Intel's Future Chips: News, Rumours & Reviews

[jdwii]

I doubt Intel would risk power efficiency for graphics when the majority are just fine with what they have. Instead they will continue improving both while leaving CPU performance alone or barely touching it.

I guess i do agree with tourist a bit, its kinda sad Intel can't come up with a actual iGPU that competes i think its more to do with drivers, i mean quicksync creams the crap out of Nvidia and Amd in somethings. Wonder how it would do if it was forced to do some kind of physics work?

 

[gamerk316]

To be fair, Intel is TDP constrained. But yes, their driver do stink.

 

[Cazalan]

I doubt Intel would risk power efficiency for graphics when the majority are just fine with what they have. Instead they will continue improving both while leaving CPU performance alone or barely touching it.

I guess i do agree with tourist a bit, its kinda sad Intel can't come up with a actual iGPU that competes i think its more to do with drivers, i mean quicksync creams the crap out of Nvidia and Amd in somethings. Wonder how it would do if it was forced to do some kind of physics work?

There is still a GT4e with 72UE on the lists for Skylake-S. With half the eDRAM though (64MB).

That would make it 50% faster plus whatever architectural improvements over Broadwell.

 

[turkey3_scratch]

Intel would not risk making less efficient, larger, hotter CPUs just for more power. Innovation is managing to make something both more efficient and powerful. Intel has done a fantastic job with their "U" mobile CPUs which draw in about 15W of power.

 

[juanrga]

this one image says it all with all that r&d money and they still cannot equal even a a6 7400k.
Intel is doomed.
http://techreport.com/r.x/intel-broadwell-nuc/thief.png

A 15W Broadwell-U chip achieving the 80% of the gaming performance of 65W Kaveri desktop chip clearly shows that Intel is advancing very fast in the GPU part. There is no doubt that 65W Broadwell will be faster than Kaveri on GPU performance. My doubt remain on if Skylake will caught Carrizo or not. If does then the only advantage of AMD (better graphics) is gone.

a6 7400k is a dual core 5250u is 4 core i5

No. The 5250u is a dual-core.

 

[Turb0Yoda]

this one image says it all with all that r&d money and they still cannot equal even a a6 7400k.
Intel is doomed.
http://techreport.com/r.x/intel-broadwell-nuc/thief.png

A 15W Broadwell-U chip achieving the 80% of the gaming performance of 65W Kaveri desktop chip clearly shows that Intel is advancing very fast in the GPU part. There is no doubt that 65W Broadwell will be faster than Kaveri on GPU performance. My doubt remain on if Skylake will caught Carrizo or not. If does then the only advantage of AMD (better graphics) is gone.

.

AMD will still appeal to Budget Gamers (x4 750k). The only low end enthusiast chip from Intel seems to be the G538. It is the cheapest CPU from Intel that one can overclock(on the LGA1150 socket). Correct me if I am wrong on that. If I had to do a low end build right now, I'd go with AMD... Thier prices on average seem to be lower.

 

[de5_Roy]

I doubt Intel would risk power efficiency for graphics when the majority are just fine with what they have. Instead they will continue improving both while leaving CPU performance alone or barely touching it.

I guess i do agree with tourist a bit, its kinda sad Intel can't come up with a actual iGPU that competes i think its more to do with drivers, i mean quicksync creams the crap out of Nvidia and Amd in somethings. Wonder how it would do if it was forced to do some kind of physics work?

intel isn't leaving cpu performance alone. with haswell-e they re-designed the ring bus, resulting in a scalable architecture that allows for higher core count and for easy fusing off defects. they also designed a low power 8 core die (like avoton) with broadwell cores.

There is still a GT4e with 72UE on the lists for Skylake-S. With half the eDRAM though (64MB).

That would make it 50% faster plus whatever architectural improvements over Broadwell.

imo 32MB would be enough, cheaper too. intel claims to have more gfx performance with bdw's shaders so 72 eus with moar per-shader power would theoretically be quite huge for intel. i'll get back after i research a bit on bdw's igpu. so far i noticed that intel calls haswell's "gen 7.5" - hd4600 to iris pro. from hd5300 (bdw) "gen 8" starts.
edit: aw, and dat 72EU number seems oddly familiar to me. any relation to one of those KL MICs? may be i am making too far a leap...

