|
mew force shoelace posted:You know what would go a long way towards making computers seem better? A dedicated chip and memory just to run the OS. I'd be pretty tiny and it'd take out a ton of the apparent slowness of a computer. Yeah, on an application page fault what you really want to do is go off-chip for the handler instead of just running the OS right on that core. That'd make things zippy.
|
# ¿ Sep 23, 2010 18:02 |
|
|
# ¿ Apr 25, 2024 17:26 |
|
Combat Pretzel posted:Trying to figure out what VMCS is, I ran over VirtualBox documentation that suggests that it's a feature available on all Intel CPUs with VT-x. The Virtual Machine Control Structure (VMCS) is explained in the Intel PRM Vol. 3B, if that's not quite enough to put you to sleep you can read the rest of the PRM. It's a 4kb region of memory containing guest state, host state, control bits, etc. accessed through the vmread/vmwrite instructions. Basically it's an implementation detail of VT-x and unless you're rolling your own hypervisor you shouldn't care about it. If any of the other SNB vets in FM want to meet up for lunch or at least to awkwardly stare at each other's shoes, hit me up on PM.
|
# ¿ Jan 3, 2011 20:09 |
|
movax posted:Yes; you are increasing multi and therefore increasing clock frequency, and if you bump VCore for stability, you are sucking down more power (P = IV). From a couple days back, but a slight correction: For dynamic circuits power scales with the square of voltage, P=C·V2·f, not linearly. P is linear with respect to frequency. It's not a huge difference when you're moving .05V at a time though.
|
# ¿ Jan 13, 2011 20:31 |
|
I went through ~100 heatsink replacements last week and I really don't get the hate for it.
|
# ¿ Jan 31, 2011 19:25 |
|
movax posted:So looking at the PCIe 3.0 spec, I guess I lied, there are some protocol layer changes in addition to physical layer changes. They went to 128b/130b for 8.0GT/s operation + a ton of new specifications PHY-wise for the 8.0 rate. Yeah it's nothing like gen2 where the diff of the entire spec against gen1 is like 5 lines, gen3 had to do some extreme changes to hit double the bandwidth.
|
# ¿ Aug 4, 2011 19:12 |
|
MeruFM posted:if internet becomes so ubiquitous that at any given moment, your computer is connected to multiple high quality lines, then it is a possibility. Wedesdo posted:Spoof your own server or re-write the microcode.
|
# ¿ Aug 18, 2011 20:11 |
|
Alereon posted:Where are our 8GB DIMMs already? At least 300ns away, according to the JEDEC spec.
|
# ¿ Sep 14, 2011 18:15 |
|
Agreed posted:Although it wouldn't surprise me under any circumstances, really, specialized industries are pretty incestuous and no noncompete is going to last 13 years. I know one guy who's worked for Cadence 4 times. He's never been hired there, but he's worked for 4 different companies that were bought by Cadence.
|
# ¿ Sep 19, 2011 18:18 |
|
freeforumuser posted:Word. Intel tocks (Penryn, IB) aren't really worth waiting for, it's the ticks (Conroe, Nehalem, SB) that are far more impressive. Good job getting tick/tock literally backwards.
|
# ¿ Nov 2, 2011 17:59 |
|
Shaocaholica posted:So have all previous Intel 6 core procs actually been 8 core procs with 2 turned off? No.
|
# ¿ Dec 1, 2011 23:39 |
|
More ISSCC coverage: http://www.eetimes.com/electronics-news/4236562/Intel-gives-deeper-look-into-Ivy-Bridge?cid=NL_EETimesDaily
|
# ¿ Feb 22, 2012 01:59 |
|
Alereon posted:that Intel China exec. Sean Maloney?
|
# ¿ Mar 7, 2012 01:18 |
|
COCKMOUTH.GIF posted:(why would I want built-in WiFi on a motherboard?) Yeah seriously, you want WiFi on-die.
