Upgrading to haswell

[jmachin]

I was just wondering whether a z77 board will support Haswell when it's released or would I have to buy a new board to go with it? Because I plan on building a PC in January but I want to have the option of being able to upgrade if Haswell CPUs end up being much better.

Cheers

 

[lehtv]

Z77 won't support Haswell. Haswell is LGA1150, Z77 is LGA1155.

Haswell also won't be anything earth shattering in terms of CPU performance, but will improve greatly on iGPU performance and power consumption (especially idle), among architectural improvements and features.

 

[ShintaiDK]

Haswell is a very big step forward with 256bit execution units for AVX2, TSX, SOix states and ondie VRM.

Socket is LGA1150 and as said the VRM is moved ondie.

 

[aaksheytalwar]

No chance. The new socket is 1150. But haswell will be one of the biggest jumps in recent years.

 

[Edrick]

Haswell also won't be anything earth shattering in terms of CPU performance

Then you have been living in a hole if you actually think that.

 

[lehtv]

Then you have been living in a hole if you actually think that.

That's very convincing.

All the news I've read say CPU performance will be up by 10%, possibly up to 20%, plus more OC headroom. Certainly a bigger improvement than Ivy was over Sandy, but not so big you should sweat over whether you can upgrade from Ivy to Haswell without changing the motherboard. Best to wait for Broadwell since an overclocked Ivy will be certainly be faster than a stock Haswell.

 

[aaksheytalwar]

No matter how high you oc an ivy. Haswell at stock will give a better out of the box experience with future applications. Period. Always has happened in the past and always will happen.

 

[lehtv]

Well, I'll quote you on that when Haswell is released. Did a stock 2600K outperform a heavily OC'd i7-8xx? Barely, and that's mainly because of the higher stock clock speed of the 2600K. Did stock Bulldozer outperform an overclocked Phenom II? Certainly not, even with the higher stock clock speed.

 

[kleinkinstein]

For me it'll be all about the IPC, not Mhz machismo. I'll dump my Ivy in a NY minute if it's >10%, sports solder not glue and has a healthy unlock. Thank god for ebay!

 

[Beavermatic]

probably around 5% real world improvement (if that) compared to Ivy Bridge.

either way I'll buy one... but anyone expecting anything phenomenal other than OC potential and better mobility devices for this processor is out of their mind.

 

[laputan machine]

If it sports a significantly smaller tdp footprint and sports better than trinity gpu I'll be happy.

 

[Edrick]

That's very convincing.

All the news I've read say CPU performance will be up by 10%, possibly up to 20%, plus more OC headroom. Certainly a bigger improvement than Ivy was over Sandy, but not so big you should sweat over whether you can upgrade from Ivy to Haswell without changing the motherboard. Best to wait for Broadwell since an overclocked Ivy will be certainly be faster than a stock Haswell.

Everything you said only holds true on legacy software. As soon as software is optimized for Haswells instructions, you will see upwards of 80% improvement and in some cases over 100%.

And with this type of increase available, I do not think developers will wait long to take advantage of that. (Games may be sooner than anything else since most currrent games already support FMA).

 

[Kristijonas]

AVX2, TSX, SOix states and ondie VRM.

Could someone explain the meaning of these words, please? Want to educate myself

 

[WhoBeDaPlaya]

From Nehalem -> Haswell (approx) : 1 x 1.1 x 1.03 x 1.1 = 1.25
Higher clocks and 25% IPC improvement? If it clocks to 5GHz, that's be equivalent to a 6.25GHz Nehalem

 

[Barfo]

Everything you said only holds true on legacy software. As soon as software is optimized for Haswells instructions, you will see upwards of 80% improvement and in some cases over 100%.

And with this type of increase available, I do not think developers will wait long to take advantage of that. (Games may be sooner than anything else since most currrent games already support FMA).

So...2 years or more? IB will be obsolete by then anyway.

 

[Edrick]

So...2 years or more? IB will be obsolete by then anyway.

So....pulling numbers out of your arse?

Remember, most games already use FMA instructions (video cards have supported that for many years already). It will not take 2 years in a lot of cases.

 

[NTMBK]

Everything you said only holds true on legacy software. As soon as software is optimized for Haswells instructions, you will see upwards of 80% improvement and in some cases over 100%.

That holds only for algorithms which can be parallelised like that.

And with this type of increase available, I do not think developers will wait long to take advantage of that. (Games may be sooner than anything else since most currrent games already support FMA).

Do you have a list of games that make use of FMA? Because I'm extremely dubious of that.

 

[NTMBK]

So....pulling numbers out of your arse?

Pot, meet kettle.

Remember, most games already use FMA instructions (video cards have supported that for many years already). It will not take 2 years in a lot of cases.

Good god man, are you serious? That has nothing to do with using FMA on a CPU algorithm.

If developers put FMA into their code, either they will have to provide multiple codepaths increasing complexity and cost, or they will have to abandon support for every processor shipped to date apart from Piledriver and Bulldozer. It's not happening for a long time.

 

[Edrick]

Do you have a list of games that make use of FMA? Because I'm extremely dubious of that.

Nvidia and AMD have had FMA instructions in their GPUs for many years. It is ONE of the reasons why GPUs are so much faster than CPUs. I doubt you will be able to find a game specifically say it uses FMA (since that means nothing to most people), anything that says its optimized for the latest GPUs, most likely will have it.

Now, I am not saying it will work on Haswell right out of the box (in most cases, it will need to be optimized for it), but it will not take 2 years, thats for sure.

 

[Edrick]

Pot, meet kettle.

If developers put FMA into their code, either they will have to provide multiple codepaths increasing complexity and cost, or they will have to abandon support for every processor shipped to date apart from Piledriver and Bulldozer. It's not happening for a long time.

Wrong!

FMA instruction will be able to use the legacy instruction if the CPU does not support FMA instructions. No extra work and not extra code paths.

 

[NTMBK]

Nvidia and AMD have had FMA instructions in their GPUs for many years. It is ONE of the reasons why GPUs are so much faster than CPUs. I doubt you will be able to find a game specifically say it uses FMA (since that means nothing to most people), anything that says its optimized for the latest GPUs, most likely will have it.

Now, I am not saying it will work on Haswell right out of the box (in most cases, it will need to be optimized for it), but it will not take 2 years, thats for sure.

"Need to be optimized"? How about, "is entirely not relevant to the discussion"? Have you ever written code which uses these instruction set extensions you love so much?

 

[Edrick]

"Need to be optimized"? How about, "is entirely not relevant to the discussion"? Have you ever written code which uses these instruction set extensions you love so much?

Yes. And it is clear you haven't.

Now stop attacking me and go back to your games.

 

[Edrick]

Pot, meet kettle.

Yea, right kid.

I just disagree with a 2 years or more number someone pulled out of their arse. I never put an exact number anywhere.

 

[NTMBK]

Wrong!

FMA instruction will be able to use the legacy instruction if the CPU does not support FMA instructions. No extra work and not extra code paths.

Will it ah heck. If a CPU which does not support FMA3 receives an FMA3 instruction, it will not have a clue what to do.

You can have alternate paths in your code which avoid the FMA instruction and instead issue separate multiply and add instructions, but as I pointed out that adds complexity.

 

[NTMBK]

Yes. And it is clear you haven't.

Now stop attacking me and go back to your games.

Aha. I write code full of SSE intrinsics for a living. Try again.