.

[ShuttleLuv]

.

 

[pelo]

They can still limit the BCLK, though.

This is pure speculation on my part, but it looks like the days of overclocking a locked CPU may be upon us again. By being able to manipulate the Uncore frequency separately from the CPU (and GPU) frequencies, there is a good chance that the lower end Haswell chips will become viable overclocking targets for enthusiasts once again. We may never see another 300a, but if we can have a $100 Haswell CPU running at similar frequencies to the $300 part, it will renew interest in overclocking for a lot of people.

Given Intel's recent trend of fusing off practically everything in order to produce a billion worthless SKUs, I wouldn't consider that speculation to hold any merit. Remember, he's thinking that Intel will allow a BCLK overclock yet it's the same Intel that yanks out AVX support and even hyperthreading on chips that had to have passed with both otherwise the cores would fail as well. Allowing a heavy BCLK overclock means Intel loses money on higher end chips with multipliers unlocked (something else that's also entirely dependent on fusing on and off), and that's just not a smart business decision. Don't get your hopes up over somebody speculating. Remember, this is the same Intel that charges us $20 for an unlocked multiplier and an extra $100 on top of that to enable the 4 other threads which work just fine... If you want to know what Intel will do when there's no AMD to compete on the high end, you need only look at the 2011 processor prices and the shenanigans they're pulling with fusing off parts of the chip that work just fine.

The decoupling of the L3 cache might also hurt CPU performance a bit depending on the task, but given the other improvements to the core in Haswell it's unlikely to be noticed at all.

- forgot virtualization. They purposely disable virtualization features on the unlocked 1155 models.

 

[Ultima99]

I just want some Haswell-E with quad channel DDR4 please!

 

[Liger88]

I just want some Haswell-E with quad channel DDR4 please!

lol yeah I know what you mean. I've been waiting for Haswell for 2 years. Intel has been prepping this chip since SB and so far the details look like it could be future proof for a long time, unless of course you are waiting for DDR4.

I'm really hoping 16GB, 32GB, and maybe higher memory modules start hitting the mass market by then. I may not need it, but I still want a 256GB system in 4 slots just for shits and giggles.

- forgot virtualization. They purposely disable virtualization features on the unlocked 1155 models.

This STILL royally pisses me off and makes absolutely no sense to me.

 

[Michaelius]

If we can get dual core 5 Ghz Haswell Pentiums for 80$ like in Pentium e 21x0 days it would be game over for AMD in enthusiast community.

 

[Tsumi]

I don't think it's going to happen though. There are other things besides the L3, uncore, and PCI-E bus that run off of the BCLK. The SATA and USB ports also get their signals from the BCLK in current SB/IB, and those don't like to be adjusted either.

 

[pxc]

heh, called it last month

I don't remember if AT mentioned it, but L3 is decoupled from the core clock and voltage domains, so that's one less thing to hold back overclocked frequency. The AT preview is lengthy, so that's just a teasing comment.

 

[drescherjm]

forgot virtualization. They purposely disable virtualization features on the unlocked 1155 models.

This STILL royally pisses me off and makes absolutely no sense to me.

I have not problem with that at all. If you are using VT-D you should not be overclocking. I guess for GPU virtulization this could be okay. But certainly not for a file server even if you are using zfs or btrfs.

 

[colinstu]

VT-x isn't disabled (the damn important one), only VT-d is (important only if you're using the features of it).

Intel's solution: don't get LGA1155/1150 chips that end with -k if you want vt-d. That still leaves you with non-k chips, LGA2011 chips (which you can OC), and Xeons.

People sure get their panties in a bunch over such a silly situation. If you didn't want this to be the case well... maybe some competition in the sector would help!

Haswell is going to be the next Sandy Bridge (performance bump wise) so everyone should be happy. The doom and gloom about BGA chips, the PC is dying, Intel purportedly locking everything down because this happens every time they release something new... ugh I can't stand it.

 

[LinkinMcOwnage]

I have not problem with that at all. If you are using VT-D you should not be overclocking. I guess for GPU virtulization this could be okay. But certainly not for a file server even if you are using zfs or btrfs.

According to who? Overclockers have a tendency to think "damn the rules, this is what I want"

 

[ShuttleLuv]

Eh ^^^ I have to agree most overclockers who are serious do disable stuff like that

 

[bwang]

Overclockers have a tendency to think "damn the rules, this is what I want"

^this
we're already violating specs and warranties by overclocking, but a well-OC'ed machine is pretty damn stable. My sig rig has had uptimes of several months, and it's running on a platform that's a PITA to stabilize.

 

[ShuttleLuv]

^this
we're already violating specs and warranties by overclocking, but a well-OC'ed machine is pretty damn stable. My sig rig has had uptimes of several months, and it's running on a platform that's a PITA to stabilize.

Yes true but sometimes (and I'm saying over the years up til now and beyond) there's been instances where disabling a feature helps an overclock. But if your chip is stable and you feel it's 100% stable and want to use virtualization yeah it wouldn't be a bad thing.

