|
PerrineClostermann posted:So what's up with these new Bay Trail Atoms? Why are they so damned good? The architecture of the old atom was really antiquated, which is why its performance was never anything to write home about. The bay trail has a completely revamped, more modern architecture which results in a huge performance boost.
|
# ? Jan 6, 2014 00:08 |
|
|
# ? Apr 24, 2024 19:20 |
|
What architecture was the old Atom based on? And the new one? And more importantly, does the new Atom have that nifty hardware support for fast AES?
|
# ? Jan 6, 2014 00:13 |
|
JawnV6 posted:Totally solid on the first page. The second... ehhhh. I don't think QPI is a slam dunk. It's quite heavy and hadn't quite been banged into an IP. Assuming that the EX line will be sharing architecture and resources is another stretch. Overall solid, just a little wishful on some of those features. I think Altera and Xilinx both have QPI IP "available", but it certainly isn't hard IP like PCIe in almost all their parts. I'd think you'd get to market much faster going with PCIe; more mature cores, and the PCIe RC has moved to being a first-class citizen these days instead of being a few hops away from CPU and memory.
|
# ? Jan 6, 2014 00:15 |
|
PerrineClostermann posted:What architecture was the old Atom based on? And the new one? And more importantly, does the new Atom have that nifty hardware support for fast AES? Yes, Silvermont's uarch has AES-NI. The new Atom is an entirely new design that's ISA-compatible with Westmere-era Nehalem CPUs. If it's an instruction in Westmere, it's in Silvermont. The first-gen Atom was also a new design, at the time. The Bonnell uarch had SSE3 and hyperthreading, so it was ISA-compatible with Prescott Pentium 4s. Neither is based on anything but the x86 instruction set and being low-power, really.
|
# ? Jan 6, 2014 00:18 |
|
PerrineClostermann posted:What architecture was the old Atom based on? And the new one? The old architecture was Bonnell and the new one is Silvermont. The biggest factor in the performance difference is that the Bonnell had in-order execution and the Silvermont has fully out-of-order execution and improved branch prediction, which results in like a 50% or more increase in the amount of instructions it can execute in a single clock cycle.
|
# ? Jan 6, 2014 00:33 |
|
JawnV6 posted:Totally solid on the first page. The second... ehhhh. I don't think QPI is a slam dunk. It's quite heavy and hadn't quite been banged into an IP. Assuming that the EX line will be sharing architecture and resources is another stretch. Overall solid, just a little wishful on some of those features. 16 GB eDRAM jesus christ
|
# ? Jan 6, 2014 01:19 |
|
JawnV6 posted:Totally solid on the first page. The second... ehhhh. I don't think QPI is a slam dunk. It's quite heavy and hadn't quite been banged into an IP. Assuming that the EX line will be sharing architecture and resources is another stretch. Overall solid, just a little wishful on some of those features. Yeah. With PCIe and what they've picked up in the QLogic IB / Cray IP acquisitions, it's hard to see a need for QPI in the kind of systems KL will be deployed in. I can't imagine the effort that it would take to wedge the RAS poo poo that comes with the EX platform onto KL.
|
# ? Jan 6, 2014 18:03 |
|
PCjr sidecar posted:Yeah. With PCIe and what they've picked up in the QLogic IB / Cray IP acquisitions, it's hard to see a need for QPI in the kind of systems KL will be deployed in. I can't imagine the effort that it would take to wedge the RAS poo poo that comes with the EX platform onto KL.
|
# ? Jan 6, 2014 18:24 |
|
Alereon posted:It seems like you need QPI if you want this to act like a processor attached to your system with 72 ultra-efficient x86 integer cores that also have godlike double-precision floating point performance. This isn't necessary or helpful if all you care about is using it as an accelerator card which is the kind of applications people have been thinking about so far, but if you're Intel this could enable compelling new applications, as well as ensuring people are buying Intel servers to pair KNL with. They might have 2 versions, one with cache coherency and QPI for small workstation type workloads, and one without for HPC
|
# ? Jan 6, 2014 18:33 |
|
Pardon my ignorance, but maybe they don't need PCIe at all? Most servers are shipped in a fixed configuration, what hangs off the PCIe bus other than network and storage controllers? If those had QPI interfaces (or storage was removed and network was QPI), and they could because Intel makes both, I don't think a lot of customers would miss PCIe. Where am I wrong? Also, does Knights Landing have AES-NI on each core?
