|
Alereon posted:I'd like to see how well 8 graphics cards (or even 4) work when sharing only 16 PCI-E lanes between them. Bridge chips like the NF200 and Lucid Hydra help, but that only goes so far. Maybe they stuck the switch on the PCH but that's still stupid because you only have 8 lanes there and DMI caps you at x4 anyway.
|
# ? Dec 29, 2010 18:17 |
|
|
# ? May 8, 2024 07:22 |
|
Eh... Maybe a CUDA cluster that needs inter-node communications? And with this post, I've caused pre-emptive headaches in thousands of university IT helpdesks that will get requests for such systems.
|
# ? Dec 29, 2010 18:18 |
|
Anyone knows whether there are virtualization improvements coming with SB? My Google-Fu leads me always to old Nehalem uArch slides and pages. --edit: vvv Multiple PSUs with power-on chaining. Combat Pretzel fucked around with this message at 18:37 on Dec 29, 2010 |
# ? Dec 29, 2010 18:20 |
|
Heh, it still won't run Crysis though. But seriously, who the hell needs 4 to 8 video cards? I guess it would be handy for scientific calculations but I can't think of anything else. And is there even a power supply that could run all of that? I wish we could get some idea what prices are going to be at launch. The i5-2500K is supposed to retail for $216 but if there are going to be plenty of them to go around what gives them the excuse to price gouge? I don't recall the Wolfdale chips being gouged when I got one right after they dropped and they were slim pickings for a few weeks too. I had a staring contest with newegg for half a day to get an E8400.
|
# ? Dec 29, 2010 18:33 |
|
Combat Pretzel posted:Anyone knows whether there are virtualization improvements coming with SB? My Google-Fu leads me always to old Nehalem uArch slides and pages. Maybe some minor uarch improvements. I think VMCS cache is a feature for the Xeon sku only.
|
# ? Dec 29, 2010 19:36 |
|
Raptop posted:I think VMCS cache is a feature for the Xeon sku only.
|
# ? Dec 29, 2010 20:36 |
|
spasticColon posted:But seriously, who the hell needs 4 to 8 video cards? I guess it would be handy for scientific calculations but I can't think of anything else. And is there even a power supply that could run all of that? Numbersmasher: Whisperstation: Octoputer: 8 NVIDIA Tesla GPUs—Over 8 TFLOPS 2 Quad- or Six-Core Intel® Xeon Nehalem/Westmere CPUs Up to 144GB DDR3 memory All from Microway.
|
# ? Dec 29, 2010 20:37 |
|
I imagine it'd be handy for storage machines with room for lots and lots of SAS expanders... but with SAS you just daisy chain the fuckers to begin with, so I dunno. Maybe lots of Infiniband cards? The use cases for so many PCI-e x16 slots are basically only for high-end, high-density servers like what you see in blade-based systems. I know there's people out there looking for 1TB of RAM in a machine, maybe there's a way to get that much RAM via PCI-e interconnects to RAM.
|
# ? Dec 29, 2010 22:11 |
|
fuseshock posted:If anyone hasn't seen these yet and wanted to start scoping out the new P67 motherboards, here are a few links: This was posted in the System Building, Upgrading, and Parts Picking Megathread. Just sharing it with this thread. I really hope those prices on ASUS boards on google is just a case of price-gouging.
|
# ? Dec 30, 2010 01:16 |
|
Bob Morales posted:
Serious academic computing: where science meets blingin'-rad blue LEDs.
|
# ? Dec 30, 2010 01:21 |
|
Combat Pretzel posted:All upcoming SB board from Asus do have EFI. And yes, you need EFI for +2TB drives, at least if you don't want to partition it. If the boot partition's below 2TB, then it should be OK I think. If you want to partition more than 2TB of space, even on a single partion, you have to use GPT. You cannot boot to a GPT partition with a BIOS, so you can't use more than 2TB on your boot drive with BIOS. mobby_6kl posted:but everyone is also forgetting about the page file. I have 6 gigs of RAM installed now, and according to windows 6143 MB are allocated right now, with 9214 being recommended. Windows' recommendations there are kind of out of date. It depends on what you're running, but 1-2GB is probably a better size for that much RAM.
