Just got a new nehalem box in for test yesterday. Experiences so far:
1. POST from power-on takes 45 seconds. *45 SECONDS*. Now, I had it said to me at SCALE7x last year from someone from Intel that all new BIOSes on Intel chips are really EFI underneath -- is this indicative of what we are to expect? If so, it's awful. It's 15 times slower than what we had ten years ago, and 50 times slower than what we can do today on coreboot. 2. POST from reset took infinity, sometimes, and 60 seconds others. 3. With no keyboard attached, we got the 'Hit any key to continue' message. I can't this problem still occurs. 4. No PS/2 connectors, which might be ok, except: 5. USB keyboard doesn't work well enough to get to the config screen. Oh, it works, just not right away. We can't get to the BIOS setup. 6. It's too dumb to realize that if there really are no storage devices attached, it really ought to try the network. And we can't force the 'boot from net' setting yet. See (5).
Experiences on a MAC with EFI 1. "How come it's so long to be responsive?" 2. The only way to make it bearable is to load REFIt. 3. there's a file on the "EFI FLASH partition". It defines the partition tables. Yep, you might think that it could just read the drive, and no, that's not good enough: it can't boot an installed linux unless you rebuild this file too.
So, the score so far: we've had a superior open source solution for 10 years, and there are lots of sectors starting to listen (I hear from more each week), but we're still stuck with Old Fashioned BIOS on our desktops and servers. :-)
I expect this problem to change this decade, just not sure when :-)
ron
Imho it's just a sad sign for our industry. There is a great, fast, cheap and very easy to implement solution for x86 bootfirmwares. coreboot, but somehow the industry doesn't seem interested. Instead every company invents the wheel over and over again and they never do it good or even right. One would think that coreboot could give a company a real advantage in the market, but still - nobody even tries (some exceptions in special market segments excluded).
I hope that changes soon, not only "this decade", and I hope that AMD stays on course and tries harder.
Anyway, you're doing a great job, keep up the good work!
On 2010-03-05, ron minnich rminnich@gmail.com wrote:
Just got a new nehalem box in for test yesterday. Experiences so far:
- POST from power-on takes 45 seconds. *45 SECONDS*. Now, I had it
said to me at SCALE7x last year from someone from Intel that all new BIOSes on Intel chips are really EFI underneath -- is this indicative of what we are to expect? If so, it's awful. It's 15 times slower than what we had ten years ago, and 50 times slower than what we can do today on coreboot. 2. POST from reset took infinity, sometimes, and 60 seconds others. 3. With no keyboard attached, we got the 'Hit any key to continue' message. I can't this problem still occurs. 4. No PS/2 connectors, which might be ok, except: 5. USB keyboard doesn't work well enough to get to the config screen. Oh, it works, just not right away. We can't get to the BIOS setup. 6. It's too dumb to realize that if there really are no storage devices attached, it really ought to try the network. And we can't force the 'boot from net' setting yet. See (5).
Experiences on a MAC with EFI
- "How come it's so long to be responsive?"
- The only way to make it bearable is to load REFIt.
- there's a file on the "EFI FLASH partition". It defines the
partition tables. Yep, you might think that it could just read the drive, and no, that's not good enough: it can't boot an installed linux unless you rebuild this file too.
So, the score so far: we've had a superior open source solution for 10 years, and there are lots of sectors starting to listen (I hear from more each week), but we're still stuck with Old Fashioned BIOS on our desktops and servers. :-)
I expect this problem to change this decade, just not sure when :-)
ron
-- coreboot mailing list: coreboot@coreboot.org http://www.coreboot.org/mailman/listinfo/coreboot
Hi,
I noticed that the HP DL380 servers that I've been working with lately wait a random amount of time before running the BIOS (up to 120sec) so that if you boot up many machines at the same time they won't have the power spikes at the same time when starting the fans and the disks.
Still, the BIOS itself is quite slow too, taking quite a lot of time, especially when it initializes the RAID controllers. I didn't measure it but it may be even more than a minute, including all the delays that can be skipped if you press Escape.
Regards, Cristi
OK, we found a keyboard that works. The BIOS is convinced there are two keyboards and two mice attached. There is one keyboard and no mice Just great.
ron
On Fri, Mar 5, 2010 at 8:58 AM, ron minnich rminnich@gmail.com wrote:
Just got a new nehalem box in for test yesterday. Experiences so far:
- POST from power-on takes 45 seconds. *45 SECONDS*. Now, I had it
said to me at SCALE7x last year from someone from Intel that all new BIOSes on Intel chips are really EFI underneath -- is this indicative of what we are to expect? If so, it's awful. It's 15 times slower than what we had ten years ago, and 50 times slower than what we can do today on coreboot.
