Is Desktop Haswell a dud? How many are not "upgrading"?

[Enigmoid]

I am talking about the haswell desktop.. not mobile.
I know of all the mobile accomplishments, however were not looking for mobile on desktops.

I say they pulled a bulldozer, because like the Athlon -> Bulldozer migration, were seeing the same results on the intel scale.

We got a slightly faster chip.. nothing to go WOW about... with a LOT more heat... exactly like how bulldozer was when first released.

A long time ago... well not so long time ago... maybe 3 yrs ago.. a lga1160 was identical in most regards to a LGA1366 chip.
Before that LGA771 was identical to LGA775.

Haswell is nowhere near a bulldozer. Bulldozer saw a reduction in IPC and a substantial increase in power use on a die shrink and a new architecture.

Haswell is dissapointing I agree, very dissapointing but nowhere near to a step back for the average consumer.

 

[PCJake]

Only a week ago, I had my entire Z87 build planned out and I couldn't wait to upgrade. Now that I've done the cold, hard research and learned more about Haswell, I just can't bring myself to do it. They left us enthusiasts out in the cold, plain and simple.

I'll still enjoy getting a GTX 780, of course, and my new Seasonic PSU and Corsair case are sweet. But now all I can do is hope that Intel puts out just one enthusiast-oriented CPU in the next year/year-and-a-half. Hopefully the Haswell-E is a pure enthusiast chip, that would make me very happy.

 

[ehume]

To answer OP's question, I am upgrading . . . to a 3770k.

Given IDC's testing, Sandy & Ivy may be in the optimum zone for overclocking. Reducing process size vs increased leakage. It looks like the chips will get hotter from now on as they shrink.

And cooling solutions will not fade into the sunset.

 

[blackened23]

To answer OP's question, I am upgrading . . . to a 3770k.

Given IDC's testing, Sandy & Ivy may be in the optimum zone for overclocking. Reducing process size vs increased leakage. It looks like the chips will get hotter from now on as they shrink.

And cooling solutions will not fade into the sunset.

Personally, I feel you're making a mistake by opting for the 3770k. The 3770k gets very hot as well while overclocking, let's not kid ourselves - you can easily get 90C temperatures at 4.6ghz. It may be very slightly lower than haswell but make no mistake - there is a definite thermal limit with IVB as well.

IMO, I think the IPC increase with haswell will offset any potentially lower overclock - I went from a 2600k to a 3770k and the 3770k scores better in all synthetics despite having a 300mhz lower overclock. I expect the same of Haswell. Aside from this, Z87 is just a much better platform than Z77 is - having only 2 native intel SATA 6G ports is just completely frustrating and WILL limit you in terms of RAID setups and SSD caching. Z87 is just better all around, and like I mentioned earlier the Haswell should be the same or better in performance even IF you get unlucky with your overclock. Again - overclocking with IVB isn't all roses compared to Haswell. It is a lot of the same thing, it DOES get hot as well; I can attest to this with my own 3770k.

 

[poohbear]

Well this is what happens when there's no competition in the desktop space. Intel is competing in the mobile space (tablets) where there is plenty of competition and business to be had. Nobody buys desktops anymore, its a dead market.

- i5/i7 "K" owners - "K" chips don't support TSX-NI, vPro or VT-d extensions for no technical reason whatsoever

i wasn't aware of this. Are u saying the "k" series don't have the extra instruction sets that are supposed to distinguish Haswell from IB & SB? I thought they were the saving grace of Haswell for the "future", or are those different instruction sets?

 

[TuxDave]

Well this is what happens when there's no competition in the desktop space. Intel is competing in the mobile space (tablets) where there is plenty of competition and business to be had. Nobody buys desktops anymore, its a dead market.

i wasn't aware of this. Are u saying the "k" series don't have the extra instruction sets that are supposed to distinguish Haswell from IB & SB? I thought they were the saving grace of Haswell for the "future", or are those different instruction sets?

There's been quite a large amount of misinformation being tossed around on the forums so I don't blame you for being confused. SOME of the extra instruction sets are not available on the K series. AVX2 is there along with the FMA instructions. Those are the two big performance boosting ones. TSX is not there. It's not clear if quad core plebeians like you and me would benefit from them.

 

[Revolution 11]

There's been quite a large amount of misinformation being tossed around on the forums so I don't blame you for being confused. SOME of the extra instruction sets are not available on the K series. AVX2 is there along with the FMA instructions. Those are the two big performance boosting ones. TSX is not there. It's not clear if quad core plebeians like you and me would benefit from them.

Is it not clear because there are no benchmark results/data yet for TSX or because quad-core users will most likely not be using the applications that require or will benefit from TSX instructions?

