Broadwell/Skywell Release - New Intel Details

drescherjm said:
I assume NDA and you can not say anything about it.
Click to expand...

Correct.

Unless 14nm sucks at overclocking, I think the 6960X is going to be a big enough e-peen upgrade over a 5960X to justify it.

Edit: * subject to change since this crap is still a long way off so who knows what sort of gimping or other market decisions happen.
 
So anyone want to chime in:

Skylake vs Haswell-E/Broadwell-E

Spare the additional cores and extra PCI-E lanes is their really enough of driving force for one to consider Haswell-E/Broadwell-E over Skylake?
 
samm said:
So anyone want to chime in:

Skylake vs Haswell-E/Broadwell-E

Spare the additional cores and extra PCI-E lanes is their really enough of driving force for one to consider Haswell-E/Broadwell-E over Skylake?
Click to expand...

More cores and pci lanes is pretty much the main attraction for -E. There's also SATA ports if for some reason you need 10 instead of 6. Even Skylake won't support 10.
 
lucidrenegade said:
Blame no competition for many, many years.
Click to expand...

Dear God, this. I hope the rumors of an AMD/Samsung buyout are true. AMD gets money to pull it's head out of it's ass in the CPU segment and Samsung gets rights to AMD GPU tech for integration into ARM devices.

Overall it'd be a nice shot in the arm for competition.
 
melk said:
Highly unlikely. I'd be surprised if they break 5 ghz.
Click to expand...

I agree, Intel's 3D transistors give them more processors per wafer but they remain a bear to cool. Same with Ivy Bridge and Haswell, die shrink or not, they just don't shed heat fast enough.
 
JDanser said:
Dear God, this. I hope the rumors of an AMD/Samsung buyout are true. AMD gets money to pull it's head out of it's ass in the CPU segment and Samsung gets rights to AMD GPU tech for integration into ARM devices.

Overall it'd be a nice shot in the arm for competition.
Click to expand...

I don't think AMD is interested in competing in the high performance desktop/server segment anymore, neither is Samsung. If anything, Samsung would use AMD for their mobile segment.
 
Just a reminder, not sure but some of you were looking for over lockable multi core ? We've been OCing and 48-64 core systems for some time in the folding section, granted not the same as Intel but if you really want the cores there are way to OC lol
 
dr/owned said:
Correct.

Unless 14nm sucks at overclocking, I think the 6960X is going to be a big enough e-peen upgrade over a 5960X to justify it.

Edit: * subject to change since this crap is still a long way off so who knows what sort of gimping or other market decisions happen.
Click to expand...

are you allowed to give an idea what an upgrade from a SB-E 3930 to a Broadwell-E may bring? 35-40%? I assume quite a bit because of the 5960 remark :)

3930 just keeps going on an on and on.. nothing worth upgrading to (for a reasonable price)
 
lutjens said:
What I'm saying is that people who pay for a high core count Xeon chip should be able to use the huge amount of headroom that's present in these chips. As I said above, the E5-2699 V3 when run at 105MHz bus speed (2940MHz) run at 0.9V and run quite cool. I know such a chip wouldn't match a chip with a lower core count in pure clock speed, but a modest clock speed between 3.5GHz and 4.0GHz with a slight voltage increase and should be easily obtainable under water without risking damage or instability.
Click to expand...

I don't really get where you say there is large overclock headroom. No one really knows and it goes against well known and established chip fabrication experiences. Quite frankly any time you add more cores, the likelihood of one core being defective at a high clock rate exponentially increases per additional core.
 
Here's hoping Skylake-K comes out before the end of the year (and is significantly faster than Haswell or at least overclocks better). To be brutally honest I don't think any of us care about Broadwell. The most strenuous things I do with my computer tend to be gaming so the high-end consumer chips are what I normally buy (-E series being slower on a per core basis and all).
 
broadwell is fine.

I think most folks will just wait for Skylake though. I think Intel kinda shot themselves in the foot by basically releasing them at the same time.

makes no sense to me to have broadwell sku's and skylake sku's selling at the same time.
 
makes no sense to me to have broadwell sku's and skylake sku's selling at the same time.
Click to expand...

To me the only reason to broadwell-k will be in the same socket as haswell and same ddr3 while skylake-k will be in a new socket and require ddr4 other than that people can wait a month or so..
 
