Broadwell, is there a point?

[VirtualLarry]

Broadwell, is just a 14nm die-shrink of Haswell, right? Nothing new architecturally? Are there going to be Broadwell-K chips, that might overclock better than Haswell-K chips? Or are they primarily for mobile (laptops, convertible full-size tablets like MS Surface Pro 3, etc.)?

 

[master_shake_]

power consumption is the point

that computer you are using is a dinosaur get on a tablet and consume!

j/k

 

[tweakboy]

Skylake LOL

 

[Hulk]

If history repeats itself then Broadwell will be a die-shrink with minor CPU improvements over Haswell.

That being said...

In his Haswell architectural article from October 5, 2012 when writing about Haswell's GPU Anand wrote "the big plumbing/redesign will be Broadwell in 2014."

When Anand drops bit hints like that he's usually on the money so I'm expecting minor CPU improvements with Broadwell and something more significant on the integrated GPU front.

 

[Lepton87]

Any source that it's just a die shrink? The last time Intel did a straight die shrink was Nehalem>Westmere, otherwise we always had IPC improvements in the 5-10% range

 

[CHADBOGA]

Broadwell on the desktop will hardly be a compelling upgrade for anyone who has a Sandy Bridge or newer CPU.

The IGP will be improved, but not to the point to impress any gamers, & I suspect very little IPC improvement.

It might overclock better, as I have heard that the hit to overclocking which came with FinFets, only comes once, so perhaps the 22nm to 14nm transition will give us the same overclocking gains as the 45nm to 32nm transition.

 

[NTMBK]

The IGP is the big change. A while back there was a commit log to the Linux Intel drivers listing a lot of the changes to Broadwell's IGP, and they sounded pretty big.

Other than that, yeah, power consumption. It's meant for laptops and tablets.

 

[Homeles]

I'm personally expecting 14nm to overclock well in comparison to Haswell and Ivy Bridge. Maaaaybe it'll overclock higher than Sandy Bridge. Other than that, there's not really much to add to what others have said.

Any source that it's just a die shrink? The last time Intel did a straight die shrink was Nehalem>Westmere, otherwise we always had IPC improvements in the 5-10% range

Westmere wasn't actually a straight die shrink either. The changes just weren't noticeable for general usage. AES-NI is pretty huge though. Encrypting drives without it is rather painful.

 

[Venomous]

Broadwell on the desktop will hardly be a compelling upgrade for anyone who has a Sandy Bridge or newer CPU.

The IGP will be improved, but not to the point to impress any gamers, & I suspect very little IPC improvement.

It might overclock better, as I have heard that the hit to overclocking which came with FinFets, only comes once, so perhaps the 22nm to 14nm transition will give us the same overclocking gains as the 45nm to 32nm transition.

Well I will have to disagree with you here about sandy... She's already bottlenecking with the high end gpu's as is. Until everyone is onboard with one API that works across all hardware, we will always need an upgrade path consisting of a new chipset and CPU. Those little 10% IPC bumps keep us in the game. When that stops, the entire GPU market will stop and this is as good as it will ever be again.

 

[imported_bman]

If history repeats itself then Broadwell will be a die-shrink with minor CPU improvements over Haswell.

That being said...

In his Haswell architectural article from October 5, 2012 when writing about Haswell's GPU Anand wrote "the big plumbing/redesign will be Broadwell in 2014."

When Anand drops bit hints like that he's usually on the money so I'm expecting minor CPU improvements with Broadwell and something more significant on the integrated GPU front.

I wonder if the IGPU will be bandwidth starved even with a L4 cache. It will be interesting to see what boost Skylake will get from using DDR4. I have been waiting a while to buy an Ultrabook that is meets certain standards, which are now starting to be met (1440p+ res, 10+ battery life, ssd, 8gb ram, wifi ac). Hopefully Intel gives a road map at IDF 2014 spring, then I can decide if to wait for Skylake or buy a Broadwell Ultrabook.

