Slashdot Mirror
Intel Removes "Free" Overclocking From Standard Haswell CPUs
crookedvulture writes "With its Sandy Bridge and Ivy Bridge processors, Intel allowed standard Core i5 and i7 CPUs to be overclocked by up to 400MHz using Turbo multipliers. Reaching for higher speeds required pricier K-series chips, but everyone got access to a little "free" clock headroom. Haswell isn't quite so accommodating. Intel has disabled limited multiplier control for non-K CPUs, effectively limiting overclocking to the Core i7-4770K and i5-4670K. Those chips cost $20-30 more than their standard counterparts, and surprisingly, they're missing a few features. The K-series parts lack the support for transactional memory extensions and VT-d device virtualization included with standard Haswell CPUs. PC enthusiasts now have to choose between overclocking and support for certain features even when purchasing premium Intel processors. AMD also has overclocking-friendly K-series parts, but it offers more models at lower prices, and it doesn't remove features available on standard CPUs."
AMD also has overclocking-friendly K-series parts, but it offers more models at lower prices, and it doesn't remove features available on standard CPUs.
It is also significantly slower buck for buck in real life workloads.
That's because they are not number one. Like Avis, they have to try harder.
“He’s not deformed, he’s just drunk!”
Is there anyone besides a small group of people who benefit from higher clock rates? Most people I know would pick battery life over performance on mobile devices. Desktops have been "powerful enough" for at least the past 5 years. Is it just about bragging rights at this point?
Now that AMD is no longer a threat to them, they can go back to their old tricks again.
Guh. Premium, not primium! And annecessary = unnecessary. I suck.
Well, "free" clock headroom aside, Intel removing features from the K series parts (VT-d, etc.) has been going on since Sandy Bridge I believe. Basically, if you want the best of both worlds you will want to invest in an Extreme Edition processor. As quick search on ark will show, the 3770K does not have VT-d while the 3930K does.
-Reed
This is why AMD can not die just think of what intel will do with out AMD in the market.
I've never found overclocking to be worth the trouble. Anytime there's a stability issue with an overclocked PC, there's always that nagging doubt that all my troubleshooting is for naught, because it was a fluke bit fail due to the overclocking. Life's too short- skip the anxiety and run your processor at it's rated speed.
If Intel will ever be allowed to become monopoly again, it will produce extremely pricey and extremely limited processors. Everybody should love AMD, because it is the only thing stopping Intel from selling them shit wrapped in golden paper for thousands $$.
They will have to deal with the ARMs market than?
The K-series parts lack the support for transactional memory extensions and VT-d device virtualization
Yeah, well, fun fact... a lot of enthusiasts like myself like things like VMWare, which depend on this kind of thing. Deleting those features from the unlocked line means I just won't buy them... one of the big drivers for overclocking is to run virtualization. You might think it's "just gamers" doing this, but a lot of us do network and system administration and deployment and like the ability of having a "lab in a box" offered by current processors. You take that away and you're going to find your bottom line hurting, possibly more than a little.
I don't know which of your marketing assclowns came up with this idea as a revenue generating measure, but it's going to backfire in their face and I hope when it does you fire their ass, apologize, and never try this again. You're only succeeding in driving us towards commodity hardware like AMDs offerings... All they need to capitalize on the market you've just shit on now is offer mainboards with multiple sockets for their CPUs and make the mainboards cheap and the core system very energy efficient... and not only will the enthusiasts ditch you, but so will the data centers...
You're opening a can of worms here. Bad plan, darlings.
#fuckbeta #iamslashdot #dicemustdie
Current K rated CPU lose this and possibly some other features. I didn't pay attention to this and found out the hard way when I couldn't run an overclocked ESX-i Sandy machine. Pissed is an understatement! There's no good reason to do this other than to screw with the marketplace.
I've switched to a XEON CPU of Ivy heritage and GL finding a board for one of those that runs ESX-i and can be overclocked. Nearly every machine I own is overclocked and has been for many years and it pisses me off to get jerked around like this by Intel.
