I rarely comment on these but I gotta say - that outro was fantastic. I exhaled nasally out loud.
Great episode. WouldBuyAMDStock/10
I made a remix out of that outro...
EDIT: New version with more knock knock: Clyp link
Clyp.it link to the old version of the remix
4 Likes
needs the knock knock pop pop tho? lol
1 Like
Actually they are in there but i reverb'd them maybe a bit too hard..
So yeah it may need more of that.
1 Like
And now I have a new ringtone
edit: I seriously can't stop smiling.
1 Like
xD check out the updated version too with more "knock knock"
1 Like
Seeing as Seagate had a SSD with a 10GB/s (yes, gigabyte with a capital B) PCIe x16 SSD at CES, I don't think it's anywhere close to right sized.
http://www.seagate.com/enterprise-storage/solid-state-drives/nytro-xp7200-nvme-add-in-card/
http://www.anandtech.com/show/10555/seagate-introduces-10gbs-pcie-ssd-and-60tb-sas-ssd
Ryzen has a total of 24 PCIe 3 lanes. That's 16 for GPU, 4 for NVMe and 4 for the chipset. You can have a NVMe 4x drive, that connects to the CPU, plus SATA drives, that connect to the chipset.
The SATA drives won't be bottlenecked by the 4x connection between the chipset and CPU, because nobody uses 10 Gb/s ethernet.
http://wccftech.com/amd-confirms-ryzen-cpus-unlocked-overclocking/
24 PCIe Gen 3 lanes (from the CPU)
6 SATA3 6Gb/s ports (2 from the CPU and 4 from the chipset)
As long as you have 4 SATA drives or fewer, you can use a 4x NVMe drive
2 Likes
Since something like that would be ridiculously expensive, I don't see why anybody would put that in a consumer motherboard.
That high end Asrock board looks like it has 5 Gbit Ethernet, just like the similar new Intel board they have. Ryzen also has USB 3.1 directly on the CPU, so no need to use the chipset for that either if you don't want to. It's 3.1 gen 1 and four ports, for common peripherals it is plenty. I could easily only use the CPU USB, just plug one in to my 10-port hub and the other three can be free for some high speed external storage or similar. 3.1 gen 1 is 5 Gbit after all.
I think it looks good for a consumer platform, much better than AMD platforms has looked in a long, long time.
Because it's not about that one product, it's that it is what we can expect to see on the market in general. And I doubt that it's going to be expensive for long. Storage technology is moving at a blistering pace.
Yeah, x16 drives sounds cool and all, but it isn't something for us mortals. And it won't be until we get much better machines (and drivers, polling instead of interrupts probably). It will be several years, AMD will have moved on from AM4 then as they talk about a four year lifespan.
So, why won't x16 drives be in the general market any time soon? One PCIe 3.0 is almost 1 GB/s throughput (in both directions, dual simplex). Drives that can actually use 16 lanes will be very, very expensive when they become available (a Nvidia Titan will probably be significantly cheaper). An x8 drive today goes for about $3000 today? And those use about half the available bandwidth. That x16 Seagate drive does not saturate x16 PCIe, and is very much an enterprise product. Not something that would be marketed towards ordinary consumers or even enthusiasts. And frankly I don't think anyone who doesn't run typical server loads should buy one (assert mode etc). We would need much betters CPUs even to be able to use that bandwidth. We see CPU bottlenecks with current x4-x8 drives, a x16 drive that actually uses the 16 lanes of bandwidth would mostly be idle in even a high end workstation.
(Remember that the perf number they cite are at typical server workloads. Not something that will be readily apparent on a desktop where Queue Depths will be more in the order of 1-2 instead of the perf number optimal 32. A Samsung "gumstick" M.2 NVMe will probably give you the same noticeable perf on a desktop.)
Later on in the year AMD will launch their server platform. That will probably be more suited to enterprise drives.
Would it be possible, that AMD has a "X390" chipset in the works, going balls on the wall in connectivity?
About the right size in the future argument... Ehm, Wendell, as you said, most people are using spinning rust. I replaced the failing hard drive for a friend of mine with a cheap ~120GB SATA SSD (not a file hoarder, he only plays League of Legends, so it is ample storage for him), and boy, wasn't it fast! For the "unwashed masses" even a good SATA SSD is enough. And not all professionals, especially those running a small business, can afford a high end workstation. I currently do my thesis for my degree in mechanical engineering in a Core 2 Duo with 4GB of RAM, an AMD HD2600 XT and a 7200 RPM mechanical drive for OS/applications, and a 5400 RPM drive for storage. It sucks in AutoCAD, but I currently can't afford a proper workstation, and neither do most professionals where I live (Greece), at least without some form of subsidies.
"640K ought to be enough for anybody."
Dude you're wrong. In 3 years, SSDs have gone from exotic to commonplace. And the throughputs have more than quintupled. Blah blah server. Blah blah enterprise. Bullshit. It's coming for the desktop. It's going to be cheap. And it's coming sooner than you think.
For anybody not, but a lot of people would be happy even with that for some years after it wasn't considered high end. In an age where upgrades were much more meaningful BTW.
Atari ST and Amiga came out with 512kB of RAM, if I recall correctly, and had GUIs too, not command prompts.
Hi Wendell!
I'm an actuary, and I have an edit for this show :)
I think that there was a misunderstanding about the role of an actuary based on the conversation that took place on the show.
Actuaries do many things, but mostly we price, reserve for and invest proceeds from the insurance business. We derive the statistical models to simulate the economy and future claims (be they for options on investments, mortality related, or health related).
The article about automating claims processing relates to staff who read incoming claim data and enter it into systems. They can decide whether there exists sufficient evidence to pay the claim. Actuaries don't usually have anything to do with this process; it's usually "lower level" employees who read hundreds of claim files a day and spend the vast majority of that time on data entry.
That said, actuaries are beginning to make use of AI techniques; we term them "predictive modelling".
Instead of hypothesis testing where I will test if x has a relationship to y, we can just use supervised learning to find those relationships for us. Though, given the opaque nature of deep learning, it's of limited use to us. The output from these methods usually needs to be vetted since they tend to produce a lot of garbage. As my first professor on the subject once said "the difference between a good actuary and a bad actuary is common sense".
The hardware for supporting AI is what we really need, doing stochastic on stochastic projections of economic models requires thousands of cores to achieve in realistic time frames. At the moment we use server farms of CPUs, but GPGPU is offering us a method of scaling this workflow (where we're not doing large operations we're just doing a lot of them) .
1 Like
Even NVMe 4x is too expensive for most people.
If you are expecting, that 16x will be affordable before Ryzen's successor is launched, then you are delusional.
That's true for SATA SSDs. NVMe is still too expensive.
SATA was an already existing and widely used standard long before SSDs came along.
NVMe is relatively new and most PCs that are currently in use just don't have an NVMe slot. So right now NVMe is still exotic. Give it a few more years and it'll become quite common.
As for the PCIe x16 drives, I'm not sure if those are going to take off. PCIe storage has been tried for quite a while now and it never really became a thing. Maybe once we start saturating PCIe 3.0 x4 there may be no other way to go forward, but I'm not too sure about that really.
I was talking about the price of drives. But you are right, motherboards with NVMe are expensive and rare.