Ivy Bridge vs. Haswell

[blueberryjudy]

I do a lot of parallel g++ compiles, I currently have a Core i7-3770, would there be a substantial improvement in performance going to Core i7-4770? Even a 10% decrease in compile times would mean quite a bit to me.

 

[jacktesterson]

Not worth the cost/effort in my opinion. You'd also need a new motherboard.

 

[ChronoReverse]

The overall IPC improvement of Haswell is about 10-15% when not using the new stuff like AVX2 and TSX.

So if you're running at stock clocks and 10% is worth the cost of a new system, then it's a good upgrade.

 

[videogames101]

parallel compiles? go hex core if the cost is not a problem

 

[tweakboy]

Depends what your doing. Do you do video editing pro audio .. If not and no autocad and rest of them then all you need is a 8 thread cpu 4770k.

For someone like me it was smarter to do what I did, because of what I do....

 

[CHADBOGA]

The overall IPC improvement of Haswell is about 10-15% when not using the new stuff like AVX2 and TSX.

I think the IPC improvement of Haswell is more like 5 - 8%.

 

[ChronoReverse]

Yeah, I double-checked and you're right. Not sure why I was thinking that high.

 

[BrightCandle]

Phoronix run Linux benchmarks and they always run g++ compilations with parallel settings so its about the best benchmarks I know of for g++:

As you can see there is a fairly significant jump to Haswell. Its not quite as good as going 6 cores but its more universally available performance than with a hexacore that really only works well in very specific programs.

I think the reason Haswell does so well comparatively is two fold. One is the increased ports that allow a lot more parallel instructions and the second is the branch prediction. Compilers are complicated, branchy and generally big messy programs.

 

[krumme]

I think the IPC improvement of Haswell is more like 5 - 8%.

Agree 5-8%. We tend to oversell the ipc improvements from sb to ib and especially from ib to hw.

 

[BallaTheFeared]

Looks like Haswell is giving a nearly 20% reduction in time to completion in BC's post.

 

[exar333]

I actually might suggest the Xeon 1270/1275 v3 instead of the 4770. It is pretty much the same price, and also supports ECC (if you want it).

With the 5-8% IPC improvements over IB, the 1270/1275 has 100mhz on the 4770 (3.5 base instead of 3.4) but does turbo to the same frequency. This could mean another ~3% in sustained workloads, on top of the 5-8% IPC gains over 3770. That's pretty close to your 10% difference, and often the Xeon is better-binned anyways. That could net you another 1-4% in performance as well, if it runs cooler and sustains more cores with more mhz, longer vs. the 4770.

 

[Arkaign]

Yep, that is definitely better. Doesn't seem like OP overclocks, and that's probably good for doing operations where you can't afford errors anyway. Haswell OC ceiling is low enough on average that the Xeon even makes a lot of sense for a gamer with a typical mid-high GPU as well. But in particular the instruction sets and binning make the Xeon a no-brainer for a serious work-PC CPU in that price range.

 

[blueberryjudy]

Phoronix run Linux benchmarks and they always run g++ compilations with parallel settings so its about the best benchmarks I know of for g++:

As you can see there is a fairly significant jump to Haswell. Its not quite as good as going 6 cores but its more universally available performance than with a hexacore that really only works well in very specific programs.

I think the reason Haswell does so well comparatively is two fold. One is the increased ports that allow a lot more parallel instructions and the second is the branch prediction. Compilers are complicated, branchy and generally big messy programs.

Thanks... I think this benchmark is huge to convince the rest of my lab to upgrade.