 

[VirtualLarry]

Well I will have to disagree with you here about sandy... She's already bottlenecking with the high end gpu's as is. Until everyone is onboard with one API that works across all hardware, we will always need an upgrade path consisting of a new chipset and CPU. Those little 10% IPC bumps keep us in the game. When that stops, the entire GPU market will stop and this is as good as it will ever be again.

You're speaking of gaming. What if mining isn't just a fad, and, in the future, it drives GPU growth?

After all, you can mine on what, 7 R9 290x with an AMD Sempron CPU?

If I were AMD, i might actually consider adding a few fixed-function hashing features to their future GPUs.

 

[Venomous]

You're speaking of gaming. What if mining isn't just a fad, and, in the future, it drives GPU growth?

After all, you can mine on what, 7 R9 290x with an AMD Sempron CPU?

If I were AMD, i might actually consider adding a few fixed-function hashing features to their future GPUs.

Well miners hardly use bandwidth though. It would still affect the crowd because development would begin a decline. Motherboards that offer all these extra pcie slots would start to trickle off and the domino effect begins.

All I'm trying to say is, until the bottlenecking addressed on all gpu's, together as one API, were going to need CPUs/chipsets to accomplish it in current times.

 

[MisterMac]

Here's to hoping 14nm will be the tank that mature 32nm was.

My logic says no tho - because it's a new process what's to say it will OC well?

Historicly all good OC nodes were mature ones - weren't they?

 

[mrmt]

Here's to hoping 14nm will be the tank that mature 32nm was.?

It's not mainly a question of maturity, but what kind of performance targets one wants to achieve within the framework of node design. That said, Intel's Mark Bohr already stated that 14nm will be optimized for the mobile space, far more than any other Intel process was. What's left for discussion then?

 

[imported_Thorburn]

Any source that it's just a die shrink? The last time Intel did a straight die shrink was Nehalem>Westmere, otherwise we always had IPC improvements in the 5-10% range

Nehalem>Westmere wasn't a straight die shrink - even ignoring the increase in core count you had the addition of AES-NI and a bunch of Virtualisation improvements in the CPU cores.

 

[Homeles]

Here's to hoping 14nm will be the tank that mature 32nm was.

My logic says no tho - because it's a new process what's to say it will OC well?

Historicly all good OC nodes were mature ones - weren't they?

Penryn overclocked well, and that was introduced on the brand new 45nm process.

I don't know about prior to that. 45nm was great because of HKMG, which is something the industry desperately needed to combat leakage current. It also had tremendously improved PMOS Idsats, likely due to continued improvement of their strained silicon. 32nm, despite introducing "nothing new" technologically, brought similar improvements over 45nm that 45nm had brought over 65nm.

22nm has been a real stinker because FinFETs perform worse than planar FETs at high voltage.

Essentially, I'd argue that it's got little to do with maturity, and has lot more to do with the underlying microarchitecture and the innovations introduced with that node. I think the importance of process maturity is overstated in the tech journalism industry. When we see products launch from Intel, AMD, Nvidia, and others, the nodes are fairly mature at those times.

 

[psolord]

Broadwell on the desktop will hardly be a compelling upgrade for anyone who has a Sandy Bridge or newer CPU.

The IGP will be improved, but not to the point to impress any gamers, & I suspect very little IPC improvement.

It might overclock better, as I have heard that the hit to overclocking which came with FinFets, only comes once, so perhaps the 22nm to 14nm transition will give us the same overclocking gains as the 45nm to 32nm transition.

Both my Q9550 and I7 860, being 45nm parts, run happily at 4Ghz. My 2500k can run happily at 4.8-5.0 Ghz. So broadwell should be able to hit around 5.5Ghz in order to have a similar increase in max clock speed.

5.5Ghz along with a 25% IPC increase over Sandy, would peak many users' interest into upgrading. Still support for very fast ram, or a tripple channel solution, would be required to actually give this ipc and clock increase, into true performance.

 

[Hulk]

I'm going to make some Broadwell predictions right here in writing so I can look back and see how I did.