|
# ? Jan 6, 2014 18:35 |
|
I would imagine there's AES-NI, since it's Silvermont cores and those have AES-NI. Common server PCIe thingies:
|
# ? Jan 6, 2014 19:23 |
|
Alereon posted:It seems like you need QPI if you want this to act like a processor attached to your system with 72 ultra-efficient x86 integer cores that also have godlike double-precision floating point performance. This isn't necessary or helpful if all you care about is using it as an accelerator card which is the kind of applications people have been thinking about so far, but if you're Intel this could enable compelling new applications, as well as ensuring people are buying Intel servers to pair KNL with. With 2 QPI links, you have about 64 GB/s between socks, compared to 300 GB/s to local memory (with current gen Phi.) With eDRAM, you're going to see a more significant performance difference between on- and off-socket performance. You're going to want to keep thread memory access local to the socket. If you're doing that, its only moderately harder to spawn those processes on an accelerator and set up some sort of shared memory access between host and accelerator. If you use PCIe, you can get 32GB/s bidirectional between card and host, and easily get 2 sockets + 4 accelerators or more in a single system. With QPI, you're limited* to 2 sockets (4 if you drop to one link per socket.) Factory Factory posted:In servers, only the first socket's PCIe is broken out anyway. Only on very cheap systems. Malcolm XML posted:They might have 2 versions, one with cache coherency and QPI for small workstation type workloads, and one without for HPC That would be two very different designs, and very expensive for a small market.
|
# ? Jan 6, 2014 20:57 |
|
Intel announced dual-boot Android/Windows tablets at CES...for those of you that care about that sort of thing. Oh and McAfee mobile products will be free...for those of you that care about that sort of thing. This sounds cool though http://www.theverge.com/2014/1/6/5282472/intel-announces-edison-a-computer-the-size-of-an-sd-card/in/5046515 quote:At CES 2014, Intel CEO Brian Krzanich announced Edison, "a full Pentium-class PC" that's the size and shape of the SD card you might otherwise put in your camera. It's powered by a dual-core Quark SOC, runs Linux, and has built-in Wi-Fi and Bluetooth connectivity, according to the company. Intel even has a specific app store designed for Edison, and a special version of Wolfram that will come to the tiny computer WhyteRyce fucked around with this message at 04:34 on Jan 7, 2014 |
# ? Jan 7, 2014 04:31 |
|
All it reads is "Please put our hardware in your phone, we ported android for you!!!" I don't think anyone is seriously thinking about putting any intel hardware out in phone factors.
|
# ? Jan 7, 2014 05:16 |
|
WhyteRyce posted:Intel announced dual-boot Android/Windows tablets at CES...for those of you that care about that sort of thing. At last, the answer to the question that no one asked.
|
# ? Jan 7, 2014 05:19 |
|
incoherent posted:All it reads is "Please put our hardware in your phone, we ported android for you!!!" I don't think anyone is seriously thinking about putting any intel hardware out in phone factors. Not only have there been handsets with Intel Atoms already, but Asus announced some high-end handsets at CES that use Bay Trail.
|
# ? Jan 7, 2014 05:24 |
|
incoherent posted:All it reads is "Please put our hardware in your phone, we ported android for you!!!" I don't think anyone is seriously thinking about putting any intel hardware out in phone factors. Motorola did and it worked out quite well. A chunk of Samsung tablets also use Intel chips. Honestly, what's going to drive Intel adoption in the handheld space is going to be limited fab capacity at places like TSMC. Demand is going to continue to ramp and Intel has fab capacity to spare. All it will take is for an OEM to face SoC shortages of Qualcomm chips and have to make a choice between shipping a product late or using an Intel SoC.
|
# ? Jan 7, 2014 05:25 |
|
Factory Factory posted:Not only have there been handsets with Intel Atoms already, but Asus announced some high-end handsets at CES that use Bay Trail. The T100 has been doing very well if Amazon's top seller charts are an indication of anything. And Intel pretty much has the Windows tablet market locked up...which doesn't mean a lot right now but maybe they can start making progress with Android tablets now that they have good hardware and better software support. WhyteRyce fucked around with this message at 06:13 on Jan 7, 2014 |
# ? Jan 7, 2014 06:08 |
|
re: QPI on KNL--a cache-coherent interconnect for an accelerator is the number one feature request of everyone that has ever programmed an accelerator, including HPC sites. Intel clearly knows that. it makes ports dramatically easier, as it removes 95% of the terrible side effects of existing GPU ports. I would be absolutely shocked if KNL is not on QPI.
|
# ? Jan 7, 2014 06:09 |
|
Professor Science posted:re: QPI on KNL--a cache-coherent interconnect for an accelerator is the number one feature request of everyone that has ever programmed an accelerator, including HPC sites. Intel clearly knows that. it makes ports dramatically easier, as it removes 95% of the terrible side effects of existing GPU ports. I would be absolutely shocked if KNL is not on QPI. Yeah cache coherency owns. Hypertransport has been open ip for a while.
