cancel
Showing results for 
Search instead for 
Did you mean: 

X99 beyond Broadwell-E?

Korth
Level 14
Writing this after seeing Menthol's Kaby Lake @ 5.0GHZ and above, Yes it does thread and googling up everything I could find about future Intel processor/chipset roadmaps.

Menthol's Kaby Lake i7-7700K appears to spec 4C, 4.2GHz, 8MB L3, dual-channel DDR4, 14nm, 91W ... overclocked to 5.0GHz on an "ASUS Z-170 board". He also comments that "supposedly the IMC has been improved and runs memory at high speeds".

Skylake, Kaby Lake, Coffee Lake, Cannon Lake, Skylake-X, Kaby Lake-X ... none of these codenames looks like they're gonna support LGA2011-3 HEDT.
And none of them looks (to me) like a "proper" upgrade to Broadwell-E - fewer cores and threads, fewer (yet slightly faster) iMCs, smaller caches, latest-and-greatest iGPUs wasting big space on die, more iPCIE lanes (on their limited-PCIE chipsets), more thermal/power efficiency, fancier extensions and features, etc. A whole mixed bag fulla tradeoffs and choices when compared vs the mighty i7-6950X.

Indeed, Broadwell-E/-EP itself doesn't really compare spectacularly vs Haswell-E/-EP. More of this but less of that. Or more of that, but not enough more to justify a $$$$ processor "upgrade" cost. Not Intel "tick", not Intel "tock", just Intel passing time with half-assed minor incremental tweaks and refinements. New mid-binned die on new process, not proven high-binned die on new process.

Is a newer and faster and bigger and mightier LGA2011-3 anticipated? Perhaps a sort of 10nm "Broadwell-X" or "Broadwell Refresh"? An LGA2011-3 CPU with specs equal or better in every measurable capacity, a true manly upgrade, not yet-another-non-linear retrograde/upgrade path filled with general and specific tradeoffs, diced up and sorted on some "intended application" marketing model?

Same cores. I could say "more cores, not fewer", but 8C/16T or 10C/20T (alongside multiple GPU cards) is already overkill crunch power.
More cache, not less. Intel generally does well on this, but sometimes they pretend that 8MB > 25MB.
Faster clocks, not slower. Broadwell-E vs Haswell-E has a poor showing. Why can't Intel just make existing Haswell-E parts run faster and cooler (and overclock better?) with a fab shrink and some Uncore tweaks?
Uncore iMCs. Faster iMCs to natively support "real world" DDR4 performance. DDR4-2400 is good, but DDR4-3000 (or faster) is better.
Uncore iPCIe controllers. How about 48 (or more?) PCIe lanes to fully support 3-way SLI and/or more fast storage? 40 (even 44) just ain't enough.
Upgraded FIVR. Yes you heard it here first, the FIVR is somewhat circumvented by some motherboards (with "overclock-grade" variant LGA2011-3 sockets and VRMs) and much maligned for package space and package heat ... but for those of us unwilling to drown our CPUs in LN2 it's a superior tool for processor control and overclocking. Why make fully unlocked fully overclockable "-X" parts without FIVRs?
Completely ditch the iGPU. Don't even think about it. Even a lump of thermal interface goop would be better than wasting space on unpowered silicon. X99 ain't for laptops and HTPCs.

I bought X99 with the expectation that over the next few years the platform would mature and offer useful upgrade options. I didn't buy (much) Z170 because it's less suitable for my particular needs. I don't intend to rush out and buy "X299" just to start another costly hardware beta tester early-adopter cycle - especially not if I see "lesser" specs on the processors designed to run on it.