 

[Edrick]

Is it not clear because there are no benchmark results/data yet for TSX or because quad-core users will most likely not be using the applications that require or will benefit from TSX instructions?

Give it time, they will come.:sneaky:

 

[Evilviking]

I5 750 @ 4ghz here. I think I'm going to wait it out

 

[Arkaign]

I have the feeling that by the time the new instruction sets are commonly used in games Haswell will be a gen or three old.

Haswell is a fantastic mobile part. At the same time, for the desktop overclocker with a discrete GPU, it's pretty much zero improvement. If they had simply went back to solder over the TIM paste, I'd wager that we'd have a much warmer enthusiast response. 4.8Ghz air overclocks on 4770K at reasonable temps without throttling would be very very well received in combination with the light IPC bump. The problem is that my 2700K runs at 5Ghz at reasonable temps with merely a 212+, which is like a $20 heatsink. I run it with an H100, but that was a purchase intended to cool my 3770K, of which I went through about a half-dozen before permanently going back to 2700K for my 1155 setup.

IDC I'm sure will show us the true potential of Haswell when delidded. It's just not something I want to personally do. The crappy TIM situation was enough to make 3770K worthless to me. Love my Ivy notebook though Will probably go Haswell notebook before long.

 

[Arkaign]

I5 750 @ 4ghz here. I think I'm going to wait it out

Welcome to AT!

Yes, unless you have a godly GPU, you're probably GPU-limited in the vast majority of gaming situations anyway. Need USB3? PCIe card, boom. Need Sata 6.0Gbit? PCIe card, boom. Pretty easy to take all i3/i5/i7 systems and make them really strong with OC + SSD + needed components.

Say you're a gamer, who already has that i5 750 @ 4Ghz, but has only 4GB of ram and a 5870. Now say you have a budget of $600 to make things better. If you decided to do it with Haswell, well you'd blow almost all the $ just on the board and chip even with an i5 non-K. All to stick with a 5870.

Now take that same $600, get a 7950 OC Edition + 650W PSU + 16GB DDR3 + SSD, and upgrade your current PC. That system would kick the ever-living crap out of a barebones Haswell + 5870 upgrade.

 

[Sheep221]

It is not what you expected, but also you didnt get it promised. But then again I cant be bothered less when I know my hardware gets relevant for another more years. I dont get all this rant, people used to spend money on the upgrading just for the sake of upgrading, as none of the complaints claims their computer is very slow for what they are doing and they are desperately awaiting arrival of faster ones.

 

[WhoBeDaPlaya]

I nearly chickened out and had bought two 3770K/Extreme6 combos at MC before Sunday (in case they took off the discount).

Glad I went with the 4770K. Only ended up costing me ~$80 more per setup (with a Z87-GD65, would have been $32 less with the G45) and I managed to eke out about what I would expect of a 3770K (4.6GHz).

So ~8% higher average performance, newer platform and a chance to play with something new - it's all good
Of course, a couple of months from now when a D0 stepping is introduced, I'll look like the suckers who bought C0 i7 920s

 

[Arkaign]

^ Ah, you could always go delidded like a boss, per IDC

 

[WhoBeDaPlaya]

^ Ah, you could always go delidded like a boss, per IDC

I would, but unfortunately...

 

[Arkaign]

LOL ^^

Funniest post in AT CPU history. Infinite reps.

 

[PCJake]

Haswell is a fantastic mobile part. At the same time, for the desktop overclocker with a discrete GPU, it's pretty much zero improvement. If they had simply went back to solder over the TIM paste, I'd wager that we'd have a much warmer enthusiast response. 4.8Ghz air overclocks on 4770K at reasonable temps without throttling would be very very well received in combination with the light IPC bump. The problem is that my 2700K runs at 5Ghz at reasonable temps with merely a 212+, which is like a $20 heatsink. I run it with an H100, but that was a purchase intended to cool my 3770K, of which I went through about a half-dozen before permanently going back to 2700K for my 1155 setup.

IDC I'm sure will show us the true potential of Haswell when delidded. It's just not something I want to personally do. The crappy TIM situation was enough to make 3770K worthless to me. Love my Ivy notebook though Will probably go Haswell notebook before long.

Completely agree. I've had my i7-930 running at 4.2GHz (50% overclock) 24/7 for two years straight. It's 100% stable and the hottest core doesn't go above 75C on air. If I was sure I could get 4.6GHz or so with similar temps on the 4770K, I'd go for it, but I know that ain't gonna happen without taking that lid off.

We all know better than to say Intel didn't achieve what they set out to - certainly, they did. I just don't happen to give a crap about those achievements.

 

[Keromyaou]

To answer OP's question, I am upgrading . . . to a 3770k.