|
# ? Jan 7, 2014 09:50 |
|
Is thrashing bad? Thermally? As in I need to process 1000s of files. Each file read puts the CPU idle since there's nothing to crunch. Once the file is read the CPU (all cores) go to 100% for 2-3 seconds. A file is written out so the CPU is idle again and the process repeats for days/weeks/months. I've monitored the temps and it swings from ~35C to ~70C between work units. That can't be good right? Over 1000s of iterations that last days and weeks? Could the thrashing be bad for the heatsink? Would a heatsink with more mass help so it can buffer the temp swings?
|
# ? Jan 7, 2014 19:26 |
|
Thermal cycling is perfectly normal and does not hurt the CPU at all. Don't worry about it.
|
# ? Jan 7, 2014 19:46 |
|
Can someone explain what an 'R' CPU is? Specifically I'm reading about the Gigabyte Brix Pro and one of the CPU options is an Intel i7-4770R. I'm guessing its a desktop CPU but ARK is telling me its FCBGA1364 so its soldered. Does it have the usual desktop IHS or is it naked?
|
# ? Jan 12, 2014 18:54 |
|
Shaocaholica posted:Can someone explain what an 'R' CPU is? Specifically I'm reading about the Gigabyte Brix Pro and one of the CPU options is an Intel i7-4770R. I'm guessing its a desktop CPU but ARK is telling me its FCBGA1364 so its soldered. Does it have the usual desktop IHS or is it naked?
|
# ? Jan 12, 2014 19:08 |
|
Alereon posted:This is a naked, soldered-on desktop CPU with Intel Iris Pro 5200 graphics. That is 40 shaders plus 128MB of eDRAM acting as L4 cache, versus the typical 20 shaders without L4 cache you get on a desktop CPU. Performance is on the order of a Geforce GT 650M, which is quite respectable. I take it these parts haven't been used much in the market? When I search for that part I only get the Brix Pro and its been out for 2 quarters. It would be neat to have something like that in a Mac Mini form factor but I know that even the Bris Pro is fatter than a Mini without a power supply either. Still, those millions of dollars of engineering effort should be good for something.
|
# ? Jan 12, 2014 20:04 |
|
Shaocaholica posted:I take it these parts haven't been used much in the market? When I search for that part I only get the Brix Pro and its been out for 2 quarters. It would be neat to have something like that in a Mac Mini form factor but I know that even the Bris Pro is fatter than a Mini without a power supply either. Still, those millions of dollars of engineering effort should be good for something. Do note that the BRIX Pro 4770R did throttle under CPU+GPU load in Anandtech's test. It's likely not going to throttle in real-world usage, but Turbo won't work as well as it would if the CPU was properly cooled. Then again, that's still lightyears ahead of the 15W TDP CPUs in the Brix S. Alereon fucked around with this message at 20:29 on Jan 12, 2014 |
# ? Jan 12, 2014 20:25 |
|
Should I wait until the Haswell refresh or Broadwell-D come out to upgrade my antique Phenom and switch to Intel, or are they likely to be more expensive?
|
# ? Jan 18, 2014 10:05 |
|
atomicthumbs posted:Should I wait until the Haswell refresh or Broadwell-D come out to upgrade my antique Phenom and switch to Intel, or are they likely to be more expensive?
|
# ? Jan 18, 2014 10:47 |
|
Anyone know when the Bay Trail desktop and server stuff (NUC, motherboards, etc) will be generally available? I'm not seeing too much besides "Q1 2014".
|
# ? Jan 18, 2014 22:20 |
|
MSI showed a board at CES suggested for end of the month. Only dual core, though - quad core supposed to come a few weeks later.
