As much as I wanted a Haswell rig, now I'm thinking twice about it.

[VirtualLarry]

I was considering a Haswell quad-core, but since I already own two Q9300 CPUs and two 1045T hex-core CPUs, one in a 990FX board with all the features (except PCI-E 3.0), I'm thinking, I don't game anymore, what's the point.

I might upgrade the Thuban to a Steamroller CPU, when and if they are released. 8 cores x 4Ghz sounds pretty sweet to me. Especially since many things, especially games, are becoming more multi-threaded, and not less. But I already said I don't game much, so that matters little.

I'm also thinking that with DDR4 on the horizon, that it might not make sense to upgrade to a platform with DDR3, just before that comes out, if there are any tangible performance benefits to DDR4.

 

[Mars999]

DDR3 is already faster than DDR4 when it arrives, unless they release the 4266 flavor...

Voltage will be less 1.2 IIRC vs. 1.5

2133–4266 MT/s compared to DDR3's 800 and highe

 

[VirtualLarry]

One other thing that I didn't mention, that really turns me off regarding mainstream Haswell, is lack of cores. I mean, Steamroller = 8cores, Haswell = 4 cores. In my mind, for heavily-threaded applications, there is no comparison. Just like the 8350 compares favorably with the 3770K when each is at stock, when streaming via Xsplit while playing online games.

Edit: I guess what I'm saying is, Haswell is out of the question, but not Haswell-E (depending on price).

 

[Homeles]

SR FX will probably sport up to 10 cores. If it ever becomes an more than a fantasy, that is. I have a feeling that we won't be getting an updated roadmap until AFDS in November.

 

[ShintaiDK]

DDR4 aint really gonna take off until 2015 with Skylake from Intel. And 2014 its all gonna be used for servers. And I doubt its much different in the AMD camp.

Skylake will also have PCIe 4.0 btw. Classic always something better on the horizont.

 

[tweakboy]

Hey Mars thats 2016 2017...

 

[frozentundra123456]

One other thing that I didn't mention, that really turns me off regarding mainstream Haswell, is lack of cores. I mean, Steamroller = 8cores, Haswell = 4 cores. In my mind, for heavily-threaded applications, there is no comparison. Just like the 8350 compares favorably with the 3770K when each is at stock, when streaming via Xsplit while playing online games.

Edit: I guess what I'm saying is, Haswell is out of the question, but not Haswell-E (depending on price).

Who knows what Haswell will bring? But 8 core FX even in heavily threaded applications only beats 3770k in a very few selected applications and loses in a lot of others. Steamroller looks promising if they can bring the 20 percent performance improvement they are claiming and get it out on time. But those are two big ifs.

I agree though Intel should bring out a six core Haswell at a reasonable price for the mainstream platform.

 

[Rezist]

Isn't Haswell going to allow overclocking of all chips again? Kinda wanted to buy a low end quad and do a mild to moderate OC 4Ghz~ish

 

[Soulkeeper]

I'm most excited about what AMD does after AM3+/FM2
Granted ddr4, pcie 3.0, die shrinks, etc. might not be so massive by themselves, but together they add up.
I really wanna see the sockets merged where we have the choice of dropping in an APU or a CPU and don't have to lock ourselves in.

I guess we have to wait another year+
PC hardware is just a big waiting game. Really we don't need to upgrade, we are just that 1% who do it for no good reason at all ...

 

[SocketF]

I wont expect Steamroller for AM3+.
Vishera is all over the current roadmap for 2013 and next year we'll only get new opterons so far. But these opterons will need a new platform for sure. PCIe3, probably DDR4 .. definitely not AM3+ compatible. If the memory controller could use DDR3, too, then it will be rather FM2 compatible.

Buying hasell or not .. well depends what you do. Buying the Haswell i5 with 4 Threads and then (much) later upgrade to the 14nm version with 8 threads would be a good upgrade path imo.

@Soulkeeper:
I agree.

 

[Rvenger]

I think this i5-3570k will be my last quad core. Its either going to be IVB-E, HSW-E or Steamroller... which ever comes out first.

 

[Homeles]

Steamroller looks promising if they can bring the 20 percent performance improvement they are claiming and get it out on time. But those are two big ifs.

