|
AnandTech posted an update on the Clover Trail+ Atom SoC for high-end smartphones and midrange tablets. It's probably the last time we'll see this old vintage 2008 Atom core as the flagship of the low-power product line. Short version: Two Atom cores (instead of Medfield's one) and a more powerful GPU, competitive with the one in the Apple A5X and A6. Similar single-threaded performance, similar light-load power use, updated memory controller, up to 1080p displays, a firmware re-write for the image processor, and a (much needed, still slightly embarassing) baseband update. The article really gives the context very well: quote:Final Words
|
#
?
Feb 25, 2013 09:46
|
|
|
|
| # ? Apr 26, 2024 19:53 |
|
|
Am I the only one who's underwhelmed with the specs of the Haswell SKUs so far? I'm looking on CPU World, and I just don't see much (if any) difference in TDP or clocks of Haswell compared to Ivy. Of course we'll have to wait until release to see the real difference in power consumption and performance, but at least right now, I feel like it's kind of pointless to wait for Haswell if you really need a new machine.
|
|
#
?
Feb 25, 2013 16:38
|
|
|
I want haswell simply for the integrated GPU. Someday i will have my dream of an ~11" notebook with a 2650x1600 screen 6+ hours of battery life and the ability to run most games at decent settings at 1080p.
|
|
#
?
Feb 25, 2013 16:54
|
|
|
My Rhythmic Crotch posted:Am I the only one who's underwhelmed with the specs of the Haswell SKUs so far? I'm looking on CPU World, and I just don't see much (if any) difference in TDP or clocks of Haswell compared to Ivy. Assuming we're looking at this, I see a 35W quad core i7, and that's pretty neat. But frequency was never really going to go anywhere, because for the past few product cycles it's been about building CPUs wider and more powerful isoclock, not with a higher clock rate. And the TDP - you have to remember that this is the same node as Ivy Bridge, 22nm, and the TDP is nearly the same even while the chip itself is getting bigger and the GPU is getting tons beefier and the package is absorbing a bunch of VRM bits. 14nm will take TDPs back down quite a bit.
|
|
#
?
Feb 25, 2013 16:57
|
|
|
I was under the impression that the TDP will remain the same but there will be very substantial increases in battery life and graphics.
|
|
#
?
Feb 25, 2013 17:03
|
|
|
High-end desktop parts will (probably) have a 84W TDP and midrange desktop will go from 55W back to 65W, mostly because of the VRM and larger die size. If you're running the chip at 100%, the total power used will be higher. Performance per watt will be higher regardless, though, because of the general 10% IPC increase and, if using AVX2 and other new instructions, very large throughput increases in floating point and encryption math. Haswell's power optimizations will absolutely smash idle power and power-per-job in light loads, though, resulting in a general increase in battery life for anyone who isn't, say, constantly transcoding video on battery power. And if manufacturers equip their screens with framebuffer DRAM, that can take advantage of the IGP's S0ix idle and let the GPU idle whenever the screen image is static, further helping battery life. This is kinda the trend in low-power computing right now - build chips with low average power and high TDPs. When you need a lot of oomph, use power gating and turbo and OS process scheduling to blast through things in one high-power burst of compute, then return to power-gated idle. It's a lot more power-efficient than keeping a lower-power, slower CPU running all the time. That's one of the under-the-hood updates in Windows 8, for example:  That said, the increase in TDP for the VRM sections will be offset at least somewhat by the reduction in VRM circuitry on the motherboard.
|
|
#
?
Feb 25, 2013 17:17
|
|
|
Misogynist posted:I think this is the same thing that everyone was saying back around 2007 when the consensus was that they would never support Microsoft Exchange on iPhone. bull3964 posted:Someday i will have my dream of an ~11" notebook with a 2650x1600 screen 6+ hours of battery life and the ability to run most games at decent settings at 1080p.
|
|
#
?
Feb 25, 2013 18:20
|
|
|
The enhancements in power management do look interesting, I will give Intel that. However if we compare this Ivy Bridge part: Core i7-3770T / 4 core / 8 thread / 2.5 GHz / 3.7 GHz / 8 MB cache / 45 Watt To this Haswell part: Core i7-4765T / 4 core / 8 thread / 2 GHz / 3 GHz / 8 MB cache / 35 Watt Those two are the most direct comparison I could find. It just doesn't look that compelling. However there might be other comparisons that are more favorable towards holding on for Haswell.
|
|
#
?
Feb 25, 2013 20:35
|
|
|
My Rhythmic Crotch posted:The enhancements in power management do look interesting, I will give Intel that. However if we compare this Ivy Bridge part: Edit: The Haswell 4770T has 10-15% higher CPU performance than the Ivy Bridge 3770T with ~8W lower power usage, and on a 45W processor that is not trivial. This does not consider graphics or anything else so may be misleading until we get a real review. Alereon fucked around with this message at 20:54 on Feb 25, 2013 |
|
#
?
Feb 25, 2013 20:48
|
|
|
Gains through efficiency are where its at
|
|
#
?