|
# ? Dec 30, 2010 03:11 |
|
Space Gopher posted:What? No, it doesn't, at least not on Windows. As long as hibernation is enabled, hiberfil.sys exists on the root of the boot drive, equal in size to the amount of physical RAM in the system. Even if you could delete it, you'd still have a requirement to keep at least that much free space set aside, which boils down to the same thing: less space available for the rest of your data. Sorry, I am used to Linux, an OS designed by people who you can't read a newspaper through their ears.
|
# ? Dec 30, 2010 03:29 |
|
Nonpython posted:
The only difference between Linux and Windows here is that Windows guarantees adequate disk space to successfully hibernate, whereas Linux does not (necessarily).
|
# ? Dec 30, 2010 03:56 |
|
Zhentar posted:Windows' recommendations there are kind of out of date. It depends on what you're running, but 1-2GB is probably a better size for that much RAM. My Windows 7 laptop has 8 gb of ram but the hibernation file is 5 gb and the pagefile is 3 gb. I dunno what's up with that but it hibernates fine and the pagefile is fine. This is the automatic settings Windows did, not something I did myself.
|
# ? Dec 30, 2010 03:58 |
|
Nonpython posted:
Linux works the same way. A lot of Linux distros require the free space on the swap partition, rather than set aside as a user-visible file, but it has to be there. You have to have free disk space if you're going to suspend to disk. But great job trying, kiddo. Maybe one day, if you save up your pennies, you'll be able to afford a real desktop OS. madprocess posted:My Windows 7 laptop has 8 gb of ram but the hibernation file is 5 gb and the pagefile is 3 gb. I dunno what's up with that but it hibernates fine and the pagefile is fine. This is the automatic settings Windows did, not something I did myself. It's possible to use powercfg to reduce the hibernation file size. This is typically safe, because a lot of data in RAM is easily compressible, and it's possible that your laptop came configured that way (especially if it has an SSD out of the box). I believe Windows can throw out cache data if necessary when hibernating, too. Of course, if you turn the hibernation file size way down and don't have enough space to hold the contents of RAM, it'll bluescreen.
|
# ? Dec 30, 2010 04:32 |
|
Hmmm, I've been thinking... The does the SB GPU have embedded memory for the main framebuffer, or is it also stored in main memory? If latter is the case, I hope that various virtualization manufacturers remember that there's an on-die GPU and think about repurposing it. Like this, one guest can actually get hardware accelerated graphics without much issues and messing around. Xen and KVM already have "mostly" working code to passthrough primary and secondary adapters to their guests, I'd guess at least VMware is also toying around with that stuff.
|
# ? Dec 30, 2010 13:07 |
|
Combat Pretzel posted:Hmmm, I've been thinking... The does the SB GPU have embedded memory for the main framebuffer, or is it also stored in main memory?
|
# ? Dec 31, 2010 01:44 |
|
Bob Morales posted:
Yep, built a modeling computer for the lab that I work in last summer, 4 GTX 295s, renders in a few minutes what takes my home computer hours and hours. It also has the ability to instantly crash windows 7 unless you turn off a video card.
|
# ? Dec 31, 2010 02:01 |
|
Another thing that Sandy Bridge will be good for is GPU computing, since you can use the onboard video for display rendering and the Tesla card for computing. You have to use special compute-only drivers that doesn't support display output, because Windows won't allow you to perform any compute operations that take longer to complete than the driver timeout window (otherwise it thinks the driver hung and restarts it). The compute-only driver isn't a WDDM (Windows Display Driver Model) driver, so it isn't affected by this. Basically this means you can get away without needing a separate videocard for display in Tesla-equipped systems.
