Haswell: Don't expect significant performance gains

Is it time for Moore's Law to R.I.P?

It's been dead for while now, unless you like integrated graphics.

Nehalem performance improvement over Penryn per-clock:

0-10% improvement in lightly-threaded tasks
Up-to %40 improvement in massively-threaded tasks. But this is due to the reintroduction of HyperThreading + the new ring bus, so you can't expect this huge increase in the future, now that they've solved the inter-core communications bottleneck and introduced a way to keep the cores fed using extra threads.

Sandy Bridge performance improvement per-clock: %5-10.

Ivy Bridge performance improvement per-clock: %5-7.

I'll tell you, I'd be happy just to see another %5-10 bump.
 
Last edited:
Moore's Law has always been about being able to double the density of transistors per area in a set time frame. That doesn't necessarily translate to doubling performance, or power efficiency, or features.
 
Well the reality is that chip performance has gotten good enough to satisfy most users needs/uses, so that now users prefer mobile platoforms for what used to be done at a desk - email, web browsing, etc.

I'll be moving to the next enthusiast platform once it's released (the successor to 2011 - whatever it is) for my gaming rig.
 
Ivy Bridge performance improvement per-clock: %5-7.

I'll tell you, I'd be happy just to see another %5-10 bump.

So do you think 5ghz OC's could be achievable on the norm with Haswell? Or will that require waiting for the next enthusiast platform?
 
I'm hoping they bring back a six core CPU with Haswell and from this info:

35, 45, 55, 65, 77, and ~100W+ (high-end) TDP desktop processors.
One might assume they are doing just that. A Haswell six core CPU should be right around ~100 watts. Well, I hope. :D
 
Well let's say 10% at the lowest, that should still be about 17% better than Sandy which isn't even that bad. Also, if they can easily OC to 5ghz then that makes them even better. If it's not that great though, I can just wait for the next.
 
I hope Intel pulls the proverbial head out of the ass and goes back to using fluxless solder under the IHS.

Regardless, I think I'm good until Skylake or even Skymont. Maybe even after that. :/
 
I'm hoping they bring back a six core CPU with Haswell and from this info:

One might assume they are doing just that. A Haswell six core CPU should be right around ~100 watts. Well, I hope. :D

How many times do we have to say it? You won't see a 6-core on the mainstream platform any time soon. Intel has already confirmed this in their roadmaps for Haswell.
 
I think the marketplace trumps Moore's Law. Right now, the money is in mobile stuff like tablets and laptops so energy efficiency and such are much more important that what we here at [H] are interested in. Sad to say but battery life seems more important to Intel and AMD than sheer speed and performance.
 
I think the marketplace trumps Moore's Law. Right now, the money is in mobile stuff like tablets and laptops so energy efficiency and such are much more important that what we here at [H] are interested in. Sad to say but battery life seems more important to Intel and AMD than sheer speed and performance.

Moore's law refers to the transistor density and fab node advancements but not performance. Tick>tock on 2 year cycles only promises benefits in density.

Moore's law is the observation that over the history of computing hardware, the number of transistors on integrated circuits doubles approximately every two years

Haswell is a 4 core part. The workstation platform might have six cores but who knows what the price of those will be :p

The performance estimates are not minimum of 10% but gains in the lower double digits are possible, meaning both IPC and clock speed gains might hit over 10% depending on the workload.
 
How many times do we have to say it? You won't see a 6-core on the mainstream platform any time soon. Intel has already confirmed this in their roadmaps for Haswell.

You have to say it a lot so stop being so annoyed about it. Roadmaps get updated and I'm hoping for a surprise!
 
You have to say it a lot so stop being so annoyed about it. Roadmaps get updated and I'm hoping for a surprise!

It doesn't happen that way. These things are planned years beforehand. In order for Intel to go "Oh shit, let's make a 6-core mainstream part!" it would take one-to-two years of work and a new socket (it wouldn't fit 1150) so technically it wouldn't even be a mainstream part at all if you throw in the costs of such an endeavor.

MasonD, the 1155/1150 mainstream desktop parts are determined by the laptop and ULV chips. If you want to know what the desktop will get just look down to the 35W and 45W TDP chips because it'll be the exact same thing but with higher clock speeds.
 
