Intel 8-core Haswell-E Slated For 3Q 2014

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
$1000 chips, is that all?

I believe my i7 3920XM was 885 GBP on top of the base i7 for my system, plus VAT.

I got it upon release.

But yer gets what ya pay for. 4.7GHz on a laptop is kinda special I think...

Gotta feed those hungry GPUs...
 

The paste Intel puts under the IB/Haswell IHS is already better than just about anything else on the market. The real problem is the thick gap between the CPU die and IHS because the IHS is propped about 100 microns off the CPU die by the glue/resin/epoxy bonding the IHS to the CPU substrate. The temperature improvement people see by replacing the stock paste with (actually worse) aftermarket paste comes from removing most of that ~100um gap.
 
More interested in the 6-core CPU than anything else. For some reason, I'm imagining myself saying something along the lines of "How much? Shut up and take my money!"
 
According to the leaks, Haswell-E will only officially support up to DDR4-2133, which is 1066 MHz = 2133 MT/s.

IneEZ7K.jpg
 

The very name and definition of DOUBLE Data Rate implies and states that the data rate is double the clock rate. This does mean 1066MHz for DDR4's starting 2133MT/s data rate.
 
That site is describing the speed of DDR4 that will be avaiable. But that's not the same as what Haswell-E will officially support.

Just like regular Haswell only officially supports DDR3 up to the speed of 1600 MT/s , even though there is DDR3 on the market at 3000 MT/s and even a little higher.

And generally Intel is very conservative with its official support specification. Haswell can in practice use memory a lot faster than DDR3-1600. Stands to reason Haswell-E will be able to exceed its official spec a lot too.
 


thanks a lot.:)

 
I am REALLY REALLY waiting and hoping for that X99 + new CPU to finally upgrade my high-end X58 RIG. I almost did it when X79 Came out but I decided to get maximum for my Buck and skip generation and go straight from X58 to X99, but 1K CPU ? Come on when let's say i7 920 came out it was less than $500 and yes may be Extreme Edition latest and greatest then yes 1K is a may be , then again right this second you can go spend 1K on CPU if you don't know how to OC your regular CPUs 😀
 


That is why I'm waiting on the 6 core Haswell-E. It's going to be interesting to see how these are priced out once the market settles down after launch. I'm interested in seeing just how good these CPUs are going to perform compared to the refreshed versions, or even the current ones.

But what's tugging away at me right now is the X99. I want that. Extra cheese, please? Thank you.
 

One thing though: X99 doesn't seem to come with SATA Express support. Kinda disappointing.
 

I would expect it to follow the same price points Intel has been using for the past ~10 years with most retailers sticking closely to Intel's MSRP as they have been doing for the past 7+ years.

Since Haswell-E is based on the same process as IB-E and I cannot think of any reason for Intel to make it any cheaper particularly considering how Intel has not discounted previous-gen chips on new product launches in ages and IB-E was priced the same as SB-E even with the process shrink, it will most likely get priced exactly the same or slightly above so expect hex Haswell-E to cost the usual ~$600 for mid-range enthusiast/prosumer chips. The $300 price point is already occupied by the i7-4770 so there won't be a hex-core Haswell-E anywhere near that price point.

Skylake is where things should start getting more interesting with Intel's commitment to making quads the new entry-level standard for mainstream desktop chips, which means something along the lines of i3 gaining two cores but losing HT, i5 either gaining two cores or HT, i7 gaining two cores and i7-E up to four.
 

More threads is better for heavily threaded workloads and nearly all modern server CPUs do support 2-8 threads per CPU core. IIRC, the UltraSparc T4 supports up to 8 active contexts out of a hardware pool of 32 contexts. Intel's Xeon Phi uses four threads per core and forgoes most hardware associated with single-threaded performance like branch prediction and out-of-order execution. HyperThreading at only two contexts per core is only the tip of the iceberg; at the other end of the spectrum, you have high-end GPUs these days that support 2000+ threads per die.

So, more threads certainly has its place.

Why does Intel feel the need to raise the bar on core count with Skylake? I do not expect mainstream software to require that much parallelism within the next five years or so assuming developers really start pushing that aggressively so my take is that Intel is simply running out of options to make people want to upgrade their PCs and at the rate ARM SoCs are catching up, laptop/PC CPUs will need to be at least that powerful to justify their price premium over embedded alternatives in situations where x86 compatibility is either optional or unnecessary: people who are not shackled to x86 or Windows will not be impressed by Wintel-based products that are only somewhat more powerful than their phone or tablet and this is a quickly growing chunk of the market.

Skylake is still nearly two years away. By then, SoCs will likely be another 3-4X as fast as they are today and that brings them uncomfortably close to current mainstream PCs' level which are already overkill for half of people's needs.
 
Meaning to say more cores is well ahead into the future & more threads in a CPU is now irrelevant, since the GPU is supplying more threads per die.
 

More threads in CPUs is how just about everyone in the server and high-performance computing does it. If you look at Intel's HT history, it only costs ~5% extra core surface area to unlock the 30-40% of under-used execution resources on the core that cannot be filled up with only a single instruction stream going through the scheduler. If multi-threaded programming ever becomes pervasive in a meaningful way across the industry, hardware-threaded CPUs will actually become more important than ever since scheduling multiple in-order superscalar instruction streams through the cores is much simpler and more efficient than having to deal with branch (mis)prediction and deep out-of-order speculative execution that make most modern applications heavily dependent on single-thread performance.

What are universal/GPGPU shaders on a GPU? Really dumbed-down CPUs with lots of hardware threads to keep execution pipelines busy even when individual threads are stalled... hyperthreading/SMT on steroids.

What did Intel do with Xeon Phi? They ditched most of the circuitry used to enhance single-threaded performance on regular CPUs in favor of twice as many hardware threads and 6X as many of those dumbed-down cores. The overall throughput is great for HPC and other similar applications but would be horrible for everyday desktop use with so much software being so heavily dependent on single-threaded performance.

If you break the shackles of dependence on single-threaded performance, just about everything about how desktop CPUs are currently being designed (for single-threaded performance) goes out the window.
 
Statistic says single threaded is not the norm in the computing industry, Intel is dominating the desktop mainstream, enthusiast & server computer so they claimed. Majority of computer user weather private or in government is using Intel product here in the US. So, actually majority is using multi-thread performance.
 
Status
Not open for further replies.