recovery desk, recovery

40
00:07:08.550 --> 00:07:22.789
Jirah Cox: um, even like loss prevention, you know, or data containment. Then we need to talk about having a higher tier Vdi experience which can often, you know, one key to that's going to be better graphics right through a Gpu.

41
00:07:22.800 --> 00:07:24.710
Harvey Green: All right, Absolutely

42
00:07:24.880 --> 00:07:43.880
Harvey Green: so. Ah, you know, in in my experience with Gpus. Ah, every time we talk about them it's always well, which users. Do you want to have a gpu set up? How many do you want to use, you know? And it's always

43
00:07:43.890 --> 00:07:57.109
Harvey Green: The conversation has always been about a subset of the users that you have in a Vdi type environment. And this brings up, and the second paragraph really is about.

44
00:07:57.120 --> 00:08:09.949
Jirah Cox: You know now that there is no real scalability limitations. I want you to fit sixty four people on a single gpu. Now we don't necessarily have to exclude people

45
00:08:10.090 --> 00:08:14.419
Harvey Green: go ahead, dry it out. I was gonna take it to you, anyway. So you go ahead.

46
00:08:14.430 --> 00:08:30.499
Jirah Cox: I don't know, and that that density, like with my background is just traditional virtualizationation design. It's always kind of thrown me for a loop. How the density fall off a cliff historically, when you have to add Gps to to your design, right? If you're talking sixteen years per card,

47
00:08:30.510 --> 00:08:38.389
Jirah Cox: maybe two cards per node. Thirty users per note is not very dense. These numbers definitely, you know, made me take notice when I can get

48
00:08:38.450 --> 00:08:47.289
Jirah Cox: sixty. Four users per card are up to right, and then up to one hundred and twenty users per node. That's a whole ton of users, right? That's actually really approaching

49
00:08:47.460 --> 00:08:56.540
Jirah Cox: like we talked about a couple of episodes ago with Ammd. Cpus. Right? I can fit. You know a comfortable enough amount of desktops on the node

50
00:08:56.720 --> 00:09:04.330
Jirah Cox: that I actually kind of quit packing users on from a risk standpoint right like How many users do I want one

51
00:09:04.340 --> 00:09:33.499
Jirah Cox: unplanned hardware failure to really affect at once Right? Two hundred and five hundred. That's probably too many users right? So um. This removes Gpu as a design constraint from a user's per node Kind of density standpoint right? And to your point really democratizes it right? But it gets it out to the masses that even these not really um huge desktops, right two B Cpu and a one gig frame buffer. That's that's pretty almost like issue low end on a per desktop basis. Right? That's just enough graphic source power to make windows

52
00:09:33.510 --> 00:09:45.690
Jirah Cox: run a little bit faster. Be a little prettier. Make some of your apps that we're going to hit that gpu like video decoding or or video conferencing, run a little faster. That's That's a very economical way to get into that.

53
00:09:46.090 --> 00:10:00.520
Harvey Green: Yeah, absolutely. So they talk a little bit about the results they got from the A. Sixteen gpus in that second paragraph two which they said yielded over a thirty percent improvement in frames per second

54
00:10:00.530 --> 00:10:13.779
Harvey Green: and latency approaching one twentieth of a second, which that's that's a mouthful there. Right Then why don't you tell us a little bit about what what they mean by frames per second, and why that matters one.

55
00:10:13.790 --> 00:10:22.470
Ben Rogers: Well, I might be off on my numbers, but then thirty, two frames per second, one hundred percent video in that you're delivering high by definition video.