|
# ? Dec 31, 2010 02:44 |
|
Alereon posted:The GPU shares L3 cache and system RAM, to my knowledge. Ivy Bridge will change this by integrating up to 1GB of RAM using stacked LPDDR2 dies with a 512-bit bus. Assuming 800Mhz LPDDR2, that's 51.2GB/sec of memory bandwidth for the on-die GPU.
|
# ? Dec 31, 2010 12:14 |
|
I read on Wikipedia that the Ivy Bridge chips will be backwards compatible with Sandy Bridge motherboards. Is this true?
|
# ? Dec 31, 2010 21:51 |
|
spasticColon posted:I read on Wikipedia that the Ivy Bridge chips will be backwards compatible with Sandy Bridge motherboards. Is this true? As of now, yes. That said, they have advertised that in the past and changed their mind when the chip actually released. Also, you have to rely on the motherboard makers to update the BIOS to support the new architecture instead of trying to sell you a new motherboard with whatever new chipset will come out with Ivy Bridge, which is a gamble at best.
|
# ? Dec 31, 2010 22:25 |
|
Has anyone seen any P67 based mITX boards? I want to do a SFF gaming box.
|
# ? Jan 1, 2011 05:43 |
|
Jet Age posted:Has anyone seen any P67 based mITX boards? I want to do a SFF gaming box.
|
# ? Jan 1, 2011 13:17 |
|
ilkhan posted:I haven't been looking for them, but there are definitely s1156 mITX boards available. There will be s1155 boards, the question is when, not if. Yeah all the designs I've seen so far have been H67 based. As you say no doubt they will come, but it looks like it will be all H67 for mITX at launch.
|
# ? Jan 1, 2011 16:28 |
|
Jet Age posted:Yeah all the designs I've seen so far have been H67 based. As you say no doubt they will come, but it looks like it will be all H67 for mITX at launch.
|
# ? Jan 1, 2011 22:45 |
|
Space Gopher posted:It's possible to use powercfg to reduce the hibernation file size. This is typically safe, because a lot of data in RAM is easily compressible I guess if somebody was really super concerned about the size/seek time of a hibernation file they could always drop 25 bucks on an 80G hard drive to dedicate to that purpose and then go hog wild. E: Space Gopher posted:Linux works the same way. Imagine a series of four file system partitions sitting on the edge of a cliff.
|
# ? Jan 1, 2011 23:34 |
|
angry_keebler posted:I guess if somebody was really super concerned about the size/seek time of a hibernation file they could always drop 25 bucks on an 80G hard drive to dedicate to that purpose and then go hog wild. Wouldn't work on Windows. The bootloader only has enough smarts to access the root of the boot drive, so that's where you have to keep the hibernation file. Besides, even if you could, an 80 gig hard drive would be slow as poo poo on the restore; old hard drives have pathetic sustained bandwidth. Putting 4-8 gigs of data into RAM at 75 MB/sec would take long enough that you'd probably be better off just cold-booting from the SSD. angry_keebler posted:Imagine a series of four file system partitions sitting on the edge of a cliff. requesting name change to "johnny fiveoses," tia
|
# ? Jan 2, 2011 00:09 |
|
Space Gopher posted:Wouldn't work on Windows. The bootloader only has enough smarts to access the root of the boot drive, so that's where you have to keep the hibernation file. Yeah you're right. I was thinking of swsusp, which as far as I know lets you drop your suspend image on any partition you define. Space Gopher posted:Besides, even if you could, an 80 gig hard drive would be slow as poo poo on the restore; old hard drives have pathetic sustained bandwidth. Putting 4-8 gigs of data into RAM at 75 MB/sec would take long enough that you'd probably be better off just cold-booting from the SSD. This is also a very good point.
|
# ? Jan 2, 2011 00:34 |
|
Most of your RAM isn't actually in use, Windows doesn't write empty pages or the disk cache to the hibernate file. The actual amount of data to read/write is probably well under a gig for most people.
