News on Ivy Bridge-E?

[AdamK47]

Originally Posted by kleinkinstein

Nice work! We've caught a troll or using his words...a damn fool! So sad when a sig line is make believe!

Yep, it's make believe.

[Idontcare]

Originally Posted by kleinkinstein

Nice work! We've caught a troll or using his words...a damn fool! So sad when a sig line is make believe!

His sig is not make believe.

[Intel17]

The reason that Gulftown came first was that it wasn't as fundamentally radically different in the uncore as compared to Lynnfield relative to Nehalem.

Right now SNB-E's uncore is a totally different animal from SNB/IVB's. I'm sure HSW-E will have a lot of goodies in the uncore that will explain the extra time.

[Ajay]

Originally Posted by Intel17

The reason that Gulftown came first was that it wasn't as fundamentally radically different in the uncore as compared to Lynnfield relative to Nehalem.

Right now SNB-E's uncore is a totally different animal from SNB/IVB's. I'm sure HSW-E will have a lot of goodies in the uncore that will explain the extra time.

Yeah, Intel is screwing with us. For what, just to make profits! Seriously though, x79 would make a bit more sense if Intel fixed and updated the chipset. Instead the boards get expensive once mainboard companies add in all the extras.

[boxleitnerb]

I believe Haswell will challenge Ivy-E in the same power envelope (with or without OC, your pick). Singlethreading - no contest, Haswell will win hands down.

Multithreading...let's see, I'll make some guesses:
10% more IPC, 10% improved HT, 15% higher clocks (again, at the same wattage). Combined this gives a plus of nearly 40%. Close to the benefit from having 50% moar coars.

However, if Haswell is a stinker too at OC, then all bets are off. Personally I hope for an easy 5 GHz.

[meloz]

Originally Posted by Idontcare

The same thing happened with successive Itanium iterations, and I doubt there were any nefarious reasons why that happened.

The delay with successive Itanium iterations were because it was Itanium. It became very apparent early in the Itanium program that it would never grow up to to be the x86 replacement, and Intel did the bare minimum with Itanium from then on. You cannot seriously compare this with x86-64.

IVB-E will share much with small socket IVB, just like SB-E did with its smaller sibling. It is not as if they are developing an all-new arch. Somehow Intel were able to get small socket IVB ready in time, but putting a few more cores, more cache, removing the iGPU and other such *relatively* minor modifications on an already proven design will take nearly two years.

If you buy into this BS I have a bridge to sell you.

[kleinkinstein]

Originally Posted by tweakboy

Skymont wtf, that is 2015 right ? Haswell is 2014. 2013 is the year of the E chip. gl

Hey now, Mom says patience is a virtue.

[NTMBK]

...what on earth happened to this thread?

[Idontcare]

Originally Posted by meloz

The delay with successive Itanium iterations were because it was Itanium. It became very apparent early in the Itanium program that it would never grow up to to be the x86 replacement, and Intel did the bare minimum with Itanium from then on. You cannot seriously compare this with x86-64.

IVB-E will share much with small socket IVB, just like SB-E did with its smaller sibling. It is not as if they are developing an all-new arch. Somehow Intel were able to get small socket IVB ready in time, but putting a few more cores, more cache, removing the iGPU and other such *relatively* minor modifications on an already proven design will take nearly two years.

If you buy into this BS I have a bridge to sell you.

"Proven for consumer use" and "proven for server use" are two different things entirely, even with the exact same chip. Validation efforts are not the same, this isn't "good enough for government work".

AMD didn't quite get that part and they created the TLB fiasco for themselves in the server space.

Intel did get that part, and you didn't see a platform-level recall with SBE like what happened with SB because they didn't rush SBE.

This is the difference between amateur hour and the professionals, AMD vs Intel.

IVB-E will come out when Intel is confident they have sufficiently validated the platform and the chip. More importantly, it will come out once key server system integrators are equally confident.

I get that this doesn't quite play into the picture you have in mind but it is the reality.

[red454]

Refresh my memory as to what the Ivy-E is expected to offer (cores, etc)...

[ShadowVVL]

Originally Posted by red454

Refresh my memory as to what the Ivy-E is expected to offer (cores, etc)...

my guess,2 cores ,higher temps,lower power consumption and a vary distant release date.

[Ferzerp]

Originally Posted by ShadowVVL

my guess,2 cores ,higher temps,lower power consumption and a vary distant release date.

2 more memory channels.

It wouldn't surprise me to see a return to solder though, so lower temps.

[ViRGE]

Originally Posted by red454

Refresh my memory as to what the Ivy-E is expected to offer (cores, etc)...

No one knows for sure. More cores is a possibility, but it's more likely it's just going to be a straight shrink along with some errata fixes (such as the PCIe controller), similar to the CPU changes for SNB->IVB.

[AdamK47]

What PCI-E problem would that be?

[ViRGE]

Originally Posted by AdamK47

What PCI-E problem would that be?

PCIe 3.0 basically doesn't work correctly. There are too many bugs in SNB-E's implementation. Which is why IVB is PCIe 3.0 certified but SNB-E is only PCIe 2.0 certified.

Intel thought they could implement BIOS/microcode work-arounds so that it could be made to unofficially work, but they've even backed off on that. You won't find any modern literature claiming any kind of PCIe 3.0 support from Intel, though motherboard manufacturers are free to unofficially enable it on if they want to (which is what lead to NVIDIA disabling it in their drivers for Kepler parts).

