Intel haswell i7-4770K preview article-TMSHW

[paulbagz]

My next CPU

-PB

[andressergio]

Lit improvement on calculation over IVY 3770K
the best scores are comparing its internal graphics so no need them

hope IMC is stronger

[kl0012]

What I found interesting in THG preview is Sisoft Sandra FMA results. Haswell does perform better than IB/SB but not nearly good enough when you look at the "theoretical" boost FMA could bring- they gained only around 18% of FP performance Vs 256bit AVX codepath.

Probably Sandra2013 doesn't support FMA3 yet (only FMA4), so the IPC improvements are responsible for current perf. gain.

Impressive improvement in graphics perf. (considering that this is just GT2). Waiting for my next laptop with GT3 (or, hopefully, convertible with passive cooling).

[LeoAndrade]

Hope that it releases soon. I'll be out of the US by july and hope to buy one before

[informal]

Probably Sandra2013 doesn't support FMA3 yet (only FMA4), so the IPC improvements are responsible for current perf. gain.

Impressive improvement in graphics perf. (considering that this is just GT2). Waiting for my next laptop with GT3 (or, hopefully, convertible with passive cooling).

I was under impression that sisoft sandra was the 1st piece of desktop commercial software to support both FMA3(FMA4 on BD/PD) and AVX2.

edit:
From THG article:

Why do doubles seem to speed up so much more than floats on Haswell? The code path for FMA3 is actually latency-bound. If we were to turn off FMA3 support altogether in Sandra?s options and used AVX, the scaling proves similar.

So if they have option to turn off FMA3 code path then it's logical it actually has support for this codepath.

edit2:
Found this on sisoft facebook page , it definitely supports FMA3.

We are pleased to announce the release of SP1 for SANDRA 2013! Because we managed to test Haswell (Crystal Well), we have some fixes and optimisations:

- AVX2 integer support (Multi-Media integer, Hashing)

- FMA3 float support (Multi-Media float/double, Memory/Cache bandwidth) [also works on Vishera]

- TSX (HLE/RTM) support (Transaction rate)

- eDRAM/L4 support (Memory, Cache, Latency)

SP6 for Sandra 2012 with all the fixes and enhancements of Sandra 2013 (but not new features) has also been published.

[kl0012]

I was under impression that sisoft sandra was the 1st piece of desktop commercial software to support both FMA3(FMA4 on BD/PD) and AVX2.

In the link I posted they explicitly mentioned FMA4, but no FMA3.

[informal]

Check my edit number 2

[R101]

What about the RAM bandwidth issue? Pre-production cpu/board, bad BIOS or..

[kl0012]

Check my edit number 2

That's the strange part - float on AVX should be ~2x faster then double since both are executed by the same hardware using the same registers, and this is true for Ivy and Sandy). It's not clear why it's not true for Haswell. Latency is not important here.

[ajaidev]

Higher TDP than IVB is bad, the more TDP room give the CPU more space to turbo in. The problem is off course the higher TDP at 84 its much higher than the 77 than IVB is rated at, if the mobo does not bring in some TDP saving it would be TDP higher than IVB.

[zalbard]

Higher TDP than IVB is bad, the more TDP room give the CPU more space to turbo in. The problem is off course the higher TDP at 84 its much higher than the 77 than IVB is rated at, if the mobo does not bring in some TDP saving it would be TDP higher than IVB.

Pretty sure I've read somewhere that Intel is taking a rigorous route of Apple [Macbooks] to fine tune motherboard specs down to the smallest parts to make them as efficient as possible.

[BenchZowner]

Not bad, given the power consumption reduction things look better ( than pure clock-per-clock performance-wise )

[eXa]

Wich power consumption reduction?

[god_43]

does not look that impressive, its more or less on par with ivb. i dont see much reason for sb or ivb users to upgrade.

[cobra_kai]

Pretty sure I've read somewhere that Intel is taking a rigorous route of Apple [Macbooks] to fine tune motherboard specs down to the smallest parts to make them as efficient as possible.

Anand wrote about it in his Haswell preview.

http://www.anandtech.com/show/6355/i...l-architecture

From the article it sounds like they really targeted the ULV parts for ultrabooks/tablets for the platform power consumption, not so much desktop or standard laptop parts.

[BababooeyHTJ]

Seems like I will grow old with my 2600K. Does Intel really not want people with a Sandy or Ivy Bridge desktop chip to upgrade? Give us at least a 6 core as a minimum already!!! And I don't need silicon spend on an igpu.

Exactly, I've been using a quad core cpu for the longest time now. Its too bad that they still don't have any plans for a hex core cpu on their mainstream platform.

[poseur04]

Anand wrote about it in his Haswell preview.

http://www.anandtech.com/show/6355/i...l-architecture

From the article it sounds like they really targeted the ULV parts for ultrabooks/tablets for the platform power consumption, not so much desktop or standard laptop parts.

Exactly what I think
Intel work hard on there graffic integration

[R101]

Just found this, it seems numbers are not quite on par with what should be expected of final product:

http://twitter.com/IanCutress/status/313593122103885824

[geo]

Just found this, it seems numbers are not quite on par with what should be expected of final product:

http://twitter.com/IanCutress/status/313593122103885824

Francois Piednoel says,

haswell major improvements come when you are more power constrained, nobody else is able to make a CPU that fast

i think the numbers will be on par with the THG review desktop processors aren't really power constrained as ulvs.

[Buckeye]

If everyone assumes that the numbers are right and a certain person says nope then what CPU where they testing in that review ?

[massman]

Not sure who to trust: THG or FP.

We all knew Intel would come out and say the review is not entirely correct - this response is nothing but expected. Whether it actually says something about the review, each and everyone one of us can decide that for themselves. As far as I'm concerned: I've seen this type of message in the past and it always turned out that the performance figures with ES were (close to) spot on anyway.

Oh well, just an opinion ...

[STaRGaZeR]

Not sure who to trust: THG or FP.

Easy: don't trust anybody until silicon is on the street.

[aznsniper911]

Hopefully it overclocks quite a bit better, considering it's built on revised 22nm tech process.

If it doesn't overclock better then it'll be another dull year for me!

[paulbagz]

Are these still on track for a June release?

-PB

[zalbard]

Are these still on track for a June release?

-PB

In limited quantities due to USB bug.