|
# ? Jan 2, 2011 04:30 |
|
ilkhan posted:ok, you're not going to find P67 mITX boards, but yes for s1155. There is no reason for a company to design a P67 board that requires a discrete GPU on a mobo that small. IF there is a nich market that needs to be catered to jetway will most likely make a poo poo board that the big mfgs will sell in their own color scheme.
|
# ? Jan 2, 2011 06:51 |
|
Since Sandy Bridge is coming out in 3 days, maybe we should prepare for the actual installation, as Windows will inevitably flip out when it learns that the old motherboard has been replaced by a new one. From googling, I get the sense that this is what you do in this situation: Boot from your windows cd and do a repair installation. If that fails, format and reinstall windows (gently caress). Anyone have a more fool-proof method?
|
# ? Jan 2, 2011 18:51 |
|
Dr. Gaius Baltar posted:Since Sandy Bridge is coming out in 3 days, maybe we should prepare for the actual installation, as Windows will inevitably flip out when it learns that the old motherboard has been replaced by a new one. From googling, I get the sense that this is what you do in this situation: To be fair, I had just put together a brand new I5 system 2 weeks ago and installed Win 7 x64 on it. Worked great for 3 days till poo poo got all hosed up. I then RMA'd my MoBo, CPU and RAM back, and took the SSD I installed Win 7 to and connected it to my old system, it booted up just fine, installed drivers and everything was great. So, you might not have to do anything at all, at least for Windows 7* *The only hardware that is the same between the 2 systems is the video card and SSD, otherwise a completely different system.
|
# ? Jan 2, 2011 18:59 |
|
Dr. Gaius Baltar posted:Boot from your windows cd and do a repair installation. If it's XP then just run SysPrep, assuming the hard drive controller uses the same driver on both systems (ie. ICH*R), otherwise you're in for a world of hurt. If it's Vista or 7 then see the post above.
|
# ? Jan 2, 2011 19:01 |
|
That's cool because I thought even with my retail copy of Win7 I would have to at least do a repair install and hope for the best. I would probably have to still reinstall my wireless card and sound card though because when I moved my wireless card to a higher PCI slot to install my sound card a few weeks ago I had to reinstall the wireless card.
|
# ? Jan 2, 2011 19:39 |
|
KakerMix posted:To be fair, I had just put together a brand new I5 system 2 weeks ago and installed Win 7 x64 on it. Worked great for 3 days till poo poo got all hosed up. I then RMA'd my MoBo, CPU and RAM back, and took the SSD I installed Win 7 to and connected it to my old system, it booted up just fine, installed drivers and everything was great. Between Windows XP and Windows Vista, there were a lot of things unified and reorganized with the goal of simplifying maintenance and imaging. You can still (and probably should) sysprep it first, but if you can't, well simply dropping it into a new box can't hurt.
|
# ? Jan 2, 2011 23:08 |
|
Unless the chipset is the same I always do a fresh install no matter what when getting a new mobo. Trying to save the original install usually seems to be more trouble then its worth in the long run.
|
# ? Jan 3, 2011 03:12 |
|
Sounds like Sandy Bridge is going to be awesome, for overclocking (turbo) at least. Overlock3D review: http://www.overclock3d.net/reviews/cpu_mainboard/i7_2600k_i5_2500k_2300_1155_sandy_bridge_review/1 i5-2500k http://www.youtube.com/watch?v=achL4hIAvpo i7-2600k http://www.youtube.com/watch?v=PSET4P6EZ30 fuseshock fucked around with this message at 06:20 on Jan 3, 2011 |
# ? Jan 3, 2011 05:42 |
|
Holy poo poo reviews be out son! Anandtech Sandy Bridge desktop i3/i5/i7 review Anandtech: Sandy Bridge: Upheaval in the Mobile Landscape
|
# ? Jan 3, 2011 06:18 |
|
|
# ? May 8, 2024 07:22 |
|
Yeah the embargo is up. Tech Report: http://techreport.com/articles.x/20188
|
# ? Jan 3, 2011 06:34 |