 

[jwcalla]

But e.g. the difference between an i5-3570 and an i5-3570K is not just VT-d and overclocking. The 3570K has the HD4000 graphics. So you might not necessarily prefer the K for the overclocking. I guess if you want VT-d and HD4000 you have to fish around the i7s.

There probably aren't too many cases where people have these exact needs, but from my POV all the segmentation is just stupid and unnecessary.

 

[Michaelius]

better gpu for K series is super stupid since it's much more likely to be used just as backup for discreet gpu compared to non K variants.

 

[SonDa5]

Sounds like performance of 4770k isn't much better than 3770k so to help out Intel will feature FSB and more ram ratios for more over clocking options.

Have to do something to make the 4770k more appealing to enthusiasts.

 

[Taranabas]

Haswell is going to be the next Sandy Bridge (performance bump wise) so everyone should be happy.

As far as I've read Haswell is going to be another small increment in performance, but a huge increase in efficiency.


As was mentioned, even if intel chooses to uncouple various bclk dependent components it's easy for them to limit adjustment. They won't allow a product to infringe on the next higher product's market space. They made that booboo back in the day with the celeron and haven't made it again since. Even the 2600K, the almighty overclocking legend, doesn't tread into SB-E territory once you include cooling and platform costs. Overclocking remains a hobby (albeit a fun one!), not a value add.

 

[pxc]

As far as I've read Haswell is going to be another small increment in performance, but a huge increase in efficiency.

A 10-15%* increase in IPC is more typical of uarch refreshes, which Haswell is. Separately, Haswell also seems to be more power efficient too**.

Shrinks/tweaks (ticks) usually have minor (3-5%) IPC improvements, not surprising since it's the same basic uarch. Haswell doesn't fit in that category compared to IB.

*Intel has repeated this figure several times. Other sources (anonymous web posters, mostly), have written otherwise. Not sure why there is a dispute with Intel's projections since apparently no one else has benchmarked one.

**this seems to be taking advantage of the "sweet spot" of tri-gate manufacturing at lower voltages. While Intel could crank up desktop Haswell's clock speeds at IB's TDP, it seems to be choosing slight clock speed bumps and using the lower voltages to reduce power. Server processors won't be constrained in the same way. Either way, spend a few dollars more for a K processor and ignore power efficiency.

 

[Tsumi]

Sounds like performance of 4770k isn't much better than 3770k so to help out Intel will feature FSB and more ram ratios for more over clocking options.

Have to do something to make the 4770k more appealing to enthusiasts.

Most likely, it'll probably feature the BCLK strap multiplier for -k CPUs only.

 

[drescherjm]

But if your chip is stable and you feel it's 100% stable and want to use virtualization yeah it wouldn't be a bad thing.

It does not prevent you from using virtualization. That is virtulization of the CPU is what is provided by vt-x. That still works. What does not is vt-d. vt-d is for assigning hardware like raid controllers or GPUs directly to a guest to avoid the emulation penalty.

 

[Michaelius]

Sounds like performance of 4770k isn't much better than 3770k so to help out Intel will feature FSB and more ram ratios for more over clocking options.

Have to do something to make the 4770k more appealing to enthusiasts.

I'm on 4500 Mhz 2500K right now - if Haswell get's 10% IPC increase over IB it will be around 15% higher than SB.

5 Ghz Haswell with 15% IPC increase over SB would be atractive, and it would be even bigger jump to everyone on socket 1366 or 1156.

 

[Tsumi]

Yeah, there are plenty of people who are still on 1156/1366 that haven't bothered upgrading because SB and IB weren't that big of an improvement.

 

[ShuttleLuv]

It does not prevent you from using virtualization. That is virtulization of the CPU is what is provided by vt-x. That still works. What does not is vt-d. vt-d is for assigning hardware like raid controllers or GPUs directly to a guest to avoid the emulation penalty.

OK you must've misunderstood me because I didn't directly mean vt-x affects overclocks but in some circumstances (i'e disabling cache, the way it's read and addressed, or even hyperthreading does help overclocks if you're pushing for that last mhz. My reply was more that overclockers can trust stabilizing their overclocks today so much that stability should not be a concern with virtualization. And yes I am well aware Sandy lacks vt-d.

 

[RamonGTP]

I'm content with multiplier only OCing. If I had to pick ONE, I'd keep multi only around. The simplicity is a bigger factor for me than squeezing out every last MHz out of the CPU.

 

[colinstu]

I'm content with multiplier only OCing. If I had to pick ONE, I'd keep multi only around. The simplicity is a bigger factor for me than squeezing out every last MHz out of the CPU.

I'm assuming Intel will have different cpu "strap" levels. With that there is a little wiggle room too. So you'll have 100mhz, 125mhz, and higher (but those won't be stable). Might be able to increase the 125 strap a little more to 130 but that's it. There really isn't anything else to the process. Multi is locked on the i7-3820 so the only way to oc is to increase the strap, which I've done.

 

[ShuttleLuv]

I'm sure Ramon understands straps his point was just that mutli overclocking (i'e Sandy/Ivy) make things more simple in getting that last mhz.