Quix said:
Here's hoping Skylake-K comes out before the end of the year (and is significantly faster than Haswell or at least overclocks better). To be brutally honest I don't think any of us care about Broadwell. The most strenuous things I do with my computer tend to be gaming so the high-end consumer chips are what I normally buy (-E series being slower on a per core basis and all).
Click to expand...

Don't give two shits about broadwell. Skylake is the only hope enthusiasts have left. It's entirely possible that Sandy Bridge will go down in history as the last great architectural improvement on silicon.
 
Whats SKUs btw would be launching with Skylake?

Intel releases their lower part/laptop parts first and then delays desktop launches upto 6 months. So anyone confirm the skylake launch would indeed be the mainstream desktops?
 
The picture shows 95W K, 65W and 35W parts released at the beginning of Q3. I expect anyone that knows if this is accurate is under a NDA and can not legally confirm or deny this.
 
Last edited:
It shows the Broadwell LGA as above the Skylake chips all the way across the board through 2016. I'm beginning to wonder if Skylake will actually be an upgrade, or will it just be an adjustment to get lower power numbers. With Haswell being such a disappointing performance uptick, if it could even be called that, I wonder if Broadwell will be anything special. Are we stuck at this performance level until a new competitor comes along in ten to twenty years?
 
wabbitseason said:
Skylake is the only hope enthusiasts have left. It's entirely possible that Sandy Bridge will go down in history as the last great architectural improvement on silicon.
Click to expand...

This. At least for gaming enthusiasts looking for per core performance improvements.
Broadwell does make sense if Skylake doesn't come with huge performance gains to justify spending in a new MB and DDR4.

If neither gives an interesting upgrade for my 2500K then I'm investing the money in a second GTX970 or perhaps a GTX980Ti ;)

Because sincerely: USB 3.1? SATA M.2 instead of PCIe? SATA Express porft for nonexistent drives? I see very little to justify a replacement for a desktop gaming PC.
 
http://imgur.com/gallery/NYiTb31

This image is a chipset comparison showing details of the Z170. It looks like a pretty huge step up in terms of I/O and PCIe usage. If accurate, all the more reason to wait for Skylake.
 
12 cores over-clocked over 5ghz would motivate me to upgrade.... till then throw the money at the GPU,..
 
Meh, I just retired my 2500k to a 4790k the other week, so i'm fine until Skylake K in 2016.
 
drescherjm said:
The picture shows 95W K, 65W and 35W parts released at the beginning of Q2. I expect anyone that knows if this is accurate is under a NDA and can not legally confirm or deny this.
Click to expand...

You're reading it wrong. The slides indicate production starts at the beginning of Q3. Probably means actual sales and whatnot are a month or two later.
 
GunzofNavarone said:
12 cores over-clocked over 5ghz would motivate me to upgrade.... till then throw the money at the GPU,..
Click to expand...

LOL, you can't even reach 5 ghz on Devils Canyon without high end water cooling, so yeah, your gonna be waiting a loooooong time.
 
lutjens said:
The largest die will be for Broadwell-EX and may be up to 24 cores (somewhere between 18 and 24). A nice chip with great potential, but again, they will be completely hard locked. Given that the current E5-2699 V3 (when overclocked by the measly 4 MHz that you can do with the locked BCLK straps) runs with a Vcore of 0.900V at nearly 3.0GHz, the incredible clock speed potential of these large die CPUs is obvious.
Click to expand...

Part of the reason that the high core count chips have such a low vcore is that they have to. There just isn't any other way to manage the heat from those dies.

You'll never be able to clock the high core-count chips as high as the 4, 6, or even 8 core parts that we can overclock. A 24-core chip would have, obviously, 3 times the heat dissipation of an 8-core chip at the same vcore/clock speed, and a whopping 6 times as much as an equivalent 4-core chip. Even with the best water cooling setups, you're still going to run up against issues of conducting that heat out of the dies.
 
AgentQ said:
Part of the reason that the high core count chips have such a low vcore is that they have to. There just isn't any other way to manage the heat from those dies.

You'll never be able to clock the high core-count chips as high as the 4, 6, or even 8 core parts that we can overclock. A 24-core chip would have, obviously, 3 times the heat dissipation of an 8-core chip at the same vcore/clock speed, and a whopping 6 times as much as an equivalent 4-core chip. Even with the best water cooling setups, you're still going to run up against issues of conducting that heat out of the dies.
Click to expand...