56
00:10:22.480 --> 00:10:50.679
Ben Rogers: So if you look at the graph before they added the Gpu. You were lucky to get to twenty-five frames per second. So going back to the to the gig of the Td. Or whatever was called. Ah, once you get into the thirty being once you get to the thirty percent delivery range, you're on this full video. So one of the things I also want to go into. The conversation is, you know, Ah, Harvey, you were talking about. This used to be a very selective group, you know that was definitely true. I had to

57
00:10:50.690 --> 00:11:05.290
Ben Rogers: the surgeons and the trauma people all the Gpu machines because they were doing, you know,

58
00:11:05.300 --> 00:11:34.250
Ben Rogers: dicom diagnostic environment. So I know that before we got the Gpus I had very little adoption in the surgery realm, because for me to see these numbers,

59
00:11:34.260 --> 00:11:56.060
Ben Rogers: it's not that I don't believe them. I believe they were done in a lab. But I wonder how that's going to wait in a real world where you have different users within an organization using different different applications, like a radiology application, or an architecture cat, or you know what's that? Sixty, four number going to look like when you get past the knowledge.

60
00:11:56.070 --> 00:12:23.949
Harvey Green: Yeah. So the the big to do around that will all will always be around what you're using for your framework brain buffer profiles that they have listed here for these. They're talking one gig um, and and you know, for a lot of users one gig might be what they need right. All of that will depend on the resolution that they're using, and how many monitors they're using the resolution on. So

61
00:12:23.960 --> 00:12:28.759
Harvey Green: if you're just trying to Do you know one thousand and eightyp type resolution

62
00:12:28.830 --> 00:12:40.219
Harvey Green: that might get you where you need to be but a lot of other, You know other display types or other, you know, workloads potentially could be

63
00:12:40.230 --> 00:12:59.409
Harvey Green: a whole lot more ah detail than than that kind of picture, and that's where we get into some of those other use cases, then, that you're bringing up. Ah! Where you might want to go up on their on their profile, and the amount of memory that's dedicated to them out of that Gpu.

64
00:12:59.420 --> 00:13:12.639
Harvey Green: And so, really, you know, if we're, if we're getting down to the brass tax like that, we're saying that these a sixteen Gpus have sixty four gigs of memory that you can divide out the way that you want

65
00:13:12.650 --> 00:13:19.489
Harvey Green: in particular for this article. They talk about doing that for as many users as he can

66
00:13:19.500 --> 00:13:39.289
Harvey Green: to Max it out to get as as dense as you can go. But to your point, yeah, we can get a lot more detail. Get a lot better utilization out of, You know. Much bigger monitors, or you know much more detailed views than like a one thousand and eighty P. Type of year. So

67
00:13:39.300 --> 00:13:47.780
Harvey Green: yes, it Yes, it will depend on the users. Yes, it will depend on the workload. But I mean going from,

68
00:13:48.020 --> 00:13:56.309
Harvey Green: you know the conversation around. Oh, yeah, We used to have Gpus, and we were lucky to get eight on there. And now we're talking about the possibility of sixty four.

69
00:13:56.320 --> 00:13:59.090
Ben Rogers: Wow, man, that's still doing it.

70
00:13:59.100 --> 00:14:03.549
Ben Rogers: That's a big increase for sure. Yeah, that's a huge increase.

71
00:14:03.910 --> 00:14:22.989
Ben Rogers: Ah, and we're also talking about. You know the Knowledge Worker is becoming the power user And specifically, you know, we were talking about teams, video conferencing. Yes, we're getting the push to get back in the office, but it doesn't mean we don't have to communicate with colleagues that are outside of our environment

72
00:14:23.000 --> 00:14:36.750
Harvey Green: absolutely. And you know, for those who end up watching this on Youtube. You'll see all three of our little faces moving around. That that just is another illustration of what we want to make sure is happening.

