The hook for this over other ones is that this is designed to have no binary blobs required, completely open and auditable firmware, etc. It's for people that care about security and are disgusted by the trash-fire that is the lowest levels of our computing stack.
The lesser expensive servers do not use the same firmware as Talos™ or even the more expensive IBM systems. The cheaper machines reduce cost by using proprietary firmware components wherever possible, and by dropping as many features as the market will bear at the lower price point. This is not the market segment that Talos™ is aimed at.
I believe their point is that the segment this is aimed at might not exist. I suppose that's why it's a campaign :) I'd also be interested for $1-2k even, but not at new car prices.
I was pretty shocked that they're asking $18k for a workstation and hoping to get $3.7 million total. I was rather excited for it but this project is stillborn in my opinion. I may end up having work order one of IBM's basic dual-Power8 systems, which are only $6k: http://www-03.ibm.com/systems/power/hardware/s821lc/index.ht...
I think it's odd that they offer no cheaper workstation configuration - that $18k price includes maxed-out ram, a higher-end workstation class graphics card (32GB ram on the card! much cheaper workstation cards exist), and an LSI SAS controller card instead of using built-in sata3. A more minimal configuration could probably shave $10k off the price (but maybe they thought it wouldn't make sense to pay $8k for a minimal config).
As mentioned on a different comment, we are investigating adding a lower-spec machine to the lineup due to popular demand. We had not anticipated the level of demand for a pre-built workstation with a less powerful configuration.
Most likely, this new configuration would scale the RAM down and replace the high-end workstation GPU with a mid-range consumer card. More information will be posted to the campaign page when we have this option available.
Kind of what workstations used to cost, took me back to the 80's and my time at Sun.
The key here is all open. From mass storage to memory to boot you can "own" the entire system. There is no way to build that at any where close to this price with existing x86 machines. Just negotiating the NDA alone to have access to the TCM area of the chip would probably be $50K in lawyers fees.
That said, I had the same thought at you, $18K ? Really? Who are they kidding? And then I thought about my PC purchases over the last 12 years. I've spent about that much (granted I did so by upgrading and replacing systems every 2 - 3 years) But my last system is now 4 years old and frankly not really in that much need of replacing.
Personally I think IBM should fund this and order 200 work stations for one of its engineering teams. Get their feedback on it. Would be an interesting experiment for them.
I am considering pledging $250, to support this effort within my means, and to be able to try things out with a VM. But I see no details on how much RAM, computing power, and storage space will be provided.
Edit: If the VM provided is anemically underpowered, the platform will be little more than a curiosity for me.
We have not worked out the exact specifications of the virtual machines at this time; we need to balance usability against overall cost and that depends to some extent on the number of signups we receive.
Initial targets were a dual core VM with 2GB RAM and 10GB storage. Would you consider this underpowered or acceptable?
The cost difference between SAS and SATA isn't all that much, and SAS has some nice benefits especially when used in a RAID type configuration.
We see this first run as targeting more niche and corporate / government type use cases, however (should funding be met) subsequent runs after this first one may be targeted more at a mainstream audience. It's far too early to tell at this stage.
Given the strong and consistent feedback on this we are looking at adding a more minimal pre-assembled option. Most likely this would involve scaling the RAM count down and dropping the GPU to a mid-range consumer card. More information will be posted to the campaign page when we have this option available.
This is a pretty redundant comment, but I echo that sentiment! Slightly more modest price tag and configuration and I'd snap one of these up in a heartbeat.
Sounds really cool, Timothy. Thanks for the prompt response. 256 gigs sounds a bit excessive for a workstation unless you are dealing with huge local datasets.
These are expensive, but the power 8 systems are really fantastic. I used to work extensively with power servers when they first started offering Linux support, and they just have fantastic performance characteristics under heavy load.
> they just have fantastic performance characteristics under heavy load
That sounds interesting, but could you elaborate? The comparisons to comparable Intel systems I've seen did not seem to suggest anything like that, it was more like sometimes Power wins, sometimes x86 does and both are quite similar performance wise. What kind of tasks did you observe to perform well?
IBM POWER looks to be the most performant libre computing platform in existence, but $4100 for Raptor Engineering's Talos is still too pricey for me -- I'd suggest US$1000 to US$1500 for plain ATX motherboard + CPU, supporting affordable standard commercial consumer components is probably the point where crowd funding is viable as a desktop computer for regular software engineers and tinkerers (well the kind of people who seriously consider buying a Novena).
But Talos is expensive as it uses a custom board, with custom open-source firmware at relatively low volume, so a more affordable Talos desktop is probably infeasible.