Build it, Drive it, Improve it! Hybridz.org
I'd imagine nobody codes for this. [TSE]
That is going to be an important feature when programmers eventually leverage it. Hardware assisted optimistic locking can make concurrency easier, safer and more efficient as the CPU takes care of coherency problems usually left to the programmer and CAS instructions. Imagine being able to give each of thousands or millions of actors in a simulation their own independent execution context (instruction pointer, stack, etc.,) all safely sharing state and interacting with each other using simple, bug free logic, as opposed to explicit and error prone locking and synchronization. This has been done with software transactional memory but it frequently fails to scale due to lock contention. Hardware based TM can prevent that contention by avoiding lock writes.
It is extremely cool that Intel is implementing this on x86.
Lurking at the bottom of the gravity well, getting old
Well, I would, for one. Unless you're using Xen or HyperV, VT-d doesn't really benefit you.
I'm starting to think GNU is the problem with "GNU/Linux" these days.
Yes, I remember the good ol' days when you can get a $100 CPU and make it work like a $800 one. I remember in particular the days of buying a cheap Celeron and having it perform like much more expensive Pentium II or even P3.
And I also remember days of headaches with stability issues, over heating and other stupid problems all to squeeze a few extra FPS out of Doom.
Nobody overclocks anymore, and if they do, it like getting a trophy for trolling a blog. Its completely unnecessary and doesn't really offer anything except a feel good, slap on the thy own back when you see your completely arbitrary and virtual benchmark numbers rise up while you ruin your CPU.
What needs the extra performance these days? You need to Tweet faster? Like on Facebook faster? Browse a website factions of milliseconds faster?
Games used to drive overclocking but GPU's are where game performance lies these days. Sure maybe overclocking your CPU by 50% might offer 1% more FPS, but who the fuck really cares, nobody with a life that is.
Intel realizes that the enthusiast market for PC's has nose dived and its obviously cheaper to produce CPU's where you don't have to worry about the kind of performance tolerances that are required for overclocking.
And I don't think "enterprise" level developers are buying cheap computers and then overclocking to get better VM performance. I mean really? If you consider yourself an "enterprise" developer then get the "enterprise" to buy you a decent workstation or VM server. I don't think your "enterprise" wants you to spend days trying to optimize performance on your workstation, I'd fire anybody that wastes any amount of time in a BIOS.
I would say Intel should focus on offering one "enthusiast" level CPU that is completely unlocked for overclocking. I mean if people want to burn out their CPU repeatedly its more money from a market segment that is drying up, but I think in general Intel or any CPU company should not have to worry about providing overclockable CPU's across their product line.
The bottom line is that benchmarks aside, if you ever looked at your Task Manager you'd probably realize that your CPU is idling at 1% usage 99% of the time, so you want to make the System Idle task run faster? I don't get it anymore.
I haven't thought of anything clever to put here, but then again most of you haven't either.
There's a big difference between VT and VT-d. Intel is only disabling VT-d (aka Directed IO) in the processors.
It is an I/O passthrough to a virtual machine (allowing a virtual machine to directly access the IO bus instead of passing through the hypervisor). Most people won't use anything like this and it's primary only found in enterprise class bare-metal hypervisors like VMWare ESXi, so it honestly doesn't have any impact on workstations running VMWare Workstation in 99.99% of situations.
From Intel:
"VT-d" stands for "Intel Virtualization Technology for Directed I/O". The relationship between VT and VT-d is that the former is an "umbrella" term referring to all Intel virtualization technologies and the latter is a particular solution within a suite of solutions under this umbrella.
The overall concept behind VT-d is hardware support for isolating and restricting device accesses to the owner of the partition managing the device.
AMD didn't come out of nowhere, they were making 8088's in 1975.
Sorry, teleporters just kill you and then make a copy. A perfect, soul-less copy.
Notice that VT-d is disabled, not VT. VT-d allows a hardware device to be passed directly from the hypervisor to a virtual machine (such as a video card). This is only used in HypverV, Xen, and (I think) VMWare ESX, none of which are desktop products. I use VMWare Workstation and Virtualbox quite often (although I'm warming up to KVM) on both AMD and Intel, with no ill effects from either side. Please be informed about what you're saying Intel is screwing us on, and you'll see that 90% of the people that use these features aren't even effected.
I'm starting to think GNU is the problem with "GNU/Linux" these days.
or someone that thinks you should be able to get overclock support and virtualization support without playing these market segmentation games.