73
00:14:36.760 --> 00:14:55.380
Harvey Green: Um! And doing that in a Vdi or virtual desktop type of environment is is as important as you know, having that sit on a laptop right in front of you, because you know, a lot of people don't necessarily know laptops that carry gpus for a pretty long time now,

74
00:14:55.390 --> 00:15:06.129
Harvey Green: but that's always one to one right, because you only have one potential, you know, one monitor, sometimes two monitors out of a laptop

75
00:15:06.140 --> 00:15:24.560
Harvey Green: mit ctl, and and that's not a huge low for a gpu to carry as opposed to something like this, where we're talking about potentially providing for sixty, four users per card so definitely different. Use cases there, but you know, giving the customers the end users one hundred and fifty,

76
00:15:24.570 --> 00:15:34.850
Harvey Green: the workers something that is a much more rich environment for them to work with and actually use that they'll be a lot happier with

77
00:15:36.390 --> 00:15:39.749
Jirah Cox: Well, and as those

78
00:15:40.580 --> 00:15:45.789
Jirah Cox: to the point of the users expectations and requirements keep growing right the way

79
00:15:45.800 --> 00:16:05.569
Jirah Cox: right? I mean every year, right? And to your point around laptops, and even even standard like with laptop Cpus now actually often are like merging with the Gpu. Yeah, you can build it on there. So so it really puts the honest on the enterprise to say we we,

80
00:16:05.710 --> 00:16:22.159
Jirah Cox: where possible, we need to be done designing Gpuless Pdi right? Because it's going to be a pretty dramatic step down in user experience compared to the standard status quo. Right? So like we have to up our game, for we're delivering to our end users

81
00:16:22.170 --> 00:16:36.109
Jirah Cox: in the article calls not even chrome, you know it looks for a Gp. By default when you when you run that right, no matter where where you run it on a laptop desktop, a virtual desktop. So it it gets important, because, like traditional data, center, grade,

82
00:16:36.120 --> 00:16:44.049
Jirah Cox: enterprise, cpus don't include that Gpu capability. Right? You add that separately where you where you need it. So you can't assume that it's there,

83
00:16:44.620 --> 00:17:02.359
Jirah Cox: and, like you like. You see, Vdi always has been right. It's always about um. What does it do for the business? Not ah, directly about like costs comparison per desktop. It's that's a rational, like a tethering kind of math to do around cost per virtual desktop.

84
00:17:02.770 --> 00:17:20.550
Jirah Cox: Right? It's more like this is unlocking capability to say I get all of the the provisioning desktop recovery, agility, scalability, um data, loss prevention outcomes of a good Eu Cbd: I environment. And now I also don't compromise on Gpu. Either it's the right way to think about this

85
00:17:20.560 --> 00:17:21.430
Harvey Green: right,

86
00:17:22.140 --> 00:17:38.690
Harvey Green: and so kind of in the third paragraph they talk more about the potential use cases and different work profiles with Ben. You've already, you know, hit and address. So once again, So you are before your time.

87
00:17:38.700 --> 00:17:40.090
Ben Rogers: Get out of the rooms

88
00:17:41.600 --> 00:17:42.990
Ben Rogers: from the drummer in the group.

89
00:17:43.000 --> 00:17:46.070
Harvey Green: That's right.

90
00:17:46.080 --> 00:18:01.290
Harvey Green: So you know the the other thing a jar of everything that you just talked through it. It kind of made me feel like either I'm getting old or they're getting faster, because I remember when a sixty, four Meg video card was like really good.

91
00:18:01.300 --> 00:18:03.590
Jirah Cox: I mean

92
00:18:03.600 --> 00:18:06.989
Jirah Cox: they could be. They're getting older. It could be both.

93
00:18:07.980 --> 00:18:14.690
Jirah Cox: Yeah. And now we're talking sixty, four gig, Which is, It's kind of crazy. It's ridiculous.

94
00:18:14.700 --> 00:18:25.389
Ben Rogers: I know my crypto Mining friends are going to be disappointed that all this technologies now coming out the time that they're taking the mining capabilities away. So i'm sure some of them are

95
00:18:25.400 --> 00:18:29.650
Ben Rogers: looking at this going man. Why couldn't it spin out, and we had the ability to mine it

96
00:18:29.660 --> 00:18:31.170
Jirah Cox: need for I drew.

