[coreboot] RISC-V HiFive Unleashed board added to coreboot - has PCI-e slots via exp board

Timothy Pearson tpearson at raptorengineering.com
Mon Jun 25 09:44:39 CEST 2018


On 06/25/2018 02:30 AM, Timothy Pearson wrote:
> On 06/24/2018 08:06 PM, Nico Huber wrote:
>> On 25.06.2018 01:55, Timothy Pearson wrote:
>>> On 06/24/2018 06:41 PM, Timothy Pearson wrote:
>>>> On 06/24/2018 06:35 PM, Nico Huber wrote:
>>>>> On 24.06.2018 23:52, Timothy Pearson wrote:
>>>>>> On 06/24/2018 03:43 PM, Nico Huber wrote:
>>>>>>> On 24.06.2018 21:37, Taiidan at gmx.com wrote:
>>>>>>>> On 06/24/2018 02:59 PM, ron minnich wrote:
>>>>>>>>> On Sun, Jun 24, 2018 at 11:47 AM Jonathan Neuschäfer <j.neuschaefer at gmx.net>
>>>>>>>>> wrote:
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> "While we’d love to provide you with this information, we believe we
>>>>>>>>>> cannot. However, we can’t prevent anyone from disassembling the fsbl and
>>>>>>>>>> copying the values sent to the blackbox DDR register map."
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> and ... there ends my interest in the hifive. A shame.
>>>>>>>>>
>>>>>>>>
>>>>>>>> I can't understand what their target audience is? who would buy such a
>>>>>>>> thing? who do they intend to sell these to? I mean the open source
>>>>>>>> people can buy the now very affordable Talos 2L and the cheap-soc people
>>>>>>>> can buy one of the many of ARM boards that litter the marketplace...I
>>>>>>>> don't get it.
>>>>>>>
>>>>>>> I don't think you can compare the HiFive Unleashed with the Talos. They
>>>>>>> really target completely different people and use cases. You could as
>>>>>>> well ask, why produce smart watches, when people can afford the Talos?
>>>>>>>
>>>>>>> Talos is a workstation it doesn't fit anywhere but a workplace where
>>>>>>> somebody else pays the power bill. So you can't even compare it to
>>>>>>> cheap ARM SBCs, HiFive aside. It's a professional product, nothing to
>>>>>>> play with, but something to work with. And it's open. It is marketed
>>>>>>> as open. It is designed to be open. It is based on an open platform.
>>>>>>
>>>>>> I just want to counter this one point.  POWER9 is absolutely not power
>>>>>> hungry.  I've seen the 8-core chips idle at under 10W, with active loads
>>>>>> maybe in the 40-60W range.  We're dogfooding one machine in a typical
>>>>>> office setting, and it dissipates nearly no heat -- it's using less
>>>>>> power than the older Xeon it replaced.
>>>>>
>>>>> Hmmm, yeah, just twist my words as you wish. I never said that it is
>>>>> power hungry compared to other workstation systems. I did not even state
>>>>> that it is power hungry at all. All I said is that it needs power and
>>>>> somebody has to pay for that too.
>>>>>
>>>>> Now you show off with random numbers that make things really weird. 10W
>>>>> for what? per chip? or per core? Whatever it is, I hope your office has
>>>>> air conditioning. And than that "it's using less power than the older
>>>>> Xeon", omg really? you're system is better than shit?
>>>>>
>>>>> Nico
>>>>
>>>>
>>>> Did not mean to offend here.  Apparently we have very different ideas of
>>>> "workstation" versus "desktop"; we'd classify some dozens of watts under
>>>> real world load per CPU as a desktop, not as a workstation per se.  I
>>>> don't see how something using this little power would suddenly put the
>>>> power bills out of reach for individual use vs. corporate use, but again
>>>> we may have very different ideas of what a computer should be.
>>>>
>>>> Personally, I would never be able to use something like a Raspberry Pi
>>>> or other low power SBC for anything other than maybe some minimal text
>>>> editing.  It's not worth my time to put up with a slow, unresponsive
>>>> system; whatever I would gain on power bills would be lost through
>>>> unproductive time and then some.
>>>>
>>>> I don't see how providing some real world numbers can be frowned upon here?
>>>>
>>>
>>>
>>> So it was pointed out to me on IRC that we don't have current power
>>> numbers published on the Wiki.  That would probably explain the
>>> confusion; subsequent firmware updates have dropped the idle power and
>>> "normal use" power significantly on the POWER9 chips.
>>
>> Sorry, I really still had the wiki numbers in mind (> 100W idle). Which
>> would be notable on a power bill. And then, 10W per 8 core CPU really
>> confused me (with 30W for the 4 core CPU in mind I really anticipated
>> you mean 10W per core). So it can compete with today's average desktop
>> system wrt. idle power. This makes the Talos II even more attractive now
>> (and I'm thinking about the space below my desk again).
>>
>>>
>>> I'll see if we can get some new at-wall measurements of our normal
>>> desktop configuration (1 CPU, no SAS, NVMe storage).  This should be
>>> well under 100W at the wall.
>>
>> I'm really happily surprised by the new numbers. If you keep the max.
>> below 160W, you could sell it with a Pico-PSU ;)
> 
> Just got done doing some initial checks on a Lite system (4 cores, 2x
> 16GB ECC registered DIMMs, 1TB NVMe drive).  Actually having some
> measurement issues due to how low the DC side load is -- the board only
> "wants" 50 watts or so at idle, but the Seasonic 1200W Platinum supply
> being used has terrible efficiency in that range, resulting in at-wall
> measurements of around 70W at idle.  Under full load it only pulls 120W
> from the wall.
> 
> If you have another, more efficient PSU in mind for these light loads
> I'm all ears. :-)  The Pico-PSU might even work here....

Follow up to the follow up....there's an oddity I need to track down
yet, in that the "master" CPU pulls ~28 watts at idle while the "slave"
CPU (if installed) draws 10 watts at idle.  There might be some part of
the nest active that was involved in bootup that doesn't need to be
active after OS start, or something similar.  I'd expect that the
primary CPU power draw (and, by extension, single CPU power draw) still
has some ways to go with future firmware updates and extensions...

-- 
Timothy Pearson
Raptor Engineering
+1 (415) 727-8645 (direct line)
+1 (512) 690-0200 (switchboard)
https://www.raptorengineering.com



More information about the coreboot mailing list