Claims they've made:
+30% OPS per cycle (I.e., +30% IPC)
-20% branch mispredictions
5-10% improvement in INT scheduling

Other changes coming:
FPU stages decreasing from 4 to 3 (potential for higher IPC)
50% larger L1 Data Cache (64KB -> 96KB) and increased associativity
Double decode width (Anand states that this will be the single largest contributor to performance gain)
Resizeable L2 cache
uOp cache

Most detailed source available: http://www.brightsideofnews.com/new...core-architectural-enhancements-unveiled.aspx

Honestly, the improvements are all there to hit that 30%, if not more. If you've read Johan De Galas' article here at AnandTech, you'd know that the largest issues with the Bulldozer architecture (clock speed [Piledriver addressed this], branch misprediction penalty, L1D$ size) will be fixed with the launch of Steamroller.

The big problem I see with these changes is power -- I have no clue how they're going to keep the power usage down. I think a lot of the success of Steamroller is going to be riding on whether GloFo decides on whether they're not going to royally screw up or not. Is it really possible to keep the same 4GHz+ clock speeds? I'm totally being an armchair CPU architect here, but I don't think it is. If they magically hit that 30% while maintaining clock speeds, they would easily have a higher performance solution than Intel's mainstream offerings, albeit at lower performance per watt.

We don't even know if they'll release an FX version for consumers, so it'll be interesting to see how things pan out. At any rate, Steamroller should promise to be a huge improvement, barring any fab disaster or crippling errata.

 

[Ken g6]

I'm looking forward to benchmarks of a Haswell i7, because from the architecture it looks to me like hyper-threading may be much more beneficial on Haswell than on previous processors. So that would get you more cores.

I also think that AVX2 may become as important as SSE2 is now - if x86 remains dominant in the market. (As opposed to ARM.)

 

[VirtualLarry]

Buying hasell or not .. well depends what you do. Buying the Haswell i5 with 4 Threads and then (much) later upgrade to the 14nm version with 8 threads would be a good upgrade path imo.

My understanding is that mainstream CPUs from Intel are still all quad-core chips, as far as roadmaps that have been leaked/released show.

 

[Mars999]

I have already made my mind up awhile ago, I am getting Haswell-E as I want more than 2 SATA3 ports, and or SATA Express also on that MB. DD4 would be great but a huge deal unless I see benchmarks that show some kind of benefit vs. DDR3 1866 other than volts.....

PCIe-4 would be great but I am not waiting until 2015 for that.

I want 8 or 10 cores from Intel and from what I read HW-E will probably have that...

Here's to hoping!!!

Now shutup and take my money INTEL!!!!

 

[Homeles]

I can't imagine Intel sticking with 4 cores after Broadwell. With Kaveri offering potentially offering 6 cores (I still don't get how they'll have the headroom for all this), they need a 6 or more core mainstream chip of their own.

 

[Vesku]

Haswell-E will be a while, Ivy-E has been delayed a few times hasn't it?

 

[jihe]

I was considering a Haswell quad-core, but since I already own two Q9300 CPUs and two 1045T hex-core CPUs, one in a 990FX board with all the features (except PCI-E 3.0), I'm thinking, I don't game anymore, what's the point.

I might upgrade the Thuban to a Steamroller CPU, when and if they are released. 8 cores x 4Ghz sounds pretty sweet to me. Especially since many things, especially games, are becoming more multi-threaded, and not less. But I already said I don't game much, so that matters little.

I'm also thinking that with DDR4 on the horizon, that it might not make sense to upgrade to a platform with DDR3, just before that comes out, if there are any tangible performance benefits to DDR4.

if you don't game then you are set for the foreseeable future.

 

[shady28]

I think we all know the basic problem. Intel doesn't have serious competition from AMD, or anyone else, in the desktop space.

So, they can sit back with pathetic 10-15% per year performance gains. Initially they appear to be just fine.

But this is a fallacy they are rapidly falling into.

Those performance gains are not enough for people to lay down hard earned cash for upgrades every 2-3 years.

At this point, the average Joe's PC will probably last 5-8 years before they will feel the need to upgrade.

Intel desperately needs to get back to their core business and push the envelope for desktops, work with MS and other companies to get things moving again, but they are too busy playing at someone else's game (ARM/Mobile) and milking existing tech (quad core, really, that's very 2008).

Meanwhile 90% of their business that is the core of their revenue slowly dies.

 

[VirtualLarry]

So, they can sit back with pathetic 10-15% per year performance gains. Initially they appear to be just fine.

As far as single-threaded performance goes, that's not exactly pathetic, from an engineering standpoint. Many of these performance increase problems are relatively "hard", although throwing more transistors at it helps.