A 15W Broadwell-U chip achieving the 80% of the gaming performance of 65W Kaveri desktop chip clearly shows that Intel is advancing very fast in the GPU part. There is no doubt that 65W Broadwell will be faster than Kaveri on GPU performance. My doubt remain on if Skylake will caught Carrizo or not. If does then the only advantage of AMD (better graphics) is gone.

i underlined the blanket statements. the 65W denotes a whole class of processors that would operate at that TDP and may have different cpu-gpu-LLC configurations, so your claim has very, very little value.

same with skylake, kaveri and carrizo - what model(s), specs and config, benches? be specific. i just don't want to read the worthless switcheroos done for the sake of trolling others ( e.g. "when i said 65w broadwell, i said about core i7 5770R not core i3 5350 with GT2!").

I'm confused. Are you pointing to me or someone else?

tourist's timing was off. don't worry about it.

 

[juanrga]

The Intel NUC5i5RYK provided us with the opportunity to take a look at what Broadwell-U can deliver when coupled with a motherboard providing premium features. The migration from 22nm to 14nm has allowed for higher base clocks while maintaining the same power envelop. The performance delta over the Haswell-U-based D54250WYKH (particularly, on the graphics side) is noticeable.

http://www.anandtech.com/show/8986/intel-nuc5i5ryk-review-a-broadwellu-ucffpc-for-enthusiasts/8

Note that the reviewed Broadwell model had some BIOS issue and that used slower RAM than other models.

 

[vmN]

http://wccftech.com/intel-preparing-dirsuptive-skylake-microarchitecture-morphcore/

This gives a hint, for something I believe Juan have mentioned.

However, utilization of MorphCore could minimize the different core designs and could theoratically universalize a standard core config. Basically one core configuration to rule them all (or atleast most of them).

 

[Cazalan]

imo 32MB would be enough, cheaper too. intel claims to have more gfx performance with bdw's shaders so 72 eus with moar per-shader power would theoretically be quite huge for intel. i'll get back after i research a bit on bdw's igpu. so far i noticed that intel calls haswell's "gen 7.5" - hd4600 to iris pro. from hd5300 (bdw) "gen 8" starts.
edit: aw, and dat 72EU number seems oddly familiar to me. any relation to one of those KL MICs? may be i am making too far a leap...

Not that odd. Its a straight 50% more EUs.

The Broadwell GT3e is 48 EU, so 48+24=72. 😉

Really nothing like their HPC cores which are MUCH bigger for a similar number of "cores". Knights Ferry is like 650mm^2. Knights landing is estimated ~450mm^2 before whatever RAM they put on it.

 

[gamerk316]

http://arstechnica.com/gadgets/2015/02/intel-forges-ahead-to-10nm-will-move-away-from-silicon-at-7nm/
http://www.extremetech.com/extreme/199636-intel-at-isscc-14nm-in-the-bag-full-steam-ahead-on-10nm

 

[juanrga]

@gamerk, this slide is also interesting

@vmN, I will copy and paste my comment on wccftech site:

Sorry, but I don't trust anything of this.

First, Morphcore, CoreFusion, and similar reconfigurable hardware approaches not mentioned in the article such as Federation, VISC, SpMT,... are still in research phase and aren't completely developed for commercial implementation.

Second, the goal of Morphore and of the other reconfigurable hardware techniques is to combine the advantages of latency-optimized cores and of throughput-optimized cores. If Skylake was to provide both latency and throughput by morphing the core, then Intel wouldn't need to develop the KNL Phi core, which has been optimized for throughput. In fact, Intel will build heterogeneous clusters made of Skylake Xeon + KNL Phi, which suggest that Skylake will be a traditional big core optimized for latency, as Broadwell, Haswell, Ivy, Sandy,...

If there is anything 'revolutionary' on Skylake my expectations are: Larrrabe-based graphics, or new ISA extensions beyond AVX512, or maybe something about memory/cache prefetching.

 

[Cazalan]

I don't really buy the "MorphCore" either but whatever a peon phd student develops has little to do with what Intel could do if they put real money behind it.

