-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 09/07/2017 01:24 PM, ron minnich wrote:
On Thu, Sep 7, 2017 at 11:03 AM Timothy Pearson <tpearson@raptorengineering.com mailto:tpearson@raptorengineering.com> wrote:
could anyone shed some light on these decision making processes? An open ISA and core design does not guarantee open silicon, and in fact one could argue that it will mean any performance improvements end up highly locked under NDA and similar to avoid competitors coming online and ruining tens of millions of dollars of investment for even one SoC improvement.Exactly. The open ISA can go both ways: pushing toward "value adds" that lock up a platform, instead of going the way we might hope, so vendors have competitive advantage. Look at page 47 of "Volume II: RISC-V Privileged Architectures V1.10" -- it basically allows a vendor to recreate SMM as it exists today, creating regions of memory irrevocably hidden from kernel. And there are certain things you can't access on riscv without an M-mode trap, which means that you can't escape the need for code in M mode.
RISCV vendors can create SMM. We need to encourage creation of a world in which they do not. But just claiming that "riscv is open so there won't be anything proprietary" is being a bit unrealistic in my view.
So what about Power? The problem I keep hearing is that power competes in the server space with x86, and not well enough. It's one of too slow, too power hungry, or too expensive right now to compete well. I am hoping that Raptor is going to show us power done right :-)
Appreciate the vote of confidence! :-)
I don't think we have to worry about POWER's speed or TDP as of POWER9. The low-end 4-core chips (16 threads, comparable with low end AMD server chips) only have a 90W TDP which is significantly less than their AMD counterpart, and speeds at isoclock are projected to increase by 1.5x-2x per-core versus POWER8. Cost remains an issue but if we see more adoption overall costs will obviously come down. It really comes down to getting people to look at POWER as something other than the power-hungry beast POWER8 was, and that's proving somewhat challenging.
- -- Timothy Pearson Raptor Engineering +1 (415) 727-8645 (direct line) +1 (512) 690-0200 (switchboard) https://www.raptorengineering.com
Timothy Pearson wrote:
The fabulous thing about RISC-V is what makes ARM successful; there can and will be multiple different silicon vendors, offering products with many different features and tradeoffs.
Some can be top performance but proprietary. Some can be transparent/open but slower.
We already have this (think cheap ARM SoCs vs. a Xeon).
Which one of ARM and Xeon do you consider transparent/open?
And which one to the degree possible with a RISC-V core?
I'm honestly surprised at the overall community betting on a long shot (RISC-V) vs. using what's available and open right now (POWER9); could anyone shed some light on these decision making processes?
Are POWER9 systems available off the shelf right now? That's the availability metric for a decision making process.
Until systems are in stock, everything is pure speculation. Preorder is really an edge case. Sorry. :\ But the good news is, I don't think anyone is mentally excluding having *more* open options - quite the opposite!
POWER and RISC-V may indeed have comparable openness, which I doubt, but if so then RISC-V certainly has the better PR/marketing.
An open ISA and core design does not guarantee open silicon
I already commented on that, no need to repeat.
ron minnich wrote:
The open ISA can go both ways
Only an open ISA can go both ways *simultaneously* - I think that many are excited about the possibility that RISC-V may, rightly so.
RISCV vendors can create SMM.
And one vendor can choose not to. And there is a market for both.
But just claiming that "riscv is open so there won't be anything proprietary" is being a bit unrealistic in my view.
For the third time (this is now getting very annoying) - noone has claimed that.
Timothy Pearson wrote:
It really comes down to getting people to look at POWER as something other than the power-hungry beast POWER8 was, and that's proving somewhat challenging.
There is no significant POWER visibility. I know one single name at one single company immediately involved with an actual potentially orderable POWER product. (You!)
I have never seen a running POWER8/9 system. I've seen two running RISC-V systems and have ordered a third.
I am able to implement (a weak variant of) RISC-V on <100¤ hardware. Today. Actually, already over a year ago!
So; POWER8/9 either do not really exist in a relevant way at the moment, or they are just not reaching me, even though I have an interest in open hardware.
I can't tell which is more accurate. And maybe there's yet another option?
//Peter
On Thu, Sep 7, 2017 at 3:37 PM Peter Stuge peter@stuge.se wrote:
For the third time (this is now getting very annoying) - noone has claimed that.
Sorry, I was not specific. No one has claimed it recenly on this list, I suppose. But I hear it all the time "not on this list" :-)
Peter, I'm glad to see you involved in RISCV, it helps ensure the right things happen ...
ron
-
Peter Stuge -
ron minnich -
Timothy Pearson