Feb 25, 2013 21:17
|
|
|
A defect has reportedly been identified in Haswell processors that prevents USB 3.0 devices from functioning after waking from S3 sleep mode. This will require a new revision (stepping) of the processor to correct and there is not time to fix this before release, meaning Haswell will launch with this bug. Intel is requiring OEMs buying processors to sign off that they accept the defect and will work around it, which may mean that Haswell systems are stuck with third-party USB 3.0 controllers at launch.
|
|
#
?
Mar 4, 2013 21:16
|
|
|
Alereon posted:A defect has reportedly been identified in Haswell processors that prevents USB 3.0 devices from functioning after waking from S3 sleep mode. This will require a new revision (stepping) of the processor to correct and there is not time to fix this before release, meaning Haswell will launch with this bug. Intel is requiring OEMs buying processors to sign off that they accept the defect and will work around it, which may mean that Haswell systems are stuck with third-party USB 3.0 controllers at launch. You would think Intel would have learned after the FDIV bug that releasing this processor is going to piss off a lot of people.
|
|
#
?
Mar 4, 2013 21:19
|
|
|
Goon Matchmaker posted:You would think Intel would have learned after the FDIV bug that releasing this processor is going to piss off a lot of people.
|
|
#
?
Mar 4, 2013 21:45
|
|
|
Unfortunately, that's probably the only place they are, the dual-chip SoC version of Haswell with the on-package PCH.
|
|
#
?
Mar 4, 2013 21:55
|
|
|
Quick question: do higher-binned chips tend to overclock better? 3570k vs the 3770k being the case in point.
|
|
#
?
Mar 8, 2013 03:58
|
|
|
The i5-3570K and i5-3770K are already top-binned parts. They overclock pretty much identically.
|
|
#
?
Mar 8, 2013 04:03
|
|
|
A slight improvement in the news for Intel: Multiple sources are reporting that the Haswell USB3.0 defect is in the chipset, not CPU. This is better than a CPU flaw, but still requires a new hardware revision to correct. However, waiting for Rev2 motherboards like we did with Sandy Bridge is a lot better than waiting for new CPUs.
|
|
#
?
Mar 13, 2013 21:12
|
|
|
Alereon posted:A slight improvement in the news for Intel: Multiple sources are reporting that the Haswell USB3.0 defect is in the chipset, not CPU. This is better than a CPU flaw, but still requires a new hardware revision to correct. However, waiting for Rev2 motherboards like we did with Sandy Bridge is a lot better than waiting for new CPUs. I wasn't keeping up to date when there was the SATA(?) bug with the Sandy Bridge chipset; did the laptops get delayed because of the bug?
|
|
#
?
Mar 14, 2013 06:38
|
|
|
unpronounceable posted:How will this affect laptops? If I were updating my desktop, I'd probably be alright, as I never put it to sleep, but I put my laptop to sleep constantly. Yes they were.
|
|
#
?
Mar 17, 2013 18:15
|
|
|
New benchmarks of Haswell from Tom's Hardware Performance seems like 10%-ish above Ivy, desktop GPU performance is still below Trinity. Wonder how the more powerful mobile Haswell GPU will fare.
|
|
#
?
Mar 18, 2013 17:50
|
|
|
Interesting that the desktop lineups only include GT2. I guess GT3/GT3e will be mobile-only? I hope someone makes those available on mini-ITX boards or NUC-likes for HTPC use. Also, how can they possibly release this information without someone somewhere violating the hell out of an NDA? Also also, what loving bullshit that the K variants don't have TSX.
|
|
#
?
Mar 18, 2013 17:58
|
|
|
Factory Factory posted:Also also, what loving bullshit that the K variants don't have TSX. K variant Ivys and I think Sandys didn't have vt-d either, Intel just likes loving with us I guess. Also note the odd i7 with no hyperthreading and a dual core i5, just to ruin my sperging over a nice and consistent naming scheme.
|
|
#
?
Mar 18, 2013 18:33
|
|
|
Not to mention the Pentium G850, a dual-core non-HT processor with HD 2000 graphics, the only one in the Pentium line to do so. It's basically i3 performance at 2/3 of the price. Around the office we're pretty sure that it's just an i5 with two dead cores. I really hope we see that trend continuing.
|
|
#
?
Mar 18, 2013 18:39
|
|
|
If I am coming from a dual core Wolfdale chip, would it behoove me to wait it out until whatever the i5k SKU for Haswell is or should I just go with Ivy? I occasionally Photoshop and do some gaming, but I don't really do anything like virtualization or the like.
|
|
#
?
Mar 18, 2013 19:04
|
|
|
I'm curious why there's not more info and "leaks" about upcoming mainboards already.
|
|
#
?
Mar 18, 2013 19:34
|
|
|
KKKLIP ART posted:If I am coming from a dual core Wolfdale chip, would it behoove me to wait it out until whatever the i5k SKU for Haswell is or should I just go with Ivy? I occasionally Photoshop and do some gaming, but I don't really do anything like virtualization or the like. Combat Pretzel posted:I'm curious why there's not more info and "leaks" about upcoming mainboards already.
|
|
#
?