97
00:18:32.080 --> 00:18:51.720
Harvey Green: So you know the next paragraph we're talking through about prior to the ability to provision Gpus at scale disaster, recovery for these use cases traditionally needed to be very, very carefully thought out, and planned with

98
00:18:51.730 --> 00:19:04.490
Ben Rogers: complicated workflows than m-packers rpos and rtos. We'll we'll throw you another one here and then alphabet suit, Rpos and Rtos. What are we talking about here?

99
00:19:04.500 --> 00:19:27.680
Ben Rogers: Why are you recovering the

100
00:19:27.690 --> 00:19:41.389
Ben Rogers: You've impacted the business in some way, shape or form? And That's why you have a business continuity plan to, you know, Deal with that outage. And how do you run the business without being one hundred percent on the technology side,

101
00:19:41.400 --> 00:19:54.100
Ben Rogers: where I think this is very fascinating. And this was one thing that we were kind of teen up before we got the record button on was. This used to be very difficult in a Dr. Situation, because you had to have life for life.

102
00:19:54.110 --> 00:20:08.290
Ben Rogers: And the reason you had to have like for like is, you know, in my environment, doing a very surgeon. They didn't accept outages, you know, if they were in a or and they needed to look at the scan, and the only way they could get to that so, and was through a citric session,

103
00:20:08.300 --> 00:20:17.210
Ben Rogers: and they had any doubt in the quality of the scan, you know they would cancel the operation. That's how meticulously these doctors are.

104
00:20:17.220 --> 00:20:45.859
Ben Rogers: And so for me. Uh, you know, it had to be whatever wherever they landed, whether it be in the primary site or

105
00:20:46.460 --> 00:21:07.409
Ben Rogers: it used to be, You know you couldn't fail. You wouldn't want to fail the user and and that had a gpu to a session that didn't have a gpu. It sounds like that. Some of the capabilities might be there, or this this this paragraph. I think I might have read a little bit more into it that's actually there. So I asked the question,

106
00:21:07.420 --> 00:21:19.499
Ben Rogers: Is this getting as easy? Where, if you'd have these gpus and all these boxes. You could sell these users over. No problem. It doesn't matter. They'll land on a session and continue to move along like like they would.

107
00:21:20.080 --> 00:21:24.510
Ben Rogers: And what would happen if they landed on a session that didn't have a Gp:

108
00:21:25.250 --> 00:21:43.490
Jirah Cox: Yeah. So this is highlighting that in the past. It's been a um even with like, for like clusters keep use on both sides. Ah, zest, recovery, planning, and activation has been a little bit more challenging because of the extra steps needed to say, If we have a Gp over there, I want to have a gpu, or when it feels over here, make all that work.

109
00:21:43.500 --> 00:21:52.189
Jirah Cox: So this is highlighting that that the the real win here right is that now you know, Gpus for virtual machines on Hv.

110
00:21:52.200 --> 00:22:04.109
Jirah Cox: Our first-class citizens right so you can fail those over from a cluster to another cluster. Pick up more gpus and keep her out and running, so you know faster, fail over less down time for your those most demanding customers like your hell outing ben

111
00:22:04.120 --> 00:22:26.210
Jirah Cox: um, and then also i'm pretty sure. Ah, that you also can fill in from Gpu to non gpu, if you want to know, like a lower cost, Cdr. Cluster as well. Right? That's also perfectly valid. Ah, lots of companies to design their Dr. Environment to not be one hundred percent right to the save. Some cost there um, and deliver good enough, but not identical performance to in a in a failure scenario.

112
00:22:26.330 --> 00:22:56.310
Harvey Green: Right? Yeah. And I think that's important to highlight, too, is that, you know, before with Gpu it was all it was always you tried to try to have like for like, and then you crossed your fingers and stood on your head and hope that everything worked the way that was supposed to. Uh. Now you, bill over, you know, from a gpu enabled on a tennis cluster to one that doesn't have gpus, so you don't have to incur that additional costs like Jared was just talking about

113
00:22:57.090 --> 00:23:18.959
Harvey Green: um and and have that additional complexity in your Vr environment. If you don't need it right. There will be plenty of cases where you do still need it. Um like in Ben's case, you know, talking through a medical facility, and potentially, having, you know doctors and nurses need that kind of rich experience for being able to look at. You know scans x-rays things like that.