The big selling points of X99 vs Z170 (currently the best Skylake chipset) are massive processor power, massive memory bandwidth, and massive (up to 40) PCIe lanes. Great for "workstation" use because of the raw information crunching power, great for gaming because of the possible multi-GPU configurations. The X99 crowd don't want Intel's fanciest new HDMI iGPU garbage, they do want to bridge 2 to 4 high-end GPU cards for maximum GPGPU and fps. The X99 crowd don't want boring JEDEC DDR4-2400, they do want to run extreme DDR4 at full spec alongside a processor overclock. The X99 crowd is willing to pay premium price for premium performance, their systems are built around every context of "high power". So while increased power efficiency is always good, the mindset is that a 98W processor is NOT better than a 138W processor when runs at a lesser spec, especially since coolers can be fitted onto either part.

So my little letter to Intel:
I don't want to buy X299 or Skylake or Kaby Lake or any other lake. I do want to buy a better CPU for my X99. A CPU which supports at least the same cores/threads, at least the same cache, faster clock, faster quad-channel DDR4, and every PCIe lane the X99 chipset can support (without PLX). I'm not talking about selling future-binned 'Lake CPUs in LGA2011-3 packages just because they fall outside a mainstream spec or have bad iGPU circuitry or not enough working cores to Xeon or whatever. I am talking about actual LGA2011-3 CPUs made from actual LGA2011-3 CPU dies, without any "bad", "dark", "deactivated" or other "failed" silicon wasting space and warping power/thermal output efficiencies, I want 8C/10C parts made from 8C/10C dies. Quit trying to sell lower-power, higher-efficiency, leaner, lighter, smaller CPUs - quit trying to sell recycled, rebadged, and repackaged castoffs from other bins - such parts have their place, but that place is not in high-powered X99 platforms. I know I'm just one (not very loud) voice you can ignore, but hey I don't have to keep buying your stuff if I feel it's always "dead-end" tech doomed to obsolescence every time a new model rolls out your door.
"All opinions are not equal. Some are a very great deal more robust, sophisticated and well supported in logic and argument than others." - Douglas Adams

[/Korth]
4,576 Views
4 REPLIES 4

Arne_Saknussemm
Level 40
Pretty standard stuff by now...

X79 2011 has SB-E and IB-E

X99 2011v3 has had it's 3 years Haswell-E and Broadwell-E

X299 will be the next cycle...new socket, LGA 2066, and will support Skylake X and later Cannonlake X with Kaby lake X fillers in there somewhere..

Then 2020 and LGA 2076 and Ice Lake and off we go again...

Skylake X will have 10 and 8 core variants replacing Broadwell E

it is what it is...

Zen is coming too;)

Korth
Level 14
Yeah, just ranting. Understood whatcha saying, it is what it is.

X99 had Haswell-E, has Broadwell-E - might even get one last refresh - but next-socket X299 is already underway.

Devil's Canyon i7-4790K (Q2/2014): 4C/8T, 4.0GHz/4.4GHz, 8MB cache, Dual-iMC (32GB DDR3-1600), PCIe 3.0 (16 lanes), DMI 2.0 (PCH), iGPU (350MHz HD 4600), 22nm, 88W
Skylake i7-6700K (Q3/2015): 4C/8T, 4.0GHz/4.2GHz, 8MB cache, Dual-iMC (64GB DDR4-2133), PCIe 3.0 (16 lanes), DMI 3.0 (PCH), iGPU (350MHz HD 530), 14nm, 91W
Kaby Lake i7-7700K (Q1/2017): 4C/8T, 4.2GHz/4.5GHz, 8MB cache, Dual-iMC (64GB? DDR4-2400+), PCIe 3.0 (16 lanes), DMI 3.0, iGPU (?), 14nm, 91W

True, each of these "top" consumer processors includes more instruction extensions (with increasingly better AVX implementions), more capable iGPUs (with more powerful video streaming/playback codecs, 3D processing, resolutions, etc), and a variety of other new Intel technologies (such as native Optane support). And each is paired with an increasingly powerful Intel chipset (such as Z97, Z170, Z270) with newer/faster/better power, storage, USB, and PCIe interfaces - stuff which still requires a little bit of dedicated silicon on the processor die. And the jump from Devil's Canyon to Skylake is bigger than the jump from Skylake to Kaby Lake, we apparently won't see any more "big" differences until 10nm Cannonlake is released (Q3/2017?, Z370?).

