Haswell model specs leaked

[Smoblikat]

One thing Intel could do is sell unlocked multiplier dual core chips.

Add $15 premium to Pentium or i3 and I would definitely consider buying.

Or intel could stop being a greedy buttface and make all of their procs unlocked like AMD.

 

[NTMBK]

Remember how everyone was saying those roadmaps were "fake"? Well Olivon just posted this link on XS:
http://chinese.vr-zone.com/45571/in...roduction-date-before-computex-2013-12252012/

Looks like model numbers were correct after all. Launch is according to the above slides on June the 2nd.

That's the same website as all the other rumours in this thread, Chinese VR Zone. I'm still treating it as suspect until I see confirmation from other sources.

 

[ShintaiDK]

Or intel could stop being a greedy buttface and make all of their procs unlocked like AMD.

And history shows we would suffer from counterfeits again. Thats why unlocked needs to be the highest bin.

 

[shadow_k]

http://www.techpowerup.com/177817/Intel-Haswell-and-Broadwell-Silicon-Variants-Detailed.html

Seems like desktop computers won't be getting GT3 :|

Broadwell looks very promising.

 

[Axon]

I'll wait for the benchmarks, but none of this is surprising.

 

[Idontcare]

Or intel could stop being a greedy buttface and make all of their procs unlocked like AMD.

And history shows we would suffer from counterfeits again. Thats why unlocked needs to be the highest bin.

Or Intel could just become a non-profit company, the antithesis of a greedy buttface, and sell ALL their chips "at cost" and put the counterfeiters out of business

Maybe one day UNICEF will get into the x86 CPU business...until then, Intel is the company to see

 

[ShintaiDK]

Or Intel could just become a non-profit company, the antithesis of a greedy buttface, and sell ALL their chips "at cost" and put the counterfeiters out of business

Maybe one day UNICEF will get into the x86 CPU business...until then, Intel is the company to see

LoL

Then we can donate i5 3570K to africa for 10-15$. :awe:

 

[thm1223]

I'll wait for the benchmarks, but none of this is surprising.

Based on the info in this thread can people reasonably infer what the CPU improvement will be between, say, the 3770k and Haswell 4770k?

 

[Axon]

Based on the info in this thread can people reasonably infer what the CPU improvement will be between, say, the 3770k and Haswell 4770k?

Not from this thread, no. There have been some hints here and there that IB was not worth the upgrade over haswell by reviewers, but we will see. With AMD down for the count, I doubt Intel cares.

 

[ShintaiDK]

Based on the info in this thread can people reasonably infer what the CPU improvement will be between, say, the 3770k and Haswell 4770k?

Depends what you run. Linpack for example will be over twice as fast due to the entire core is extended from 128 to 256bit width, inclusing 2 extra issue ports. Unoptimized code should be 10-15% at the same clock. Any scalar code should easily be converted to AVX2 for twice the speed.

 

[NTMBK]

Any scalar code should easily be converted to AVX2 for twice the speed.

Why would scalar code benefit from wider vectors?

(Also, not a single game or application to date has shipped with AVX2, so that "legacy" category is going to be "the majority of programs you run for the next several years" )

 

[inf64]

Yep if we are lucky AVX2 will be supported in 2015 as AVX is today .

 

[ShintaiDK]

Why would scalar code benefit from wider vectors?

(Also, not a single game or application to date has shipped with AVX2, so that "legacy" category is going to be "the majority of programs you run for the next several years" )

AVX2 is vector instructions of the current scalar. Plus added gather support that also replaces 17 old instructions. And its very easy to convert current scalar code for AVX2. AVX1 code that we got today will also benefit greatly.

 

[ShintaiDK]

Yep if we are lucky AVX2 will be supported in 2015 as AVX is today .

AVX2 code have been produced for quite some time already tho in compilers.

 

[inf64]

I meant commercial desktop applications that desktop users use daily . If we see ONE that supports AVX2 on June the 2nd I would be really surprised (pleasantly). AVX is pretty useful for fp vector stuff and only a few desktop apps support it today(I don't refer to server market,that's a different ballpark).

 

[mikk]

That's the same website as all the other rumours in this thread, Chinese VR Zone. I'm still treating it as suspect until I see confirmation from other sources.