|
# ¿ Apr 23, 2012 03:13 |
|
incoherent posted:Kneecapping IVB so there is a big difference between IVB and IVB-E. Yeah IVB-E doubling the cores, using a bidirectional ring, completely rearchitecting I/O... all fluff. This internal heat transfer poo poo literally nobody sees except a couple overclockers, that's where there's some tricky segmentation happening.
|
# ¿ Apr 30, 2012 15:13 |
|
All the p-state stuff is well documented in the ACPI spec. It defines the communication between applications, the OS, and hardware regarding power states. It's basically the evolution of the MPS spec, APM, all that fun stuff. It's a great read if you're some kind of nerd.
|
# ¿ Jul 6, 2012 05:21 |
|
movax posted:I unfortunately have experience with ASL now, I will die forever alone and unloved, like a true super-nerd. Some program manager somewhere decided we should support ACPI-mediated PCIe hot-plug as well Shaocaholica posted:Do Macs use ACPI? I ask because Apple isn't on the list of companies that contribute to it.
|
# ¿ Jul 6, 2012 19:29 |
|
Shaocaholica posted:So Apple doesn't contribute to ACPI because of how new they are to x86? Why do you think ACPI is x86-only? You're also vastly misunderstanding the depth of information that ACPI can provide. The extensions have mostly been to support entirely new interfaces like PCIe cards.
|
# ¿ Jul 6, 2012 22:44 |
|
Apple would have been fools to not use little-endian when they jumped over to x86. Therefore all little-endian machines are x86.
|
# ¿ Jul 9, 2012 21:49 |
|
Richard M Nixon posted:I'm coming from a Haswell i7 You're from the future?!?!
|
# ¿ Aug 3, 2012 20:51 |
|
Henrik Zetterberg posted:Still rolling on my old Yorkfield and other hardware just as old, but just got my complimentary IVB i7-3770k from work. Pairing this up with a DZ77RE-75K and 240gb 520 SSD. rhag posted:I rarely touch a laptop, and when i do, the power of the device is irrelevant for the task. I prefer to have a CPU that is focused on doing whatever is doing as fast as possible, leaving other things (such as graphics) to the dedicated boards/devices.
|
# ¿ Oct 23, 2012 15:46 |
|
WhyteRyce posted:We're entering an era where all the new engineering students just have smart phones and tablets and never have done any actual work or tinkering around on their own The boards I learned on seem hilariously quaint by comparison. Parallel port JTAG, a separate board for USB, Windows MFC programming... so painful. Henrik Zetterberg posted:I'll tell you what though, eliminating validation on a whole package saves approximately 1 million headaches for me. Bye LGA!
|
# ¿ Nov 29, 2012 18:59 |
|
Haswell 8w?
|
# ¿ Jan 4, 2013 19:22 |
|
Alereon posted:the current 17W products have rather anemic performance "Anemic" compared to what?
|
# ¿ Jan 4, 2013 20:17 |
|
Fruit Smoothies posted:I thought intel were just being ultra-picky about which parts they make OEM use to get their "certification" rather than making it more SOC like. Certainly on the haswell front. Could be wrong, though He's talking about Rosepoint.
|
# ¿ Mar 24, 2013 14:39 |
|
Is there some separate thread for the Brian Krzanich election? Or just no interest in discussing it?
|
# ¿ May 2, 2013 20:26 |
|
Otellini announced he was retiring back in November, oddly timed since he could have gone a couple more years if memory serves. His reasoning was along the lines of there's a Big Decision about the future of Intel that the next CEO will have to live with the consequences of for their entire tenure and he didn't want to make it for them. I no longer work for the company, but I'm still not sure how much of that is public or not so apologies on being vague. I'm a little surprised they went with Krzanich, the incredibly safe choice of promoting the current COO. The very possibility of an outside hire, even if "outside" meant Pat Gelsinger, was really interesting. I think I'm trying to read a lot more into the phrase "open-minded approach to problem solving" than was intended, but he might have tipped his hand on that Big Decision.