There's also more localized heat transfer problems. With huge die, some cores are going to be surrounded on all sides by other cores. What we see on 4 core die with the middle cores being hotter than the outer cores...same thing scaled up.


Random tangent, but if we all banded together and ordered a few million bucks worth of chips, Intel could/would make a custom brew of -EX. 300W TDP, 18 cores, 4 Ghz. Start saving pennies now :)
 
DigitalGriffin said:
I don't really get where you say there is large overclock headroom. No one really knows and it goes against well known and established chip fabrication experiences. Quite frankly any time you add more cores, the likelihood of one core being defective at a high clock rate exponentially increases per additional core.
Click to expand...

The Vcore range for the HCC chips has been defined as 0.65V to 1.3V. If one were to give an 18-core 1.2-1.25V, you open the door to much greater frequency potential, as long as you can manage the heat generation. While having more cores doesn't help with attaining the highest clock rate, it's more a function of transistor ability across the entire die. An entire die is analyzed when manufactured for quality with lower yielding portions often being deactivated as required (depending on how much of the die rates highly). A 14 core may have all cores functional, but 4 cores were found to be lower yielding (but not defective) so there are deactivated to keep the die's yield up, rather than being forced to remark the entire chip as a slower, lower grade (and lower margin) 18-core (although either are an option depending on demand). Chips with lower yielding portions are often harvested into other chips, which is one of the reasons why there are a huge number of lesser SKUs that are offered. The transistors of top bin examples of a chip (like the E5-2699 V3) typically have a high quality throughout the die and can support higher speeds at higher TDPs (that aren't currently offered) if they were given a higher Vcore.


AgentQ said:
Part of the reason that the high core count chips have such a low vcore is that they have to. There just isn't any other way to manage the heat from those dies.

You'll never be able to clock the high core-count chips as high as the 4, 6, or even 8 core parts that we can overclock. A 24-core chip would have, obviously, 3 times the heat dissipation of an 8-core chip at the same vcore/clock speed, and a whopping 6 times as much as an equivalent 4-core chip. Even with the best water cooling setups, you're still going to run up against issues of conducting that heat out of the dies.
Click to expand...

MCC and HCC Xeons are typically much tighter chips than the consumer graded chips (have less internal leakage) due to being manufactured with much greater care, as the penalty for leaky/defective transistors is much higher for chips with a large transistor count. They run cooler, but when they hit their wall, they hit it more quickly. The superior die quality is why I purchased an E5-1680 V2 (based off the MCC 10-core die) rather than a V3 8-core, even at twice the price (along with more L3 cache and no FIVR). Even the 8-core V3 Xeons, while being the best of Haswell-EPs LCC dies, are not of the same pedigree as the MCC and HCC chips.

I agree that an 18-core won't clock as high as a 4, 6 or 8 core, but still has considerable headroom that could be used with proper cooling, headroom that is present due to Intel's low TDP requirements. The low Vcore that is used currently is kept low (along with the low clock speeds) to keep the CPU within it's specified TDP envelope.

dr/owned said:
There's also more localized heat transfer problems. With huge die, some cores are going to be surrounded on all sides by other cores. What we see on 4 core die with the middle cores being hotter than the outer cores...same thing scaled up.
Click to expand...

As I've said before, the higher core count chips are made to a higher standard than the low core count chips and have much less leaky transistors. To compare the transistor consistency and quality of a run-of-the-mill i7 to that of an HCC Xeon is like comparing apples to oranges.

Random tangent, but if we all banded together and ordered a few million bucks worth of chips, Intel could/would make a custom brew of -EX. 300W TDP, 18 cores, 4 Ghz. Start saving pennies now :)
Click to expand...