As far as I can tell the sole purpose of EFI is to make it easier for hardware vendors to shovel more junk into the BIOS by removing the hurdle of hand-coding 16-bit assembly.
But while EFI might accelerate the trend, it's not the only villain. Someone noticed a 9x growth in boot time on qemu recently (http://lists.gnu.org/archive/html/qemu-devel/2010-03/msg00546.html ). Even on a virtual platform with no actual hardware to initialize, boot time will grow unless someone is actively pushing the other way.
Ultimately the system board vendors are responsible for the BIOS in the boards we buy. They are the ones cutting deals with Intel and AMI and Phoenix, and can exert the necessary leverage. But they won't, until they see 1-second cold boot as a feature that will sell more boards.
--Ed
I think every UEFI/EFI implementation will boot to "old school" boot mode when it can't find any EFI/UEFI-compliant boot-device/boot-partition. It would take too long though but at least the fallback is there.
-Darmawan
On 3/9/10, Ed Swierk eswierk@aristanetworks.com wrote:
On Fri, Mar 5, 2010 at 8:58 AM, ron minnich rminnich@gmail.com wrote:
Just got a new nehalem box in for test yesterday. Experiences so far:
- POST from power-on takes 45 seconds. *45 SECONDS*. Now, I had it
said to me at SCALE7x last year from someone from Intel that all new BIOSes on Intel chips are really EFI underneath -- is this indicative of what we are to expect? If so, it's awful. It's 15 times slower than what we had ten years ago, and 50 times slower than what we can do today on coreboot.
As far as I can tell the sole purpose of EFI is to make it easier for hardware vendors to shovel more junk into the BIOS by removing the hurdle of hand-coding 16-bit assembly.
But while EFI might accelerate the trend, it's not the only villain. Someone noticed a 9x growth in boot time on qemu recently (http://lists.gnu.org/archive/html/qemu-devel/2010-03/msg00546.html ). Even on a virtual platform with no actual hardware to initialize, boot time will grow unless someone is actively pushing the other way.
Ultimately the system board vendors are responsible for the BIOS in the boards we buy. They are the ones cutting deals with Intel and AMI and Phoenix, and can exert the necessary leverage. But they won't, until they see 1-second cold boot as a feature that will sell more boards.
--Ed
-- coreboot mailing list: coreboot@coreboot.org http://www.coreboot.org/mailman/listinfo/coreboot
On 3/9/10, Ed Swierk eswierk@aristanetworks.com wrote:
On Fri, Mar 5, 2010 at 8:58 AM, ron minnich rminnich@gmail.com wrote:
Just got a new nehalem box in for test yesterday. Experiences so far:
- POST from power-on takes 45 seconds. *45 SECONDS*. Now, I had it
said to me at SCALE7x last year from someone from Intel that all new BIOSes on Intel chips are really EFI underneath -- is this indicative of what we are to expect? If so, it's awful. It's 15 times slower than what we had ten years ago, and 50 times slower than what we can do today on coreboot.
As far as I can tell the sole purpose of EFI is to make it easier for hardware vendors to shovel more junk into the BIOS by removing the hurdle of hand-coding 16-bit assembly.
But while EFI might accelerate the trend, it's not the only villain. Someone noticed a 9x growth in boot time on qemu recently (http://lists.gnu.org/archive/html/qemu-devel/2010-03/msg00546.html ). Even on a virtual platform with no actual hardware to initialize, boot time will grow unless someone is actively pushing the other way.
Ultimately the system board vendors are responsible for the BIOS in the boards we buy. They are the ones cutting deals with Intel and AMI and Phoenix, and can exert the necessary leverage. But they won't, until they see 1-second cold boot as a feature that will sell more boards.
--Ed
Sorry about the double post. Something went wrong with my mail client.
Anyway, perhaps these articles by vid is a nice addition: http://x86asm.net/articles/introduction-to-uefi/index.html http://x86asm.net/articles/uefi-programming-first-steps/index.html
-Darmawan
-
Cristi Magherusan -
Darmawan Salihun -
Ed Swierk -
ron minnich -
Thomas Gstädtner