|
# ? Jan 18, 2014 23:08 |
|
My Rhythmic Crotch posted:Anyone know when the Bay Trail desktop and server stuff (NUC, motherboards, etc) will be generally available? I'm not seeing too much besides "Q1 2014". I couldn't find anything today related to the new Celeron NUC, the DN2820FYKH. I really wish Amazon at least had a preorder available, or some way to do a availability email or something. The i3 is available already: http://www.amazon.com/Intel-D34010W...ords=DN2820FYKH kri kri fucked around with this message at 04:31 on Jan 20, 2014 |
# ? Jan 20, 2014 04:28 |
Really really basic question, but what was that Intel chipset for uhh, somewhere between Sandy Bridge and Haswell that was supposed to have like 14 SATA ports built in? I hope something new in the next year or so has an overpriced 6+ core CPU for less than triple the cost of the consumer 4 core, I'm really antsy to upgrade this 2500K
|
|
# ? Jan 20, 2014 11:15 |
|
Straker posted:Really really basic question, but what was that Intel chipset for uhh, somewhere between Sandy Bridge and Haswell that was supposed to have like 14 SATA ports built in? Ivy Bridge-E is the CPU you're referring to, socket 2011, chipset X79. It doesn't have a lot of SATA3 ports natively, and is an old chipset (from 2011, aptly enough). It has 6 on chipset, but unless the chipset is workstation validated, Intel requests you don't use all of them. So most of these boards don't use a lot of Intel ones, they just tack on Marvell chipsets for the SATA3 ports. Also, X79 doesn't have native USB3, so boards with it commonly use ASmedia chipsets for the USB3. Having an ASmedia chipset on my Z68 board, I know I would not upgrade to another board unless it had Intel-native USB3. X79 isn't something you'd want to upgrade to. It's old, clunky, and all you get is two extra cores for a lot of cost; and the speed improvements in Ivy vs Sandy are minimal. I wouldn't do it unless you have something that's highly threaded that you use regularly, and have money to burn. vv It still might be X79. On the PCH itself it has: 2x SATA3, 4x SATA2, 4x SATA3/SAS ports. But as mentioned, Intel asks you not to use the SAS ports unless you pay for a validated chipset. Depends whether you think 10 is fucktonne of ports or not. HalloKitty fucked around with this message at 11:31 on Jan 20, 2014 |
# ? Jan 20, 2014 11:18 |
those were two separate questions (I'm not dishing out nearly $600 for a now-old CPU) but thanks maybe it didn't make it to production, but I definitely remember something about an Intel chipset/controller that supported a loving ton of SATA ports. It would've been in production for at least a year now, so the announcement would've been pretty old. Even at the time I remember it seemed like it'd fill a pretty small niche, since as great as plugging poo poo directly into Intel ports and not having to worry about it is, there aren't many people who need more than like 8 drives that won't just be using PCIe controllers. Straker fucked around with this message at 11:28 on Jan 20, 2014 |
|
# ? Jan 20, 2014 11:24 |
|
Does anyone know when Ivy Bridge EX is going to start hitting the market? There are a couple 3rd party sites with details but I can't find any information on Intel's own site. 24 DIMMs per socket is so sexy for per-core licensed database servers if these CPUs aren't ridiculously expensive.
|
# ? Jan 22, 2014 04:39 |
|
Chuu posted:Does anyone know when Ivy Bridge EX is going to start hitting the market? There are a couple 3rd party sites with details but I can't find any information on Intel's own site. Intel announces next month. IBM just announced the EX platform X6. quote:24 DIMMs per socket is so sexy for per-core licensed database servers if these CPUs aren't ridiculously expensive. Per-CPU pricing should be comparable to existing E7s.
|
# ? Jan 22, 2014 06:46 |
|
Anandtech has some new details about Haswell Intel released at International Solid-State Circuits Conference 2014. The most interesting bits are die size and transistor counts for various Haswell configurations, as well as info about the OPIO interface used to connect eDRAM or the PCH. TechReport has coverage of Intel's low-power graphics work discussed at ISSCC. This core is optimized for near-threshold voltage operation, which is around the minimum voltage required for the transistor to flip at any frequency. This means this GPU is optimized for ultra-low clockspeeds, though it can probably scale up to near the frequencies we're used to under load. This would make it well-suited for continuous low load, like a live desktop background on an otherwise idle mobile device.
|
# ? Feb 9, 2014 23:16 |
|
Anyone care to guess what this ~10 year old top tier mobile CPU can handle? http://ark.intel.com/products/27596/Intel-Pentium-M-Processor-765-2M-Cache-2_10-GHz-400-MHz-FSB Supposedly it was $600+ when it came out in 2004. 1080p youtube? 1080p H.264 playback? Paired with an Intel 855GM GPU/chipset. I'd test myself but I won't have the notebook for a few weeks. It was my grad school notebook Shaocaholica fucked around with this message at 03:06 on Feb 10, 2014 |
# ? Feb 10, 2014 03:04 |
|
I don't know, and very doubtful. I had a 745 at 1.8 GHz, it wasn't close to H.264, but the software could have improved. And.... I have a 1.7 GHz 735 right now. The thing's at 86% to play a 360p YouTube video in Chrome on linux, but it starts going real slow if you full screne it. Let's try the real flash player. The real flash player can play a 360p and fullscreen it, which makes sense because it could do that before. It can do 480p too, smoothly. It starts struggling at 720p. There's no hope for 1080p YouTube. Boy I sure do miss 1400x1050 14" screens.
|
# ? Feb 10, 2014 03:42 |
|
|
# ? Apr 24, 2024 19:20 |
|
Hmm. Back in school I was able to play back 720p offline h.264 on it although they were circa 2006 720p encodes.
|
# ? Feb 10, 2014 04:20 |