|
# ¿ May 2, 2013 20:46 |
|
Alereon posted:I can't imagine AMD management or shareholders going for that given they are about to start making a SHITTON of money from consoles for 5+ years. Consoles aren't huge volumes though. The PS3 shipped 70M over 6 years, Ivybridge shipped 100M in one year. Different margins and continuing engineering support profiles too. Totally agreed on the x86 monopoly/ARM issue though. ARM's in a great position and there have been a few decades of research to abstract away the ISA layer between HLL's and performance. Apple in particular has practical experience swapping out the ISA and complete control of the iOS toolchain. I still think ARM would have huge growing pains if they tried to scale to a billion+ transistors and it isn't a given they could get to today's desktop range without burning up. But a monopoly on x86 isn't worth what it used to be.
|
# ¿ May 2, 2013 21:41 |
|
WhyteRyce posted:I just assumed that Paul stepping down was do to Intel missing the mobile boat and being really late to the game. That's just me guessing with no actual information or anything. quote:I'm assuming that picking a guy with TMG experience is an indication of where the company is going and/or what they plan on doing WhyteRyce posted:And boring CEO choice where is my stock pop
|
# ¿ May 2, 2013 23:06 |
|
flavor posted:I'm not an economist, but if revenue is an indicator of scale, then I'd think Intel and AMD still have no excuse to merge. Maybe in a few years. iPhones alone have shipped >250M, iPads >100M. That's dwarfed by the number of Androids. Cell phones are a fraction of the raw count of ARM cores shipping. The margins are a different picture and a core-to-core count isn't great, but the basic picture is that ARM currently dominates mobile and mobile's on the way up. Factory Factory posted:Control over the CPU-GPU HSA, maybe. That's AMD's big push and big hope, yet Haswell having the ability to push a GPU memory-space pointer to the CPU shows that Intel is trying to do the same thing. If Intel had AMD, then Intel could do one, non-segmented HSA for x86. Especially one which conveniently only worked with Iris and Radeon hardware, leaving Nvidia at a disadvantage in the compute space it gets so much money from.
|
# ¿ May 3, 2013 16:40 |
|
flavor posted:Same thing applies to x86 with the exception of the manufacturing of the actual CPUs. It looks to me like you are trying to compare everything but the kitchen sink on the ARM side to just the bare manufacturing of the CPUs on the x86 side. I could go on, but the short version is that ARM is a diverse ecosystem of several companies. x86 is the Big Two and hardly anybody else. If you're going to call my analysis specious it might help you to actually supply, you know, a fact or two? flavor posted:I'd just like to see the actual numbers with a good analysis and not just commonplace "mobile is on the way up" projections where somebody uses a ruler and draws a straight line to 2050 based on the 2007 and 2012 values or similar. Intel openly acknowledges this in a lot of ways, so I can't imagine why you're holding this point in contention. At the 2011 investor meeting, then-CEO Paul Otellini asked "600 smartphones were sold. Who made the most money? Intel, because someone had to buy a Xeon to support the backend." There are a lot of industries where a high-end manufacturer ceded the low-end to cheap competitors, who ramped up on the huge volumes, got some experience, then beat the high-end player at their own game. Steel, manufacturing, etc.
|
# ¿ May 4, 2013 00:16 |
|
flavor posted:It looks a little bit like you're trying to push some kind of narrow limit around x86 and a wide one around ARM, and everyone who brings this up meets with condescension. If you feel you have very convincing arguments, let them speak for themselves. ARM is an ecosystem. Intel is an IDM. Shaocaholica posted:What exactly is a stepping anyway? Is it like a different layout of the same logic?