CPU performance vs. Haswell - Barring any new instruction based performance increases I'm betting on a 2 to 3% improvement in IPC over Haswell. This is based on Haswell just receiving an improved branch predictor, wider execution engine, 2x increase cache bandwidth, and better memory controller. Unless Intel can somehow pull a rabbit out of it's hat I think we're topping out on legacy code IPC-wise. The low hanging fruit has been gone for 5 years now and it's "blood from a stone" time.

iGPU - 50% increase in performance across the board when looking at new version of HD Graphics, 4000 series, and 5000 series. ie 4600 moves from 20EU's to 30. Perhaps some new architectural improvement will provide 10% additional performance gains over the 50% I quoted above.

Overclocking - No change from Haswell. Intel seems to be having enough trouble with 14nm so I don't expect much headroom from a frequency perspective. And the heat transfer issue which began with Ivy at 22nm will only be exacerbated at 14nm.

Those who felt Haswell was disappointing are going to be even more disappointed with Broadwell as I think it will offer virtually nothing for the desktop.

For mobile users on the other hand I think it will be great. I'm projecting 25% better battery life compared to Haswell at the same performance level and better overall CPU and GPU performance due to 1). Lower power usage means higher clocks are possible in mobile, and 2). Improved GPU.

 

[The|Hunter]

This is the only real difference vs Haswell arch wise.


Broadwell will introduce some instruction set architecture extensions:

• Intel ADX: ADOX, ADCX and MULX for improving performance of arbitrary-precision integer operations
• RDSEED for generating 16-, 32- or 64-bit random numbers according to NIST SP 800-90B and 800-90C
• PREFETCHW instruction.

 

[Ken g6]

I'm hoping Broadwell will bring a CPU with a Crystal Well cache to the desktop. But I have my doubts that it will. 🙁

 

[Sweepr]

I'm hoping Broadwell will bring a CPU with a Crystal Well cache to the desktop. But I have my doubts that it will. 🙁

According to this (rumour?) Broadwell-K will feature GT3e graphics. A $235 ''Core i5 5670K'' with slightly better CPU performance and a big leap on the graphics side (vs Haswell GT3e and especially Haswell GT2) would make one hell of an interesting chip. Comparisons should be fun. 🙂

 

[2is]

I'm personally expecting 14nm to overclock well in comparison to Haswell and Ivy Bridge. Maaaaybe it'll overclock higher than Sandy Bridge. Other than that, there's not really much to add to what others have said.

What are you basing these expectations on? It's been years since we've seen a die shrink from intel give better overclocking results.

 

[witeken]

iGPU - 50% increase in performance across the board when looking at new version of HD Graphics, 4000 series, and 5000 series. ie 4600 moves from 20EU's to 30. Perhaps some new architectural improvement will provide 10% additional performance gains over the 50% I quoted above.

Broadwell GTx will have 20% more EUs. So GT2 will have 24 EUs, GT3 48, and so on. But, the default GT will increase from GT2 to GT3. So the i7 5770 will have 48EUs, 2.4 times more than Haswell.

Also, Broadwell will have Crystalwell. And on top of all that, gen8 will bring huge architectural improvements.

Overclocking - No change from Haswell. Intel seems to be having enough trouble with 14nm so I don't expect much headroom from a frequency perspective. And the heat transfer issue which began with Ivy at 22nm will only be exacerbated at 14nm.

There might be some overclocking improvements.

 

[piasabird]

I have seen all these promises of improvement and then you can only get those fancy IGP advancements in embedded motherboards or AIO gear.

 

[PPB]

IMO, it will be something like this:

OC potential: Same as haswell.
Power consumption: Lowered across all (sane) clocks/voltages.
iGP performance: on desktop not much will change, on mobile probably we will get the first decent performing iGP from intel on a sub $400 CPU.

Here's hoping the broadwell iGP improvements reach every segment of the market, and those supposedly big gains are not seen in just some $500 CPU part 0.01% of the market will be buying while the most mainstream processors are stuck with crippled parts.