[VR-Zone] Mainstream desktop CPUs future evolution - Haswell

[dma0991]

More performance or more integration?

• Performance enhancements like 20%++ per-core speed jump after counting the slight IPC improvements and frequency gains, as well as at least double the graphics performance
• Reduced power consumption
• LGA1150 for desktops and rPGA947 & BGA1364 for the mobile market for Haswell
• Increased instruction per cycle parallelism with new execution ports and better prefetching & branch handling for per-thread performance improvement
• Voltage regulators are to be fully integrated in the CPU
• Display ports are all right there on the CPU die

Intel doesn't seem to show any signs of slowing down despite AMD's lack of competition. Certainly not much is gained in the performance end but definitely not lacking in trying to integrate more into the processor. I wonder if Intel is consolidating to create a SoC like processor. :hmm:

 

[lehtv]

Interesting. I wonder if integrating the voltage regulators fully in the CPU is going to have any effect on overclocking?

 

[OVerLoRDI]

Interesting. I wonder if integrating the voltage regulators fully in the CPU is going to have any effect on overclocking?

My thoughts exactly. It seems like Intel goes back and forthbon whether or not it wants to support overclocking.

 

[boxleitnerb]

They see 20% in each step (SB->IB, IB->HW)...ambitious. Hope this is true and Intel doesn't slow down.

The OC thing is weird, true. But I guess the VRM thing is just about cost, not specifically for or against overclockers. Still, with the baseclock being decoupled, coupled, decoupled...decide please and stick with it.

 

[piesquared]

It's obvious intel is headed in the same direction AMD is with Fusion, SOC. Haswell will be competing against the APU after Trinity on 28nm, with much more integrated GCN so there's a significant obstacle ahead.

 

[Silverforce11]

Intel needs to improve even with no competition, because their previous gen are their biggest competitor. Why should people upgrade if it doesn't offer a big improvement?

 

[podspi]

More performance or more integration?

Intel doesn't seem to show any signs of slowing down despite AMD's lack of competition. Certainly not much is gained in the performance end but definitely not lacking in trying to integrate more into the processor. I wonder if Intel is consolidating to create a SoC like processor. :hmm:

As CPUs become a smaller proportion of the total system cost, we'll see more integration in an attempt to capture a higher percentage of the system value.

 

[dangerman1337]

So there's going to be another socket change? I guess this means that Socket for Enthuasist will change as well.

 

[BD231]

More performance or more integration?

Intel doesn't seem to show any signs of slowing down despite AMD's lack of competition. Certainly not much is gained in the performance end but definitely not lacking in trying to integrate more into the processor. I wonder if Intel is consolidating to create a SoC like processor. :hmm:

Familiar with SoI, remind me what SoC is again?

 

[Idontcare]

Familiar with SoI, remind me what SoC is again?

SOI = Silicon on Insulator

SOC = System on a Chip

 

[StrangerGuy]

"Voltage regulators are to be fully integrated in the CPU"

I wonder what Asus & Co is going to blow hot air on this time, now that the last bastion of marketing fluff is dead.

 

[GammaLaser]

The way I understand it, the mobos will still need a VR (and manufacturers can still differentiate themselves with VR design) but the mobo will only deal with regulating two voltage supplies going into the CPU (memory + everything else).

 

[IntelUser2000]

Interesting. I wonder if integrating the voltage regulators fully in the CPU is going to have any effect on overclocking?

The same site says it should be easier to overclock: http://vr-zone.com/articles/intel-cpus-overclocked-sandy-easy-ivy-easier-haswell-easiest-/13733.html

 

[dma0991]

Leaked slides reveal Intel Haswell platform

More leaked info about Haswell. The max TDP for desktop platform have increased from 77W Ivy Bridge to 95W Haswell due to 2 extra cores(?).

Edit: Probably not

The Haswell desktop processors will have two or four cores and a TDP between 35W and 95W.

Does that mean that there will be no 6 core Haswell? :hmm:

 

[Arachnotronic]

The enthusiast platform will get the 6+ core HSW. Did you guys honestly expect Intel to do "moar cores" in the mainstream? After the shit BD got? No, it's all about IPC/power consumption.

 

[lol123]

I'm happy to see there will be a 95W desktop SKU. That means they won't be completely focused on scaling down for power efficiency improvements. For 6 and 8 cores we will probably have to wait for the enthusiast/performance server lines, which I also think is the right way to go.