It doesn't happen that way. These things are planned years beforehand. In order for Intel to go "Oh shit, let's make a 6-core mainstream part!" it would take one-to-two years of work and a new socket (it wouldn't fit 1150) so technically it wouldn't even be a mainstream part at all if you throw in the costs of such an endeavor.

I'm trying to accept (lol) what you two are saying here. I have heard it before too. I just don't see them not having it designed already and not even making one (even if it's not being released to the public ATM, they still got to have produced them) when they did it for Gulftown and SandyBridge-E. It seems like keeping with the pace they'd do it with an IvyBridge-E chip or Haswell chip at the very latest (shrugs).

MasonD, the 1155/1150 mainstream desktop parts are determined by the laptop and ULV chips. If you want to know what the desktop will get just look down to the 35W and 45W TDP chips because it'll be the exact same thing but with higher clock speeds.
So what are they referring to with the 100+ watt chip from Haswell? Server side only? :(
 
I'm trying to accept (lol) what you two are saying here. I have heard it before too. I just don't see them not having it designed already and not even making one (even if it's not being released to the public ATM, they still got to have produced them) when they did it for Gulftown and SandyBridge-E. It seems like keeping with the pace they'd do it with an IvyBridge-E chip or Haswell chip at the very latest (shrugs).

So what are they referring to with the 100+ watt chip from Haswell? Server side only? :(

Enthusiast chips with unlocked multipliers and high stock clock speeds. TDP has as much to do with stock speeds as it does to do with number of cores. Remember, Haswell will also have a significantly beefed up iGPU, and that factors into the TDP rating as well.

They won't have 6-core mainstream chips because it's to differentiate the mainstream and server segments. As much as you are wishing it to happen, it won't.

TDP rating doesn't really refer to the CPU power as much as it refers to minimum board power delivery design and minimum heatsink cooling capacity.
 
After owning a i7 920 the one thing I have issues with is the heat output, enough to make whatever room it's in uncomfortable
 
It doesn't happen that way. These things are planned years beforehand. In order for Intel to go "Oh shit, let's make a 6-core mainstream part!" it would take one-to-two years of work and a new socket (it wouldn't fit 1150)
Unless the die is truly gigantic (and its Intel, so that isn't going to happen) Intel could stuff 6 cores into a s1155 package. But it would be bottle necked by the platform at that point.
 
We're in kind of a stalemate lately because of a few things, windows 7/8, and the xbox developed pc games to name a few. They'll be a few cpu bumps but anything major won't be for a bit I reckon. We need more pc centric games and apps and a few major changes.
 
I don't have any plans to upgrade my desktop to Haswell anyway, just don't think it would be worth the cost. I do however fully intend on upgrading my C2D MacBook Air as soon as there's one with Haswell in there. I'm already impressed with the battery life on it, and it will only get better. Unless Apple gets stupid with it's obsession with thin, and takes the opportunity to make smaller batteries. Crossing my fingers that that won't happen.

My next desktop upgrade will likely be built around Skylake.
 
It's been dead for while now, unless you like integrated graphics.

Nehalem performance improvement over Penryn per-clock:

0-10% improvement in lightly-threaded tasks
Up-to %40 improvement in massively-threaded tasks. But this is due to the reintroduction of HyperThreading + the new ring bus, so you can't expect this huge increase in the future, now that they've solved the inter-core communications bottleneck and introduced a way to keep the cores fed using extra threads.

Sandy Bridge performance improvement per-clock: %5-10.

Ivy Bridge performance improvement per-clock: %5-7.

I'll tell you, I'd be happy just to see another %5-10 bump.

In many apps and games, Nehalem is up to 35% faster than Penryn per clock. It didn't really show at release due to lack of optimizations and/or GPU limited game benchmarks. Sandy Bridge was another 10-15% per clock. I guess Haswell will be 15% faster per clock over Ivy and bring 5-10% higher clocks.
 
Last edited:
The most recent Intel leaked powerpoint/information puts Haswell at up to 10% faster per clock over Ivy Bridge.
 