114
00:23:18.970 --> 00:23:32.889
Harvey Green: You probably want them to be able to still do that, even if you're in a in a bail-over standpoint. But if that is you know, just for Gyra playing, you know. Call of duty. He's just gonna have to wait.

115
00:23:35.500 --> 00:23:38.290
Jirah Cox: That's That's an important. That's the critical use case for me.

116
00:23:38.300 --> 00:23:43.790
Ben Rogers: It's critical to you, but it's not critical to the business.

117
00:23:58.300 --> 00:24:07.479
Jirah Cox: I was going to different directions. They act Division will consider it very important. I

118
00:24:10.470 --> 00:24:28.150
Harvey Green: all right. Um, and kind of the close of the article. Here they they lay out. Ah, what testing systems they needed. Um! How they tested what they tested, what they used. Um, the big thing that I I guess I want to bring out here, is there? They're using

119
00:24:28.160 --> 00:24:42.029
Harvey Green: the mechanics Inx three thousand series, and that's when Jarv remind me, I believe right now. The three thousand series is the one you want to go for. If you're using Gpu,

120
00:24:42.060 --> 00:25:01.129
Jirah Cox: you guys want to hear with the G in the model name. This one is the thirty one point five g. It's the nx branded node that supports the thermals for the gpu. It's a twou form factor single node, so high airflow, high power budget which you leave both of those for one.

121
00:25:01.700 --> 00:25:02.830
Harvey Green: Got it

122
00:25:03.490 --> 00:25:14.250
Harvey Green: all right. Let's see. I think I you know I think we made it here any closing thoughts, Mr. Rogers,

123
00:25:14.910 --> 00:25:37.960
Ben Rogers: you know, if I ever had to to go back into the seat where I was managing this,

124
00:25:37.970 --> 00:25:41.390
Ben Rogers: he said it, because that's a whole separate conversation with that

125
00:25:41.400 --> 00:25:47.489
Ben Rogers: to be mentioned. But you know, when you go into this licensing is one of the things you got to definitely pay attention to.

126
00:25:47.500 --> 00:25:48.690
Ben Rogers: I don't know. I think

127
00:25:48.700 --> 00:26:02.880
Ben Rogers: I think it's great. That man we're getting this technology condensed to the point that we can have sixty four users, you know, on a single gpu. That's that's incredible for me at this point, Then I would want to start seeing this in real time.

128
00:26:02.990 --> 00:26:21.999
Ben Rogers: I think this is gonna become a

129
00:26:22.010 --> 00:26:26.299
Ben Rogers: You know something it'll be. It'll be that to have instead. I wish I would have

130
00:26:26.630 --> 00:26:27.700
Harvey Green: right.

131
00:26:27.820 --> 00:26:45.999
Harvey Green: Yeah, let me throw a question out there to you, Ben, as you think about that, and put that hat on, and you look at the possibility of being able to support one hundred and twenty eight users out of two gpus and one node.

132
00:26:46.020 --> 00:26:59.340
Ben Rogers: That something you would do Well, the numbers wouldn't take long

133
00:26:59.350 --> 00:27:07.990
Ben Rogers: five nodes. I could. I could possibly be getting to that number, you know. So that's that's pretty impressive. And one thing that I didn't mention is that

134
00:27:08.000 --> 00:27:29.159
Ben Rogers: Gpus make it difference just beyond the taxing apps like Emr Radiology or Tad R. E. Mr. There were strange, ah, especially like in the drug allergy screens where they were doing a lot of comparisons. Gpus, make a difference in how fast the application naturally run. So you know, people don't like to hit click and see the spinning wheel,

135
00:27:29.280 --> 00:27:35.679
Ben Rogers: I can get it in their face. This does address that in certain courses of other applications like an

136
00:27:35.890 --> 00:27:42.060
Harvey Green: well, you know, I said, closing thoughts. And now i'm gonna open it back up

137
00:27:42.420 --> 00:27:59.220
Harvey Green: only because you you know a point that I think is important, too, that you know most people before would look at Gpu and say, Well, only that's that's only for the people. I want to have this very great rich, user experience that are using these certain programs and things like that.