AMD has superior FP capabilities. In both CAD and CAE benchmarks, honors always go to AMD for the math. But what really hit me as a big-ole liar fanboi comment was the one about CAD rendering. The majority of that is not related to your CPU, but your GPU. The portion of GPU that is CPU related still benefits from AMD chips which have the memory at the front end of the chip, compared to the Intel that has the memory pipeline as far back as possible in order to claim "we have more MHz than AMD".
Video compression really depends on who's chip the code has been modified for (if any). As with CAE math, native chip math functions are much faster on AMD.
I run annual benchmarks inside companies for Intel vs. AMD and have for over a decade. These benchmarks show real world performance of Unigraphics, CATIA, HyperMesh, MSC Patran, Ansys, and Muses. CATIA and Ansys are always the worst on AMD, as they have both been assimilated by DirectX over OpenGL with no option to force OpenGL. They still however slightly favor AMD over Intel.
I don't rely on Tom's hardware or someone else for opinion, since Tom's showed us long ago that you can't trust "independent" benchmarks for much. I have read benchmark reports from others that indicate the opposite, but have yet to have anyone recreate their results for me. I use real decks and models from real products, I don't use code exercising a subset of CPU instructions as fast as it possibly can.
-The wise argue that there are few absolutes, the fool argues that there are no probabilities.
The problem is that people buying K parts and building PCs around them are pc enthusiasts.
Is my gaming desktop going to do double duty as a production Xen server? Of -course- not. At least not at the same time.
But if I look around my home office, the cpu's that used to be in my gaming PCs ... one is in a Xen server that I'm using actively. And another is a vmware server.
But as I use both xen and vmware for work, having these 'toy' servers at home has been helpful for learning, and experimenting. I definitely want cpus that support these technologies. I expect I'll build a hyper-V unit sooner than later too.
The only question i have about intel's move is "why" is this some sort of misguided marketing nonsense, or do these features perhaps interfere with the overclockability of the K cpus. Maybe transactional memory and hardware virtualization don't over clock well ? If that' the case... I get it.
Otherwise, I'm completely stumped as to why intel is removing it.
It could become a feature once mainstream, as it dramatically helps thread scaling. I will agree that it is not much of a current selling point and may not become one for a long while, depending on threaded programming up-take or OS advantages.
Just to add a couple options.
As far as I know it's not available in VMware Workstation.
More cores are useful if, and only if, you have software threaded out enough to use it. Some workloads are, many are not. This "OMG moar cores lol," attitude is silly, and to me reeks of fanboyism. "My chosen holy grail platform does this, therefore everyone should want it!"
Also more cores aren't necessarily useful if things over all are too much slower. For example, you'd expect a T1100 to be faster than a 2600 at x264 encoding. I mean it is all kinds of multi-threaded, and the T1100 has 50% more cores. Maybe the FX-8350 too. While it isn't 6 core, it does have 8 modules so 8 threads.
Well, the reality it that they are not (http://www.anandtech.com/bench/CPU/27). The T1100 and FX-8350 are behind pretty much all modern Intel CPUs. An i5-2400 beats them out. Despite the core advantage, the speed disadvantage per core is too much.
But go ahead and keep telling yourself that you are the only TRUE kind of computer user because you care more about cores than actual performance.
I was looking into upgrading my system when the Haswell CPUs came out, and I was disappointed. Then I ordered a socket 2011 motherboard with 4 full-length PCIe slots and quad-channel DDR3. It ended up being about $100 more than a comparable Haswell Z87 chipset build, with a faster (MHz) cpu.
I got the (sandybridge-E) core i7 3820 quad core for $249, which is 3.6GHz stock, 3.9GHz turbo. Overclockers have pushed it to 5.5GHz and it is not an "unlocked" K-series cpu. Socket 2011 allows for "old school" base-clock overclocking.
AMD. We're not as big a dicks as the other guys!*
*(But we are trying out one of those "pumps"...)