So what are they referring to with the 100+ watt chip from Haswell? Server side only? :(

Don't pay too much attention to the TDP this generation as Haswell has a lot of graphical power that contributes to it thus skewing the whole-chip TDP. It's unlikely that the mainstream socket gets a >100TDP Haswell chip but even if it does it'll be a 4-core part with a very hefty contribution to that TDP by the on-die graphics (remember that TDP is measured by both)

The workstation platform is certainly going to have a >100W TDP and 6-core parts.
 
This is the natural product of having no competition in the high-end market. So Intel is chasing other markets instead. You really can't blame them as their goal is to make money.
 
Well the reality is that chip performance has gotten good enough to satisfy most users needs/uses, so that now users prefer mobile platoforms for what used to be done at a desk - email, web browsing, etc.

I'll be moving to the next enthusiast platform once it's released (the successor to 2011 - whatever it is) for my gaming rig.

Yeah, pretty much. A SB/IB oc'd to 4.5-5.x should be good enough for a little while longer.
 
This is the natural product of having no competition in the high-end market. So Intel is chasing other markets instead. You really can't blame them as their goal is to make money.

I wouldn't so much say that no competition in the high-end market is driving intel's designs, as much of it is determined by consumer demand: it's now shifted towards mobile computing consisting on laptops, netboots, ultrabooks, tablets, and smartphones. As for business demand, there's servers, workstations, desktops, laptops, but nothing that necessitates the need for ultra-performance parts to be stuck in every one of them, let alone even a smaller fraction of them. It's more like a microscopic sliver of the overall volume.

Gotta steer the ship towards the part of the coast where the biggest chest of buried treasure is.
 
This is the natural product of having no competition in the high-end market. So Intel is chasing other markets instead. You really can't blame them as their goal is to make money.

I know people have been stating this is the reason but I highly doubt it. It's more of a transition of the desktop-to-mobile and a shifting of funds to address the booming segment rather than the stable and decreasing one. Even if Intel were to make a CPU that's 3x faster than their current top end the average user wouldn't get much out of it. Contrastingly, if Intel were to create a 5W atom capable of good throughput and great on-die graphics they'd make a boatload of money and people would see the tangible benefits of such a product.

So quit blaming AMD... AMD is doing the same exact thing Intel is with regards to focusing on the mobile end of the market. It's been this way for a while now so I really don't understand the pissy attitude.
 
Intel needs to beef up integrated graphics more than cpu performance anyway, so this sounds fine. More of their die space ought to be tasked for the gpu so the market has a higher baseline performance level for graphics. Virtually no consumers are clamoring for more cpu speed because the current crop has way more performance to do what they need. The people yelling at intel for bottlenecked cpu performance don't exist, and if you find some it's like spotting a unicorn. For the people that genuinely need ever faster cpu speeds, they should focus a special line of cpus for that segment, but again, that focus will have ZERO benefit to the vast majority of people who use intel chips at this point.
 
Last edited:
Love the way Intel is thinking right now: continue to lower power consumption and increase performance when possible. I can't wait until we get around 17w quadcores, Thunderbolt graphics, all in one tiny tablet form factor.
 
After owning a i7 920 the one thing I have issues with is the heat output, enough to make whatever room it's in uncomfortable

LOL. you guys dont even know what heat is until you start bitcoin mining. Im gaming/multitasking on my I7 right now and cant even notice it running. I do however notice my other rigs with 2x 7970s, 2x5850's, 2x5830s and a 6770 bitcoin mining both by heat and the noise of the fans.
 
I'm planning on rebuilding my HTPC with Haswell once it comes out, so thermal improvements are technically more beneficial than speed improvements, but I can't help but be a little disappointed that it's not another big jump.
 
LOL. you guys dont even know what heat is until you start bitcoin mining. Im gaming/multitasking on my I7 right now and cant even notice it running. I do however notice my other rigs with 2x 7970s, 2x5850's, 2x5830s and a 6770 bitcoin mining both by heat and the noise of the fans.


I knew that.:D
 
regardless of how much faster haswell will be over ivy bridge, haswell is what im going to buy.

My i7 920 and X58 chipset need a refresh and i want native Sata 6gb/s and usb 3.0 and a healthy gain of performance wouldnt hurt ( maybe not so much in games but whatever )
 
regardless of how much faster haswell will be over ivy bridge, haswell is what im going to buy.