Also keep in mind that we very well might see 4-way hyperthreading as well as increased execution width in Haswell.

 

[Edrick]

What is Hotham 1.0 support?

Those slides are a good indication of what we will see in the platform. I am really interested to see what uArch improvements have been made over SB/IB.

 

[bryanW1995]

@lol123: why is limiting core counts on desktop the right way to go? Because AMD has ceded the high end, just like they did to nvidia in graphics. AMD's failure will be the end of our hobby...

 

[lol123]

@lol123: why is limiting core counts on desktop the right way to go? Because AMD has ceded the high end, just like they did to nvidia in graphics. AMD's failure will be the end of our hobby...

Because 4 fat cores (especially with improved SMT) are better than 8 weak cores. In the Haswell-E family there will definitely be an 8 core SKU, but that will be with a larger die, higher costs, somewhat lower clock frequency and higher TDP than in the mainstream line. Those are trade-offs that simply have to be made and I don't see why it is a problem that Intel provides both options to their customers.

AMD chose to combine all these trade-offs and resulting disadvantages in a single processor family (and with weak cores to start with) and the result is a disaster. I don't see how that helps the overclocking or enthusiast hobby at all.

 

[Edrick]

why is limiting core counts on desktop the right way to go?

Because for the average desktop user, 4C 8T is more than enough at the current time. Of course we will see more cores in the mainstream within the next 5 years, but not next year.

And users that need 6/8/10 cores, there will always be a the E series platform to choose from, and naturally comes with a higher price tag.

 

[Concillian]

I'm happy to see there will be a 95W desktop SKU. That means they won't be completely focused on scaling down for power efficiency improvements.

- Putting the VRs into the CPU will eat some of the TDP budget that isn't eaten now.
- Better graphics = larger TDP dedicated to graphics.

Your conclusion isn't necessarily correct.

The VR integration is very interesting. Mobos will presumably only have to regulate 3 voltages then. The CPU voltage will likely be significantly higher than existing voltage, which will allow for significantly better power handling of the socket (required current would decrease). This is possibly where the loss in pins is coming from compared to 1155/1156. Mobo MFRs will have very little to work with other than checkbox features to distinguish themselves (and heatsinks that look like guns / bullets / boobs). That could end up good for consumers, even if Intel jacks up prices on the chips / motherboards that have unlocked voltages. Who even cares about multipliers or Bclk locking if the voltages are locked.

 

[lol123]

- Putting the VRs into the CPU will eat some of the TDP budget that isn't eaten now.
- Better graphics = larger TDP dedicated to graphics.

Those two design features will apply (and I would say more or less equally) to the 35W, 45W and 65W SKUs and their TDP rating as well, so it's safe to say that they are not a deciding factor behind (increased) TDP. If they were, performance/power scaling like that with the same die would not be possible.

 

[gevorg]

I wonder if there will be low-power Haswell variants that can be passively cooled even at load. Would be great for HTPCs without any fans or moving parts.

 

[Concillian]

Those two design features will apply (and I would say more or less equally) to the 35W, 45W and 65W SKUs and their TDP rating as well, so it's safe to say that they are not a deciding factor behind (increased) TDP. If they were, performance/power scaling like that with the same die would not be possible.

Increased TDP can be used in many ways. The 35W version probably will have lower GPU clocks and lower CPU clocks. This will be usable at lower voltage and thus lower current. This would reduce VRM heat, GPU heat, etc...

A VRM designed to handle 95W requirements is going to put out significantly less heat at 45W, and so the VRM budget will scale with the TDP, at least to some to extent. It's like a PSU, it's waste heat scales with the power required. Efficiency changes with load as well, but the major contributor to the heat the PSU produces is the amount of power it needs to provide.

Also, if you look at the existing lineup, the 2400s is slower than the 2500k on the CPU side, but also has half the EUs on the GPU side. All portions of the CPU are likely to scale up / down with the TDP.

 

[exar333]

I wonder if there will be low-power Haswell variants that can be passively cooled even at load. Would be great for HTPCs without any fans or moving parts.

You can already do this with Celeron and Pentium SB-CPUs. Haswell derivitives would likely be even lower power with even better performance. Great HTPC options with this chip...