138
00:27:59.230 --> 00:28:21.720
Harvey Green: And now that we talk more about getting that out to the other users, and we talk about the performance of the gpu. One of the things that we didn't hit on at all is how much of a difference it makes to your Cpu to have those things offloaded and have your gpu take care of that, so I i'll pick on you. Can you speak to that a little bit, sir?

139
00:28:22.330 --> 00:28:40.049
Jirah Cox: Not in more detail. But you're right, which lets you

140
00:28:40.590 --> 00:28:43.670
Jirah Cox: strain. Get higher density, remote workloads.

141
00:28:43.680 --> 00:28:53.490
Harvey Green: Yes, yes, so so in a nutshell. There we're saying, if you add Gpus, you're going to see better performance. Even if

142
00:28:53.670 --> 00:28:58.190
Harvey Green: a user experience, they're only just seeing things look prettier on their screen.

143
00:28:59.810 --> 00:29:00.789
Jirah Cox: What's it?

144
00:29:02.390 --> 00:29:20.459
Ben Rogers: It calls to these things, You know. I was looking at that orbiter today. Ah was sixty, four years old on these boards. It was a lot to go on the boardroom and go. Hey? I didn't spend several thousand dollars to take care of it.

145
00:29:20.470 --> 00:29:34.389
Ben Rogers: That was hard. That was a hard argument to get across,

146
00:29:34.400 --> 00:29:38.390
Ben Rogers: you know. Evaluate whether you need to come out of your pajamas and get into your surgical.

147
00:29:38.400 --> 00:29:40.189
Ben Rogers: That was a compelling argument.

148
00:29:40.200 --> 00:29:53.840
Ben Rogers: But now, being able to pay sixty, four, divided by six seven grand, the numbers become very, very a different different conversation, and I don't think it would be as much of an argument as it was three or four years ago.

149
00:29:53.850 --> 00:30:13.689
Harvey Green: Yeah, absolutely. Um, that's you know. It's kind of akin to uh a Tv that does twice the resolution and weighs half as much as that. That big brick you got sitting on your desk. So yeah, they're They're definitely making strides here, and and I love to see it.

150
00:30:14.130 --> 00:30:24.200
Jirah Cox: Gyra. Any closing thoughts there. I mean. Some people need to hear things several times until it really sinks in Harvey. What episode? And Bro. You say that this is,

151
00:30:44.000 --> 00:30:54.690
Ben Rogers: I am fully committed to it until I hear differently. I had to meet myself because I was so funny.

152
00:30:55.400 --> 00:31:04.059
Harvey Green: Yeah, we'll We'll probably have Mr. Andy White sign back on the next one and we'll, We'll have him chew me out on the air if it was not fifty-nine

153
00:31:04.480 --> 00:31:06.720
Harvey Green: so

154
00:31:07.800 --> 00:31:10.190
Harvey Green: all right. I'll stay too.

155
00:31:10.200 --> 00:31:18.490
Jirah Cox: What listeners stay tuned. We got nice same bad time saying that, you know

156
00:31:18.500 --> 00:31:25.789
Harvey Green: I call it her out. Thank you guys, for jumping on with me and discussing this, and we will see you for the next one.

157
00:31:25.920 --> 00:31:27.290
Jirah Cox: I'll see you all. Thanks. Everybody.

158
00:31:27.300 --> 00:31:29.439
Ben Rogers: Thanks. Appreciate it. Yeah,