|
# ¿ May 4, 2013 03:11 |
|
WhyteRyce posted:I do love how the forecasted cost, the thing that drove the decision, was wrong. It's not quite that simple. That's back in the time frame when Intel still had an in-house ARM team, and the bet was made on x86. And the "forecasted cost" being wrong wasn't simply a matter of the numbers being off. The entire model had to change and it wasn't clear how to do that, certainly not for a <5M part.
|
# ¿ May 17, 2013 16:05 |
|
The flip side to slow to change direction being that once Intel starts focusing something, they're bringing incredible pressure to bear on the problem. WhyteRyce posted:I don't believe switching from ARM to ARM isn't quite as big of a deal as switching from x86 to ARM or vice versa, but I could be wrong.
|
# ¿ May 17, 2013 23:27 |
|
necrobobsledder posted:(I am not a believer that Intel has won from the cloud revolution unlike some writers / visionaries out there). necrobobsledder posted:I still think the tablet game is basically over though without a way for developers to port everything over to x86 quickly. Apple may ironically make x86 mobile rather relevant if they release a solid iOS for x86 tool suite something similar to Rosetta in pursuit of their "iOS ALL THE THINGS" strategy that's highly controversial.
|
# ¿ May 28, 2013 17:32 |
|
HalloKitty posted:Arbitrarily missing features from unlocked CPUs How many overclocking stress tests even exercise VMX functionality?
|
# ¿ Jun 5, 2013 16:35 |
|
movax posted:I would have been really upset if Vt-x was disabled on K SKUs; I can understand no Vt-d support though, as I'm not running PCI passthrough on my desktop (almost certainly on a dedicated hypervisor box though). Just thinking of the case where someone OC's, 'proves' it with a compute-heavy workload, then throws a VM on it and tears hair out when a leaf of a rare vmexit becomes the critical path.
|
# ¿ Jun 5, 2013 19:07 |
|
Shaocaholica posted:Passed 200 hours of prime95 but crashes on Crysis right away? I've seen that case and never have I heard anyone blame the chipmaker or the developers. Gosh, immediate crashes are easy to root cause? Validation missed a precious gem in you.
|
# ¿ Jun 5, 2013 20:05 |
|
It shouldn't be a stepping stone to design, it should be a stepping stone to architecture. And they might just be the special snowflake types who would've been equally unhappy at the reality of design work staring at the same 5 critical paths for a couple months. imo.
|
# ¿ Jun 5, 2013 23:57 |
|
Phantom Limb posted:Fair enough. I think my perception of it is also colored by the fact a lot of younger engineers at my company got shoved into doing validation since there was no other work, and they pretty much universally hated it. You've got to stamp that out. There's a bit of a difference between what companies call "validation/verification". There's the team that answers "Is the layout an equivalent circuit to the RTL?" that's mostly automated now and if you have humans doing any significant portion of it, yeah it's not the best. There's also the team that answers "Is this design doing what's expected?" that requires actual engineering since over the life of the project you have time to run maaaaaybe 1 second of real CPU time, so you have to be clever what tests you run on what portions of the chip to make sure the billions to get the first ones back enables the post-silicon teams to run multiple seconds () of quality content to flush out the nastier bugs. If you exclusively hire RCG's, treat them like second class citizens, and actively poach the best of them for other teams... really hard to imagine an outcome other than a terrible team? Ideally you've got experienced validators who engage during the early stages of architectural planning (or even take architectural ownership) to shape the project in a way that's easy to validate to cut that long pole down and let you build a wider tent.
|
# ¿ Jun 6, 2013 02:56 |
|
|
# ¿ Apr 25, 2024 17:26 |
|
Shaocaholica posted:I always thought that the vast majority of overclockers consider a speed stable only if ALL aspects of the CPU are stable. Hence the 100s of different stress tests. It's cute that you think "100s" is a big big number though It's ok, most designers are quite myopic. sincx posted:Monopoly pricing.
|
# ¿ Jun 6, 2013 23:02 |