 

[Turb0Yoda]

Traditional is Haswell I assume?

 

[de5_Roy]

http://www.pcworld.com/article/2887581/boxx-apexx-5-heres-what-a-36-core-five-gpu-workstation-looks-like.html
...so, can it play crysis 3?

 

[jdwii]

Just ordered my 4790K(and 970) decided to keep that until i see skylake numbers, broadwell seems very questionable to me in terms of desktop availability.

 

[de5_Roy]

skylake-dt industrial mobo spotted
http://www.anandtech.com/show/9006/embedded-world-2015-skylakedt-industrial-motherboard-spotted

 

[jdwii]

http://www.pcgamer.com/intel-skylake-processors-reportedly-delayed-until-late-august/

Lol as soon as i bought my 4790k to.

 

[gamerk316]

The Skylake delay isn't exactly breaking news. Intel is probably pushing back in least in part to help Broadwell.

 

[de5_Roy]

Intel to rebrand Atom chips along lines of Core processors
http://www.pcworld.com/article/2888996/intel-to-rebrand-atom-chips-along-lines-of-core-processors.html
promo slide here
http://www.techpowerup.com/img/15-02-26/75a.jpg

Intel Readies 4K-ready NUC NUC5i7RYH Desktop
http://www.techpowerup.com/210206/intel-readies-4k-ready-nuc-nuc5i7ryh-desktop.html
in case you're too tech-illiterate to decipher the uber-simple model numbering, it doesn't have iris pro hd6200, only hd6100 iris (non-pro) igpu, 3.1(b)-3.4GHz(T) mobile i7 cpu (which means castrated LLC of 4MB).

Some details of upcoming SoFIA 3G processors
http://www.cpu-world.com/news_2015/2015022501_Some_details_of_upcoming_SoFIA_3G_processors.html
slower is the keyword here...

 

[de5_Roy]

AT's directx 12 preview part 3: intel igpus
http://www.anandtech.com/show/8998/directx-12-star-swarm-intel-igpu-performance-preview

seems like intel's problem may be either in the code optimizations (drivers, games etc) or in the front end of the igpu. from bdw-related texts, i think gen 8 also has single front end hence faces similar bottlenecking issues. however, i've only come across compute stuff, nothing on gfx processing.

 

[jdwii]

De5 i think its more to do with the hardware then anything else. When will Intel start over on their igpu anyways? Lets see how Skylake goes. The 4600 is around Llano level quite behind Amd's current design

 

[de5_Roy]

De5 i think its more to do with the hardware then anything else. When will Intel start over on their igpu anyways? Lets see how Skylake goes. The 4600 is around Llano level quite behind Amd's current design

hardware is definitely part of the issue but intel claims that it redesigned gen 8 EUs for better resource utilization and compute performance. skylake will likely add more of the same.
i am parroting🙂)) a lot of stuff from here
https://software.intel.com/en-us/articles/intel-graphics-developers-guides

edit: intel at mwc: 14nm, sofia, rockchip, new branding etc.
http://www.anandtech.com/show/9026/intel-at-mwc-2015-sofia-rockchip-low-cost-integrated-lte-atom-renaming-and-14nm-cherry-trail

 

[DWW256]

What I want to know is if the upcoming Xeon E7 v3 series will use 22nm or 14nm. Intel could choose to finally move their Broadwell architecture onto something that can pack some serious power, or they could choose higher clock speeds on the Haswell platform. As far as I know, the highest BASE clock speed on any of the E7 v3 CPU's is 3.2 GHz. Does anyone know if Broadwell can handle that kind of frequency? Thanks!

 

[Cazalan]

What I want to know is if the upcoming Xeon E7 v3 series will use 22nm or 14nm. Intel could choose to finally move their Broadwell architecture onto something that can pack some serious power, or they could choose higher clock speeds on the Haswell platform. As far as I know, the highest BASE clock speed on any of the E7 v3 CPU's is 3.2 GHz. Does anyone know if Broadwell can handle that kind of frequency? Thanks!

The V3's are Haswell's so they're all 22nm.

http://www.cpu-world.com/news_2015/2015021001_Partial_specifications_of_Xeon_E7_v3_microprocessors.html