I guess for that it needs to be a desktop (not a workstation) -- so a much more stripped down system dropping expensive components such as the FPGAs that a secure internet connected server, or a typical software engineer's desktop don't need.
I really wonder what sort of work you could do on this - most platforms don't have Power8 support do they? So likely no frontend, obviously no mobile, most tooling probably isn't available either...
The less hip Linux distributions have very good support for the less popular CPU architectures. For example, Debian Jessie has an official ppc64el (ie, POWER8) release as well as mips and other platforms with low mindshare [1].
In my experience with Debian armhf of a ARM Chromebook (via crouton), all the packages repositories all work out of the box and I haven't noticed any issues. The only exception I've had has been (ironically) the Chromium browser -- but that application is famously complicated to build.
The past few years I've been pleasantly surprised smoothly developing on machines using different CPU architectures is with open-source software (thanks to good automation (build bots), robust unit testing, and hard work by Debian maintainers and their upstream developers)
Compared to the closed-source world of applications (application compatible between Windows and Windows RT, or Apple's transition from PowerPC Macs to x86 Macs), being CPU architecture agnostic using purely open-source software requires no effort at all. But being architecture agnostic using closed source applications of course suffer from the same issues as the other platforms.
This machine is way out of my price range; but I am very interested in the frequently mentioned "open-toolchain FPGAs" in their descriptions.
AFAIK there are only a handful of very specific chips that you can currently get away with not using vendor tools against, like ICE40s.
I would be willing to put a fair amount of money into crowdfunding for better open source -full- toolchains for FPGAs (especially something targeting Xilinx devices). The vendor tools, aside from being closed source, are all huge bloated disasterpieces, and none available for macOS.
Yeah and unless the FPGA manufacturers start supporting open toolchains, it can end in an instant when the parts become EOL and Lattice (maker of the ICE40 series) starts obfuscating.
The problem with open toolchains from the manufacturer's point of view is that they have nothing to gain and everything to lose. They don't want to sell chips one at a time to hobbyists. They want to sell them millions at the time to GM and other companies. In their mind, open toolchains buy them nothing but leave them open to competitors stealing their designs. I'm not saying this view is accurate, it is just how chip vendors think.
If you want Power8 stuff to come down in price to something that is reasonably affordable (even only double the cost of a current generation core i7 and $190 motherboard), what's needed is for the top-ten Taiwanese motherboard manufacturers to make standard ATX format boards for it. Period. Tyan, ASUS, Supermicro, MSI, ECS, etc.
Some of the Taiwanese manufacturers do make OpenPOWER machines. However, the pricing is in the same ballpark as Talos™ and you don't get anywhere near the complete owner control that Talos™ offers. Look into the closed-source BMC on those machines, and also the built-in peripherals that require third-party closed firmware and, in some cases, closed drivers to run.
I want one of these badly, but sadly the price is not for me. I hope they find a big enough market, but I'm not really convinced.
$4,100 for the motherboard, add $1,135 for the slowest CPU and your minimum buy-in is $5,235 or $7,450 for the top-end 12-core (we can assume that RAM, GPU, Storage costs are the same as any other system)
A less-expensive configuration has become available. For $8,750 you get an 8-core CPU, 128GB of RAM, a Radeon RX 480 GPU, and 2 x 2TB hard drives.
I really want to see this project succeed. I am leaning toward buying just the motherboard and CPU. I prefer SSDs, and don't need massive storage anyway.
39 comments
[ 4.1 ms ] story [ 89.2 ms ] threadhttps://secure.raptorengineering.com/TALOS/prerelease.php
And on a practical level we know that people who bleat about firmware blobs are not going to spend $6,000 to escape from them.
The lesser expensive servers do not use the same firmware as Talos™ or even the more expensive IBM systems. The cheaper machines reduce cost by using proprietary firmware components wherever possible, and by dropping as many features as the market will bear at the lower price point. This is not the market segment that Talos™ is aimed at.
That presentation offers several potential explainations for why that particular market segment is locked to proprietary, non-libre offerings only.
[0] https://www.crowdsupply.com/raptor-computing-systems/power8-...
As mentioned on a different comment, we are investigating adding a lower-spec machine to the lineup due to popular demand. We had not anticipated the level of demand for a pre-built workstation with a less powerful configuration.
Most likely, this new configuration would scale the RAM down and replace the high-end workstation GPU with a mid-range consumer card. More information will be posted to the campaign page when we have this option available.