Mar 18, 2013 20:05
|
|
|
Alereon posted:If you are overclocking or think you might use any of the new processor features (or are in a power-limited application), then Haswell will be a very significant upgrade. What's the advantage to overclocking? Just undoing the Ivy Bridge issues?
|
|
#
?
Mar 18, 2013 20:12
|
|
|
What Ivy Bridge issues? Overclocking is just making the chip run at a faster rate, i.e. giving it more GHz.
|
|
#
?
Mar 18, 2013 20:16
|
|
|
evilweasel posted:What's the advantage to overclocking? Just undoing the Ivy Bridge issues? Apparently Intel will enable changing DMICLK/BCLK ratios. http://www.xbitlabs.com/news/cpu/display/20120919160307_Intel_Haswell_Processors_to_Further_Improve_Overclocking.html
|
|
#
?
Mar 18, 2013 20:27
|
|
|
Factory Factory posted:What Ivy Bridge issues? Ivy Bridge overclocked worse than Sandy did thanks to the way the thermal spreader was connected to the die, I thought.
|
|
#
?
Mar 18, 2013 20:33
|
|
|
But there's no evidence that Haswell will change that.
|
|
#
?
Mar 18, 2013 21:25
|
|
|
evilweasel posted:What's the advantage to overclocking? Just undoing the Ivy Bridge issues? Edit: TechReport is also citing a "trusted source" as saying that Haswell has more overclocking headroom than Ivy Bridge. Alereon fucked around with this message at 22:08 on Mar 18, 2013 |
|
#
?
Mar 18, 2013 22:05
|
|
|
Alereon posted:Additionally moving the VRMs on-chip may improve overclocking, since it will provide cleaner power with faster response times, and there will be less variability due to motherboard VRM quality. Wait what, how will overclocking boards differentiate themselves (from regular Z87 boards) now?
|
|
#
?
Mar 18, 2013 22:19
|
|
|
Grim Up North posted:Wait what, how will overclocking boards differentiate themselves (from regular Z87 boards) now? More PCIe and (3rd party) SATA ports and heatsinks that looks like ninja stars!!!!
|
|
#
?
Mar 18, 2013 22:27
|
|
|
I always found it hilarious that the "enthusiast" mobos all had large rows of chokes, and Intel shows that some tiny components on top of the chip just do the same.
|
|
#
?
Mar 18, 2013 22:29
|
|
|
Keep in mind that you will still need VRMs on the motherboard to convert from 12V to CPU voltage, it's the last conversion step to core voltage and the multiple voltage planes that's now happening on the CPU. The big advantage, as I understand it, is you no longer need crazy high-frequency conversion hardware on the motherboard, because you're just feeding the VRMs on the CPU that will actually handle that.
|
|
#
?
Mar 18, 2013 23:26
|
|
|
I can't wait to hear the reasoning as to why TSX is disabled. That absolutely does not seem like an enterprise feature to me, it sounded more like a free performance boost with a recompile for applications with poorly written threaded code. Anyone know when information was supposed to start leaking? From this twitter thread between Ian@Anandtech and FPiednoel@Intel, THG was not authorized to publish this preview. Chuu fucked around with this message at 04:24 on Mar 19, 2013 |
|
#
?
Mar 19, 2013 04:20
|
|
|
Sandy Bridge didn't lift until launch day. Ivy Bridge stuff didn't start lifting until its original launch date (I think it was HD 4000 that was let out then?), and the CPU benchmarks didn't lift until the street date for the desktop chips.
|
|
#
?
Mar 19, 2013 05:19
|
|
|
Alereon posted:I think we can also expect better clockspeed and voltage scaling due to improvements in how the 22nm FinFET process is utilized thanks to the lessons learned from Ivy Bridge. This cuts both ways, though - Ivy Bridge benefits in yields (and thus headroom) from the lessons learned from Sandy Bridge, while the new architecture has new critical paths that aren't yet optimized (this is the beauty of the Tick/Tock cadence). Chuu posted:That absolutely does not seem like an enterprise feature to me, it sounded more like a free performance boost with a recompile for applications with poorly written threaded code. That sounds exactly like an enterprise feature to me. Consumer applications are extremely slow to uptake new processor features, because it generally isn't easy to support the new feature without breaking compatibility with old processor revisions.
|
|
#
?
Mar 19, 2013 22:28
|
|
|
|
| # ? Apr 26, 2024 19:53 |
|
|
I'd like to see some virtualization benchmarks, if there's something like this. The Intel slides mention multiple enhancements to be specified later on (I guess the official unveiling or something?) and doesn't go into detail beyond shorter roundtrip latency. --edit: Actually nevermind, I hope the various hypervisors will take advantage of the stuff listed fast. 
Combat Pretzel fucked around with this message at 22:49 on Mar 19, 2013 |
|
#
?
Mar 19, 2013 22:46
|
|