Those performance gains are not enough for people to lay down hard earned cash for upgrades every 2-3 years.

At this point, the average Joe's PC will probably last 5-8 years before they will feel the need to upgrade.

Intel desperately needs to get back to their core business and push the envelope for desktops, work with MS and other companies to get things moving again

That's the other thing though, as far as multithreaded performance goes, it's not that hard (as long as you can stay under a reasonable TDP) to just add cores, as you shrink geometry. That factor is largely what has been propelling GPU performance increases, because graphics is highly parallel as far as processing goes.

I realize that Amdahl's law limits the performance increases possible for many multi-threaded workloads, but there are still gains from 4 to 8 cores.

And some things, like distributed computing (which I participate in), can use basically as many cores as you can throw at it. (For BOINC at least.)

 

[Tsavo]

I think we all know the basic problem. Intel doesn't have serious competition from AMD, or anyone else, in the desktop space.

So, they can sit back with pathetic 10-15% per year performance gains. Initially they appear to be just fine.

But this is a fallacy they are rapidly falling into.

Those performance gains are not enough for people to lay down hard earned cash for upgrades every 2-3 years.

At this point, the average Joe's PC will probably last 5-8 years before they will feel the need to upgrade.

Intel desperately needs to get back to their core business and push the envelope for desktops, work with MS and other companies to get things moving again, but they are too busy playing at someone else's game (ARM/Mobile) and milking existing tech (quad core, really, that's very 2008).

Meanwhile 90% of their business that is the core of their revenue slowly dies.

Intel spends billions every year on R&D.

They aren't coasting.

 

[Homeles]

But this is a fallacy they are rapidly falling into.

Your entire argument is a fallacy, and is simply asking for a reach-around from the enthusiast community hivemind.

Intel has competition. It's called ARM, in case you haven't heard of it.

There's a good chance that Joe's PC would have suffered a hard drive failure by 5 years, and Joes are inclined to throw away their dysfunctional computer rather than repairing it.

What in the world would Joe do with that extra computing power anyways? Joes typically browse the internet, email, and do word processing. For those that actually need some muscle, there actually are processors to suit their needs.

You also have seem to missed the fact that enthusiasts aren't typically bound by power constraints. Just because Haswell is targeted for 84W does not mean you have to run it at 84W. There is no inherent neutering that would prevent it from running at a higher power usage, either. In fact, there's a substantial probability that Haswell will bring back BCLK overclocking, and Intel has already stated that Haswell brings overclocking improvements to the table.

You know what Joe does want? More battery life. Thinner devices. Intel knows this, and since there are a lot of Joes out there, Intel builds processors that suit the needs of Joes.

You act as if this is at your expense. It isn't. In fact, what was the number one revolution in the past decade in x86 performance? Conroe. Guess what let to that? Focusing on performance per watt. You seem to have forgotten this or are ignorant of it.

In fact, blind pursuit of performance is simply, well, stupid. It's been said that the usefulness of ILP, one of the major methods of extracting more instructions per cycle, is close to being maxed out. In case you aren't aware, power consumption grows exponentially with clock speed. Even if Intel didn't give a hoot about power and serviced your selfish needs, they'd hit a brick wall not two steps from where they are right now.

It's not like Intel's alone on this -- AMD's strategy's changed as well. But you're focusing on Intel, and I can guarantee it's because Evil Empire Intel is the low hanging fruit and because you failed to conjure the critical thinking skills to look at the whole picture.

You want something to blame for not getting exponential leaps in performance? Go yell at physics. That's what you should be upset with.

 

[frozentundra123456]

Intel spends billions every year on R&D.

They aren't coasting.

First of all, we havent seen 10 to 15 percent performance gains since SB came out, and wont see it with Haswell either.

I know Intel is focused on mobile, probably with justification.

However, I do think they have become complacent in high performance desktops and are relying too much on high IPC and hyperthreading, at the expense of increasing clockspeeds and core counts.

 

[Homeles]

Well how exactly would they increase clock speed? New manufacturing processes don't seem to be providing much benefit. They essentially need a breakthrough in materials science (à la HKMG) to get anywhere. They could implement RCM, but that's about it.

The pipeline is long enough already.

 

[JamesWoods]

Haswell is supposed to have more control over mobo components, reducing total system usage. It's not going to shut off either, it will utilize sleep modes more.