The key here is all open. From mass storage to memory to boot you can "own" the entire system. There is no way to build that at any where close to this price with existing x86 machines. Just negotiating the NDA alone to have access to the TCM area of the chip would probably be $50K in lawyers fees.
That said, I had the same thought at you, $18K ? Really? Who are they kidding? And then I thought about my PC purchases over the last 12 years. I've spent about that much (granted I did so by upgrading and replacing systems every 2 - 3 years) But my last system is now 4 years old and frankly not really in that much need of replacing.
Personally I think IBM should fund this and order 200 work stations for one of its engineering teams. Get their feedback on it. Would be an interesting experiment for them.
Edit: If the VM provided is anemically underpowered, the platform will be little more than a curiosity for me.
Initial targets were a dual core VM with 2GB RAM and 10GB storage. Would you consider this underpowered or acceptable?
Edit: In terms of CPU and RAM, this is comparable (in the sense they can be compared) to a $20/month VM at Digital Ocean.
As far as strorage, have you considered using standard drives in addition to enterprise SAS drives?
Are you trying to go mainstream with the platform as quickly as possible, or just trying to carve out a niche for now?
We see this first run as targeting more niche and corporate / government type use cases, however (should funding be met) subsequent runs after this first one may be targeted more at a mainstream audience. It's far too early to tell at this stage.
Given the strong and consistent feedback on this we are looking at adding a more minimal pre-assembled option. Most likely this would involve scaling the RAM count down and dropping the GPU to a mid-range consumer card. More information will be posted to the campaign page when we have this option available.
https://www.crowdsupply.com/raptor-computing-systems/power8-...
The price range is $1,135-$3,350.
Though I'd choose the mainboard only + CPU and add an old, passive GPU for price reasons
If you can't afford the mainboard or a full system, please pledge some money for user freedom.
That sounds interesting, but could you elaborate? The comparisons to comparable Intel systems I've seen did not seem to suggest anything like that, it was more like sometimes Power wins, sometimes x86 does and both are quite similar performance wise. What kind of tasks did you observe to perform well?
But Talos is expensive as it uses a custom board, with custom open-source firmware at relatively low volume, so a more affordable Talos desktop is probably infeasible.
I guess for that it needs to be a desktop (not a workstation) -- so a much more stripped down system dropping expensive components such as the FPGAs that a secure internet connected server, or a typical software engineer's desktop don't need.
In other IBM POWER news, Google has recently announced it's Open Compute Project POWER9 server -- the Zaius P9 Server: https://news.ycombinator.com/item?id=12709995
In my experience with Debian armhf of a ARM Chromebook (via crouton), all the packages repositories all work out of the box and I haven't noticed any issues. The only exception I've had has been (ironically) the Chromium browser -- but that application is famously complicated to build.
The past few years I've been pleasantly surprised smoothly developing on machines using different CPU architectures is with open-source software (thanks to good automation (build bots), robust unit testing, and hard work by Debian maintainers and their upstream developers)
Compared to the closed-source world of applications (application compatible between Windows and Windows RT, or Apple's transition from PowerPC Macs to x86 Macs), being CPU architecture agnostic using purely open-source software requires no effort at all. But being architecture agnostic using closed source applications of course suffer from the same issues as the other platforms.
[1] https://www.debian.org/ports/
AFAIK there are only a handful of very specific chips that you can currently get away with not using vendor tools against, like ICE40s.
I would be willing to put a fair amount of money into crowdfunding for better open source -full- toolchains for FPGAs (especially something targeting Xilinx devices). The vendor tools, aside from being closed source, are all huge bloated disasterpieces, and none available for macOS.
The problem with open toolchains from the manufacturer's point of view is that they have nothing to gain and everything to lose. They don't want to sell chips one at a time to hobbyists. They want to sell them millions at the time to GM and other companies. In their mind, open toolchains buy them nothing but leave them open to competitors stealing their designs. I'm not saying this view is accurate, it is just how chip vendors think.
Interestingly, some Asus and GIGABYTE motherboards for AMD and/or Intel processors support Libreboot:
https://libreboot.org/docs/hcl/
So in theory, if this campaign succeeds, it may catch their attention, and we may have more affordable options down the road.
That's why we need companies like Raptor Engineering! Vote with your money...
$4,100 for the motherboard, add $1,135 for the slowest CPU and your minimum buy-in is $5,235 or $7,450 for the top-end 12-core (we can assume that RAM, GPU, Storage costs are the same as any other system)
I really want to see this project succeed. I am leaning toward buying just the motherboard and CPU. I prefer SSDs, and don't need massive storage anyway.