My i7 920 and X58 chipset need a refresh and i want native Sata 6gb/s and usb 3.0 and a healthy gain of performance wouldnt hurt ( maybe not so much in games but whatever )

you could've had that with z77/IB.
 
regardless of how much faster haswell will be over ivy bridge, haswell is what im going to buy.

My i7 920 and X58 chipset need a refresh and i want native Sata 6gb/s and usb 3.0 and a healthy gain of performance wouldnt hurt ( maybe not so much in games but whatever )

you could've had that with z77/IB.

Or x79/SB-E.

My 3820 @ 4.9Ghz setup is way faster than my W3565, and before that 920 was at 4.2Ghz.

Now I am kinda.. ho-hum about newer stuff coming out in the near future because it seems like this setup will be plenty fast for another couple years.. and even if I do need an upgrade, I can go up to a 6-core.

The quickness of 2011/SB-E is a pretty good difference over 1366/x58. And it is a lot easier to tweak. Took me all of about 30 minutes to find max clock for CPU and RAM. x58... you are talking days of tweaking and retweaking just to get the settings correct to find max stable.
 
Last edited:
I too am still on my I7-920 and was hoping for a decent upgrade, having passed over mainstream platforms for quite some time. 1155 SB wasn't worth my time, IB would barely be so methinks. What I was really hoping to do was upgrade to Socket 2011 but Intel made SB-E so stupidly expensive I couldn't justify it. If they sold the 3930K (the only real upgrade for me) with the pricing structure they offered the i7-920, they would have had a sale.

Even worse, I've not seen any IB-E Socket2011 processors! Latest leaks suggest that IB-E may now be pushed back to Q3 2013, released concurrently with Haswell mainstream and may or may not release an update X79! What the hell is that! Why release a "high end" platform based on old technology?! On another site there was an Intel rep that said something about "well, we like to give all our processor tech at least a year on sale, so... since SB-E didn't release until Q4 2012, then IB-E won't be until Q4 2012 at the earliest". Fucking infuriating. Why can't they get their shit together and release the enthusiast platform either 1) Earlier than mainstream, as they did with 1366 or 2) at least concurrently with mainstream? This is the kind of bullshit that makes me wish AMD had a solid competitor for performance. If they had a chip that was 10% less powerful than the 3930K and significantly cheaper, with a good chipset, I'd buy in a second.

Even worse is that (unless someone can give me an update) I can't just pick up a Socket 2011 X79 platform and invest into it now as a good buy, as apparently it doesn't support full PCI-E 3.0 speeds? Or is that just certain boards? I was originally under the impression that X79 would have more PCI-E lanes and overall better performance than either 1155 SB or IB, but if those lanes aren't running at PCI-E 3.0 spec (where, IB is the first chipset to official suppor them) then what is the point of buying something so expensive? Please tell me that at least the Rampage IV series have full PCI-E 3.0 compliance and a ton of lanes!? Otherwise, I expect this same drama to play out again when Haswell has released and IB-E eventually comes out, but is limitedbecause of older technolgy?!

I'm tired of watching Intel plod along with a mishmashed, expensive platform of old and new. It really makes upgrading much more difficult. I can't believe they've fouled up so badly after their amazing 1366 platform; it came early, had all the new features yet to be released in mainstream, great performance, great value, lasted for ages and was worth it. Hard to decide where to go from here, especially if Haswell is not going to be releasing an enthusiast line. I'd much rather have them scrap the whole "enthusiast" socket if they're going to continue with this crap.
 
I know people have been stating this is the reason but I highly doubt it. It's more of a transition of the desktop-to-mobile and a shifting of funds to address the booming segment rather than the stable and decreasing one. Even if Intel were to make a CPU that's 3x faster than their current top end the average user wouldn't get much out of it. Contrastingly, if Intel were to create a 5W atom capable of good throughput and great on-die graphics they'd make a boatload of money and people would see the tangible benefits of such a product.

So quit blaming AMD... AMD is doing the same exact thing Intel is with regards to focusing on the mobile end of the market. It's been this way for a while now so I really don't understand the pissy attitude.

Difference is Intel is doing this because it can, AMD is doing it because it CAN'T
 
Back
Top