Here is another source for you: http://diy.pconline.com.cn/cpu/study_cpu/1212/3115955.html


Vr-zone infos were legit from the beginning, that's how a model list can look like from Intel. No surprise there.

 

[tweakboy]

Weak... Man Ivy Bridge E is going to spank all these haswell processors. GUys Haswell is only good for onboard video. Other then that, its a pointless pos from what I see.

They cant even make a 6 core 12 thread version......
They just optimized the video and its really fast now that you can play games on it. That is all Haswell is........... My CPU will spank these 4/4 and 4/8 but only 8MB cache,,,sheez.. gl

 

[NTMBK]

AVX2 is vector instructions of the current scalar. Plus added gather support that also replaces 17 old instructions. And its very easy to convert current scalar code for AVX2. AVX1 code that we got today will also benefit greatly.

Any code amenable to vectorisation will already be using SSE2+, not scalar. If the company is interested in vectorisation, but an intensive section of code is still running in scalar instead of SSE, then that implies that the algorithm is not amenable to vectorisation.

Gather will definitely be very useful in the right workload. It needs to be something with unpredictable/unaligned memory accesses by nature though (which is why graphics cards got this long ago, while CPUs took this long and still lack scatter).

The other biggy is FMA. If anything this will be even more useful, as it can be quickly dropped into almost any FP heavy workload as you're bound to have a few multiply-then-add sequences in there. (And thankfully Piledriver already supports Intel's version, FMA3, as well as their own FMA4 implementation, so they won't get into another stupid instruction set war like they did with 3dNow!.

Gather and FMA could certainly speed up code written for AVX1. However, the most useful feature of AVX2 is that it now handles integers. This was the obvious missing point of AVX1 (same way it was on SSE1), and that addition helps a lot with making AVX more useful in all cases.

I'm very interested to see what the performance of some of the media encoders/transcoders will be like when optimised for AVX2.

 

[Idontcare]

Based on the info in this thread can people reasonably infer what the CPU improvement will be between, say, the 3770k and Haswell 4770k?

For single-threaded apps, according to Intel's published data they are expecting ~12% increase in IPC across "a large suite of applications".

Corner cases will exist, of course, where you will see radically higher IPC increases, and vastly lower IPC increases.

 

[grandpatzer]

There's already definite gains to faster memory on Trinity- it's graphics performance will scale with memory bandwidth as far as DDR3-2133, whereas IB's graphics seems bandwidth saturated at ~DDR3-1600.

Is that in the integrated gpu?

I have sandy bridge system with crossfire AMD 7950, I guess I would not see that big improvement going from 1333mhz to 1600mhz?

 

[NTMBK]

Is that in the integrated gpu?

I have sandy bridge system with crossfire AMD 7950, I guess I would not see that big improvement going from 1333mhz to 1600mhz?

Yes, that is with the integrated GPU. The CPU cores aren't so starved for memory bandwidth.

 

[Bubba77]

Meh. Dont wanna wait till june to build. Sitting on a bunch pf newegg gift cards and i wanna put a new system together. My case is falling apart and only 1 usb port still works! Even tho its only an 3 year oldi7 960 I have had the new build itch for a while.....

 

[cbn]

Some info on overclocking Haswell:

http://www.legitreviews.com/news/14176/

However, high end overclocking features that allow greater BCLK tuning and which currently only appear in the top end, pricey SB-E LGA 2011 platform will be present in the mainstream Haswell desktop and laptop parts too. Significantly, the BCLK coarse ratio adjustments of 1.00, 1.25 and 1.67 which allow BCLK to change without unwanted overclocking of those SATA ports and other components, will feature in the new Haswell mainstream parts. This will help with clock speed granularity, since simply adjusting the multipliers along with limited BCLK as happens now on Sandy Bridge and Ivy Bridge CPUs leads to large clock speed jumps, preventing optimum fine tuning of system performance.

 

[ShintaiDK]

Some info on overclocking Haswell:

http://www.legitreviews.com/news/14176/

That news is 3 months old now tho

 

[Fjodor2001]

http://www.techpowerup.com/177817/Intel-Haswell-and-Broadwell-Silicon-Variants-Detailed.html

Seems like desktop computers won't be getting GT3 :|

Broadwell looks very promising.

How can you tell whether Broadwell will be promising or not by just looking at that table?