You have the right to remain sentient. If you give up the right to remain sentient, you will be elected to public office
Windows 8 Client Hyper-V REQUIRES VT-d. Otherwise there is no first-party VM solution for Windows 8 and you're have to install VirtualBox or WMware. Windows 8 doesn't default to installing into Hyper-V when the requirements are met as the parent suggests. Hyper-V is a feature that needs to be installed on all machines. Once installed then Windows 8 boots the hypervisor first then boots Win8 from the drive as a highly privileged VM. Performance for most things is near where it would be if the OS was on bare metal(thanks to the required VT-d instructions). "Host"(i.e. that highly privileged VM you boot to) 3D game performance does take a noticeable hit however even with no other VMs running so I leave the Client Hyper-V turned off most of the time. I'm guessing that Intel knows this and figures that overclockers won't give a shit about running type-1 hypervisors on their gaming desktops. Still a dick move though. Microsoft should get pissed at them but I doubt that would matter anymore or at least not as much as it used to.
Let me make this very clear: back in the days of the Athlon versus the Pentium IV, Intel had the disadvantage because the damn thing was designed primarily for SSE2, and they had a decode imbalance in the design. The Athlon had 3 x87 FPU pipes which made it superior despite the P4's faster clock...but once developers targeted SSE2, the Pentium IV matched the Athlon in FPU, and outclassed it on ALU operations (since both chips had dual 64-bit SSE2 units).
With the introduction of the Core 2, Intel switched to a 4-wide decode and DUAL 128-bit SSE2 units, allowing 2 instruction / cycle throughput, TOASTING the Athlon 64 in all matters of performance. Almost two years later AMD countered with Barcelona, which also had dual 128-bit SSE2 units, but was castrated by their 3-wide decoder. It was a match for Core 2 at the same clocks, but they couldn't match the clocks Intel had.
With the new Core series of chips, and the reintroduction of Hyperthreading, Intel wiped the floor with AMD in anything multithreaded, and they steadily increased single-threaded performance with each new iteration. Dual AVX 256-bit units in Sandy Bridge also potentially DOUBLED Intel's FP throughput. At the same time, AMD moved away from FP performance with Bulldozer, which shared dual 128-bit AVX execution units between two cores. Even with twice the cores AMD still lagged behind in peak FPU throughput, because the shared decode units meant roughly two-wide decode when all cores were heavily-loaded.
So today AMD is not the destination for high FPU throughput, and they really have not been for a decade. I really cannot understand your claims to the contrary.
Man is the animal that laughs.
And occasionally whores for Karma.
First, Bulldozer was not a high performance chip, and never intended to be a high performance chip. It was meant to be a PC based equivalent of a Niagra capable of massive threading. So let's compare apples to apples shall we?
AMD Still considers the Athalon to be the performance chip. Comparing apples to apples, maybe you are asking how a chip rated 300MHz lower be faster? First, the length of the bus needed to get from inbound to FPU is much longer on Intel. Cache is much larger on AMD, prefetch is superior especially for FP instructions. That has a lot to do with the bigger caches. Next, memory is also closer to the front of the chip. Most FPU based apps are also memory intensive. An Intel operation would start at the front of the chip and move to the back. Every memory or FPU operation requires traversing the full chip bus, then the same long ride back. That movement is not required in the AMD design, and that efficiency does make a difference.
If Intel had really doubled FPU performance AMD would have been out of business. Yet they are not, and I can still get exceptional performance off of them for heavy FPU loads. I/O and integer based, Intel beats them pretty solidly and has for quite a long time.
-The wise argue that there are few absolutes, the fool argues that there are no probabilities.
I'm glad to hear AMD was selling 8088's before Intel developed them.
Please, pretty please let me see your 8088 based machine from 1975.
Damn am I getting that old!
VT-d is not only for servers. I found it's use because of my countless cycles of attempts to dual-boot windows and linux (as in I eventually ended using just windows...repeat afte 6 months).
Now I boot linux, do the web browsing and stuff, but when I want to play, I just start my VM and play.
Linux: i5-2500 IGP
Windows: Radeon 7950 (started with 5850)
My over 80 hours of Skyrim are Xen exclusive. DeusEx HR was maybe 20-30h native, followed by more than 50h in VM.
This is my original post (closed since then): http://www.tomshardware.co.uk/forum/336186-33-full-gaming-virtual-machine
This is another thread that I joined and posted some benchmarks: http://hardforum.com/showpost.php?p=1039531303&postcount=27