Educate me on Haswell

VladDracule

[H]ard|Gawd
Joined
Feb 5, 2007
Messages
2,043
Hello all, i recently completed graduate school and am looking at a few new pieces of hardware.

My question is in regards to the ivy bridge vs sandy bridge. My main question is, will the actual PROCESSING power of the haswell see a significant improvement over the ivy bridge?

for a desktop i do not care about the increased performance out of the integrated video card because ill have a dedicated card anyways.

Are we likely to see an improvement in terms of sheer processing power in the haswell vs the ivybridge? enough to justify waiting several months?
 
Last edited:

Forceman

[H]F Junkie
Joined
Aug 7, 2005
Messages
9,243
The numbers that Anand was throwing around in his preview were 5-15% clock for clock, but with significant increases in some areas (things that can use AVX2, for instance). I don't know if it is worth waiting for or not - kind of depends on how much you are hurting now. An Ivy chip is going to tear through pretty much anything, so it's not like you need to wait.

http://www.anandtech.com/show/6355/intels-haswell-architecture/14
 

Liger88

2[H]4U
Joined
Feb 14, 2012
Messages
2,657
Pretty much what Forceman said. I don't think you're going to see major improvements over Sandy/Ivy as people would have thought. We hardly see a 5% increase with Ivy over Sandy which disappointed some, although you did get lower power consumption with Ivy which is kind of a big deal. Generally I think it's safe to say you'll always see a 5-10% improvement generation to generation with Intel and those are even some numbers Intel's been throwing around. However, there are certain things Haswell is going to include that may make that percentage skyrocket (TSX and AVX2 for instance) in some applications, but that could take a couple years to fully be realized..

All in all I think Haswell will be a future proof chip for a long time if Intel does everything right and programmers take full advantage of it. Time will tell though, but I wouldn't expect a Netburst > Nehalem style numbers anytime soon.
 

pelo

2[H]4U
Joined
Apr 23, 2011
Messages
2,911
The only significant improvement will be in the GT3 model (+ eDRAM) in iGPU performance. The IPC bump is roughly 10-15%, but that too depends on ISAs and workloads, and Intel hasn't released any clock speed figures yet. If I were to guess, I'd say clock speeds have stalled while IPC has gone up (and that's in order to retain the same TDP while bumping up the EUs on the GPU).

TSX and AVX2 would be the biggest improvement CPU-wise, but won't see wide-scale adoption for 3-5 years as it depends very heavily on developers recompiling their software. Getting a developer to give a shit about the x86 world today isn't easy...

What's your typical workload and how long do you expect to use the machine? If it's a period of 3+ years then it might pay off opting to wait for Haswell. If you're usage scenario is just your run-of-the-mill computing with some gaming then waiting probably isn't worth it. Bear in mind the chips aren't being released until late 1H, with the 35-45W laptop models being released first.
 

MrSmegzabush

Limp Gawd
Joined
Sep 21, 2004
Messages
434
It may sound a bit short sighted, but will AVX2 and TSX benefit gaming in any way? It looks like most of these added instructions benefit threaded apps. From Lynnfield to Sandy Bridge to Ivy Bridge, the improvements in gaming have been relatively marginal.

We look back at SB as a huge increase for us gamers because while it did improve on single thread performance, the gains were not as great as the die-shrink and awesome multiplier overclocking they enabled, and the chips were capable of.

With AMD less of a threat in the enthusiast segment, and intel trying to move their chips more into the mobile and gadget space by dedicating their die space to GPU, and design for lower power consumption.
 

Tsumi

[H]F Junkie
Joined
Mar 18, 2010
Messages
13,551
AVX2 and TSX will not benefit any application unless the application is compiled to use them. Bulldozer and Piledriver both have AVX and FMA instructions, and would crush IB and SB in any application that used those instructions. Problem is, very few applications do, and it will take a while before adoption becomes widespread.
 
Top