What I'm saying is that the differences (so far) aren't that big at all. All of these parts could reasonably be expected to run OC at 4.8GHz or 5.0GHz. Lots of "exciting new Intel technologies" and small incremental improvements (along with increasingly impressive specific benchmark metrics), but not so much actual change in terms of real-world computing or usage or gaming. So what I'm saying is "who cares about Cannonlake-X, another 4C/8T 4GHz 8MB part like all the others?", why not stop tinkering with new stuff and start reducing lithography on the proven old stuff? I imagine the performance potential of a 14nm Devil's Canyon or a 10nm Haswell-E. I suspect that shrinking from 22nm to 10nm (a 55% reduction?) must free up a lot of active surface space on the die, space which might be used for upgraded iMCs or PCIe controllers, space which could at the very least be stuffed with a void-filling TIM. (Plus, Intel usually targets package TDPs, so if a fab shrink reduces their 91W to 78W they'll find some way to pack 12W more features back into the part.)

What are (were) X79 "SB-E and IB-E"? I assume you weren't talking about the Thermalright Silver Arrow Extreme cooler, SB-E and IB-E variants?

Not really interested for Zen. AMD has made too many promises, too quickly, too often, and already waited far too long ... yet AMD has a poor record with results. I expect AMD Zen will appear, eventually; I expect it'll be better than AMD's current CPU/APU parts; and I expect it'll generally price much lower than Intel offerings. I don't expect AMD to suddenly veer into a decisive lead over Intel after lagging so far behind the race for so many years. The days of AMD-Intel (and AMD-NVIDIA) leapfrogging are gone forever - I think the next Intel competitor will be a big brand-fab company like Samsung/TSMC, Panasonic/TowerJazz, Acer/SMIC, SK Hynix, or TI. Intel might patent-lock their sockets and architectures - at least until they officially withdraw support and declare EOL - but imagine a world where you could upgrade your (old) computers with new processors made by new companies, or buy entire desktop motherboards (capable of full x64 emulation) which have nothing to do with Intel or AMD. Intel needs competition, it's a global company with global dominance and monopoly, it builds our civilization's technology and it's built our civilization's marketing - consumers now pay up to $1000 or $1500+ for a processor and up to $500+ for a motherboard, while enterprise pays up to $5000+ per processor and up to $2000 per mainboard?
"All opinions are not equal. Some are a very great deal more robust, sophisticated and well supported in logic and argument than others." - Douglas Adams

[/Korth]

Korth wrote:
Yeah, just ranting.


Sure..I get that...when I say "it is what it is" I do so with a sigh and in resigned tones.

I am a PC enthusiast and have changed my hardware as frequently as my underwear sometimes...err...

Anyway...lately I find myself unenthused to do so....little gain. Probably more gain changing GPUs at the moment

I got lucky with my Haswell E it does 5.0GHz so replacing it with Broadwell just wasn't going to happen...especially at that price...like you say...small increments for whole jump in cash price not attractive for upgrades....

SB-E and IB-E...Sandy Bridge E and Ivy bridge E....was another disappointing upgrade path...

I'm a bit the same with AMD...would like to like but at this point seems they aren't going to get anything together to provide a challenge to Intel...shame

Before long they will have us renting cloud computing power by the month and the hardware we have at home will be a display...

Korth
Level 14
Not at all interested in renting what I could be owning, lol. And data on the cloud is never truly secure. Always gonna be a powerful machine on my desk, lol.
"All opinions are not equal. Some are a very great deal more robust, sophisticated and well supported in logic and argument than others." - Douglas Adams

[/Korth]