If we had an unlocked variant, we'd be able to decide the TDP of the chip ourselves by what clock speed we set and the voltage we use...no custom part required. The only benefit to the chip you describe is the pre-validated nature of it...no overclocking or tweaking required to get the stated specs. If such a chip were to be offered, I'd take a tray of 10 of them myself...;) Such a chip is unlikely to be released for mass consumption as a regular SKU (although there would be a huge demand for it) due to the reality than the reference platform for Haswell-EP almost certainly doesn't consider this type of TDP to be possible. Most server and workstation boards for Haswell-EP only have 4 to 5 phase VRMs which would limit the maximum TDP of the CPU that they could support. HPs Gen9 servers and their Socket R3 workstations all have provision for five phase power, but have one phase removed (the silkscreening on their motherboards for the missing phase is still visible), so a Rev 2 would have to be executed to restore the missing phase. Currently only enthusiast based X99 boards (and perhaps the Asus Z10PE-D8) could support the current draw of a 300W chip, and Intel is highly unlikely to even consider a higher TDP chip that the reference Haswell-EP platform cannot support.
 
Last edited:
drescherjm said:
To me the only reason to broadwell-k will be in the same socket as haswell and same ddr3 while skylake-k will be in a new socket and require ddr4 other than that people can wait a month or so..
Click to expand...

Sad but true.. I've been milking this G3258 processor a long time though, and I held off by the skin of my teeth waiting for Broadwell. From my standpoint I can either dump some $ for a CPU, or wait until skylake and shit the bed paying for Mobo+Ram+CPU. If the chip turns out to be nothing special, I'm probably going to pick up the cheapest 4790k I can find to save a little coin.
 
dr/owned said:
More cores and pci lanes is pretty much the main attraction for -E. There's also SATA ports if for some reason you need 10 instead of 6. Even Skylake won't support 10.
Click to expand...

probably not.

i refused to buy haswell-refresh over haswell-e because i saw s1150 as funadmentally crippled by having only 16x 3.0 lanes from the CPU, and half a dozen 2.0 lanes from the MCH. it is quite frankly inadequate for the m.2 generation.

skylake answers this, and with 14nm with 95W thermal headroom should produce some extremely fast CPU's regardless of whether they remain 4c8t designs.

i just didn't have a year to wait...
 
The really crappy thing is that even with the extra 6 Gbps ports on X99, you have to deal with the fact that the DMI 2.0 bus will become a bottleneck in VERY short order and largely nullify the advantage of more 6 Gbps ports. One more 6Gbps SSD can be added to a two drive RAID 0 configuration that existed under X79, but that's about it...three SSDs will saturate the 2GB/s DMI 2.0 bus.

NVMe drives are an option to overcome this bottleneck...;)
 
is it NVME that will help here, or simply more PCIe on the CPU and faster PCIe on the DMI bus?
 
Just chiming in here to say thanks lutjens, your posts are intelligent and useful. Specifics, facts, and rational reasoning?? We're glad to have you here.
 
lutjens said:
The really crappy thing is that even with the extra 6 Gbps ports on X99, you have to deal with the fact that the DMI 2.0 bus will become a bottleneck in VERY short order and largely nullify the advantage of more 6 Gbps ports. One more 6Gbps SSD can be added to a two drive RAID 0 configuration that existed under X79, but that's about it...three SSDs will saturate the 2GB/s DMI 2.0 bus.

NVMe drives are an option to overcome this bottleneck...;)
Click to expand...

Guess what my next upgrade is going to be in March? ;)
 
wabbitseason said:
Just chiming in here to say thanks lutjens, your posts are intelligent and useful. Specifics, facts, and rational reasoning?? We're glad to have you here.
Click to expand...

Thank you for your kind words. I just try to offer my $0.02 and it's good to hear that it resonates with some folks.:)

dr/owned said:
Guess what my next upgrade is going to be in March? ;)
Click to expand...

You and me both...;)

I'm just waiting for a little more finality and polish on Windows 10 before making the leap (as I want bootability and detest Windows 8.1). I may just end up waiting for RTM, although the lasted Technical Preview looks pretty good.
 
lutjens said:
I'm just waiting for a little more finality and polish on Windows 10 before making the leap (as I want bootability and detest Windows 8.1). I may just end up waiting for RTM, although the lasted Technical Preview looks pretty good.
Click to expand...

You and me both. I've got $1200 in a sock drawer that took three years to get together and has survived a new mortgage and a new baby. In a perfect world, Win 10, Skylake K and Z170 all hit this summer, and I can get a worthy CPU + RAM + mobo + OS + M.2 PCIe x4 upgrade.

USB 3.1 with a couple type-C connectors wouldn't hurt either. :cool:
 
You must log in or register to reply here.
Back
Top