Given IDC's testing, Sandy & Ivy may be in the optimum zone for overclocking. Reducing process size vs increased leakage. It looks like the chips will get hotter from now on as they shrink.

And cooling solutions will not fade into the sunset.

I agree with this comment. Die size of 4770K is the same as that of 3770K (22nm). The physics operating in this die size should be the same for both processors. Since 4770K has a bit more features than 3770K (such as an improved iGPU performance), it is expected that the the temperature is a bit higher in 4770K than 3770K.

In gpu world, people don't complain the small bump of performance from one generation to the next if the die size is the same (such as HD5870 vs HD6970, GtX480 vs GTX 580, 45nm die size). However, once the die size shrunk from 45nm to 30nm, the performance increased (such as HD6970 vs HD7970, GTX580 vs Titan, I think that GTX680 is a mid-range card if you consider its power consumption).

However, as he pointed out, I feel that the benefit of die size shrinkage seems to be getting smaller after 28-30nm. More and more leakage due to quantum effect is problematic for sure when the die shrinks, I guess. Since cpu reached to 22nm before gpu will reach to 20-22nm maybe next year, we are observing this symptom already in cpu field, I feel. I am very curious how much gpu performance will increase when they start to manufacture gpu on 20nm die.

Anyway I am not surprised the performance of 4770K if you consider it is made with the same die size as 3770K. And if you don't need iGPU or other new features, there is not much reason to get 4770K instead of 3770K.

 

[galego]

- In theory it's 10% faster, in practise it's as little as 1% in many games (eg, Skyrim is only 1%, WinRAR only 4%)

About a 8% improvement in general performance, but 1-5% on gaming (dGPU).

to me...

haswell is bulldozer...

like bulldozer.. Intel fubard, but unlike AMD, they only fubard the desktop line.

Failwell is the new faildozer...

 

[CrazyElf]

Once code is written to take advantage of AVX2/FMA, you all will change your tune about Haswell. Sure, it will take some time, but Intel had to start somewhere, and they had to be first (before developers).

Lets try and being realistic here. There is only so much more juice Intel can squeeze out of legacy x86 code without new instructions.

By then though for a lot of software, the new generation of CPUs will be out ... and they will have a new set of instruction sets. It takes time for software to adopt to the latest instruction sets. It really depends on what you use your CPU for, but right now, if you were to drop in an OC'ed Haswell over an OC'ed Sandy (assume neither is a golden chip), for a lot of CPU-intensive things, you'd hardly notice the difference.

And that's the issue. People wanted performance akin to Sandy over Lynnfield, which after OC's were factored in, were on the order of perhaps 25% because Sandy ran cooler and had more OC headroom on top of the base 10% faster clock for clock. This generation brings somewhat under 10%, but also does not OC as well as Ivy or Sandy on average.

 

[Mallibu]

Failwell is the new faildozer...

You're trying too hard.

 

[Dave3000]

I most likely won't upgrade to 4th gen i7 from my i7-3930k. I might upgrade to the upcoming i7-4930k. With the upcoming consoles that have 8-core CPU's, I would rather have a 6-core CPU than a 4-core CPU.

 

[NTMBK]

Failwell is the new faildozer...

You can keep saying that, but it's not going to catch on.

 

[JimmiG]

However, as he pointed out, I feel that the benefit of die size shrinkage seems to be getting smaller after 28-30nm. More and more leakage due to quantum effect is problematic for sure when the die shrinks, I guess. Since cpu reached to 22nm before gpu will reach to 20-22nm maybe next year, we are observing this symptom already in cpu field, I feel. I am very curious how much gpu performance

Maybe. The difference is that GPU clock speed remains pretty constant. Performance is improved in other ways. Usually, "more of everything", which works great in graphics. The GPU figures out the color of each pixel, and there are a lot of pixels to work on in parallel at any one time. In fact, it's not uncommon for the GPU clock speed to *drop* from one generation or model to the next, while the amount of compute resources increase.

22nm seems to work well at lower clock speeds, which may suit GPUs perfectly.

While the CPU market has been slow, GPUs make up for it. In the last 5 years, Intel has evolved from the Nehalem to Haswell. Meanwhile, on the GPU front Nvidia went from the GTX 280 to the GTX Titan....and from 240 stream processors to 2,688 stream processors...

 

[Ajay]

Haswell has failed in some regards to live up to the hype that typically surrounds Intel CPU introductions. That's completely different than Bulldozer which not only failed to live up to it's hype, but failed to stack up well against it's own predecessor.

Some of the architectural enhancements in Haswell where expected to deliver a larger bump in IPC than we are seeing. Apparently there are compromises at the block level and maybe even the transistor level (most people don't realize that not all transistors in a CPU are created equal) that we don't know about yet. C'est la vie!