[AdamK47]

Guess I'm lucky with my 7970s. PCI-E 3.0 is working correctly.

[ViRGE]

Originally Posted by AdamK47

Guess I'm lucky with my 7970s. PCI-E 3.0 is working correctly.

Intel has never come clean on what exactly is wrong, but it's apparently some kind of manufacturing defect. Whether it works or not is almost entirely variable on a chip-by-chip basis; some people can make it work, some can't.

[Ferzerp]

Originally Posted by ViRGE

No one knows for sure. More cores is a possibility, but it's more likely it's just going to be a straight shrink along with some errata fixes (such as the PCIe controller), similar to the CPU changes for SNB->IVB.

I'm guessing you mean "more cores" in the context of SB-E and not IB? It would be odd to stick with 4 cores.

Though with the talk of the die shrink, I'm pretty sure you mean in the context of SB-E. Ditto on the PCIe comment too.

If this is clear, forgive me. I'm ill and medicated, so my comprehension may be off.

[red454]

Originally Posted by ViRGE

No one knows for sure. More cores is a possibility, but it's more likely it's just going to be a straight shrink along with some errata fixes (such as the PCIe controller), similar to the CPU changes for SNB->IVB.

I see - Seems like I read somewhere recently about 8 or perhaps 10 cores. Speculation, most likely.

It will be interesting to see what that will do to the current CPU lineup (cost wise) for the X79 platform in a year or so. So if someone were going with an X79 now with a 3820, and held out if they wanted either a better deal on say the 3930K or a mega-core Ivy-E monster.

Didn't the Sandy Bridges drop a little after the Ivy was released?

[ViRGE]

Originally Posted by red454

I see - Seems like I read somewhere recently about 8 or perhaps 10 cores. Speculation, most likely.

It will be interesting to see what that will do to the current CPU lineup (cost wise) for the X79 platform in a year or so. So if someone were going with an X79 now with a 3820, and held out if they wanted either a better deal on say the 3930K or a mega-core Ivy-E monster.

Didn't the Sandy Bridges drop a little after the Ivy was released?

SNB-E is already 8 cores. Intel does have a 10 core Gulftown processor, but it's not clear if they intend to replace it. As for SNB-E prices, this is Intel. Intel won't cut prices they'll just discontinue SNB-E processors.

Originally Posted by Ferzerp

I'm guessing you mean "more cores" in the context of SB-E and not IB? It would be odd to stick with 4 cores.

Though with the talk of the die shrink, I'm pretty sure you mean in the context of SB-E. Ditto on the PCIe comment too.

If this is clear, forgive me. I'm ill and medicated, so my comprehension may be off.

I'm talking about IVB-E as compared to SNB-E.

[tynopik]

Originally Posted by ViRGE

PCIe 3.0 basically doesn't work correctly. There are too many bugs in SNB-E's implementation. Which is why IVB is PCIe 3.0 certified but SNB-E is only PCIe 2.0 certified.

Intel thought they could implement BIOS/microcode work-arounds so that it could be made to unofficially work, but they've even backed off on that. You won't find any modern literature claiming any kind of PCIe 3.0 support from Intel, though motherboard manufacturers are free to unofficially enable it on if they want to (which is what lead to NVIDIA disabling it in their drivers for Kepler parts).

PCIe 3.0 is supported on the sandy bridge xeons (E5 series)

[gramboh]

Originally Posted by skipsneeky2

LOL@Adam.

I had people laughing at me as well when i dumped my e6750 back in 2007 for a q6600 and all i heard was that well quads are useless you wasted $300 LOL but the main game i played being UT3 at the time made tremendous use of the extra cores.

i remember the e8400 being the big rival and about a million e8400 vs q6600 threads popped across every forum on the web,well it took about 3 years but nearly every game makes use of quads and the q6600 with a oc is still a rather good processor.

Going by the experience of my old q6600 and how it aged well,i could see myself building a "foreverbox" with a 3930k build just as a pc i won't upgrade or touch it for the next 4-5 years.

Only problem with that is you have a short upgrade path with IB-E. Except in corner cases, for desktop use you would most likely be better served (and save a lot of money/power) with Haswell which is not that far off now. Q6600 was legendary price/performance I agree, I had a Q6700 up until IB launch day.

[NTMBK]

Originally Posted by ViRGE

SNB-E is already 8 cores. Intel does have a 10 core Gulftown processor, but it's not clear if they intend to replace it.

Going to 10 cores would be a fairly smart move- don't forget, on IB Intel sank a load of their die-shrink bonus transistor budget into enlarged processor graphics:



They still managed to come back with a die that was only 75% the size of SB, despite having 20% more transistors. My back of an envelope calculations indicate they could happily get a couple more cores onto an IB-E (over SB-E) and still get a similar shrink.

EDIT: Plus, going to 10 core gives a good reason for pro customers to go for an upgrade. Going from a 16-core to 20-core workstation, with slightly higher clocks and IPC, seems like a nice investment for CPU-heavy workloads.

[Dadofamunky]

Originally Posted by Ajay

Yeah, Intel is screwing with us. For what, just to make profits! Seriously though, x79 would make a bit more sense if Intel fixed and updated the chipset. Instead the boards get expensive once mainboard companies add in all the extras.

I don't think that's exactly what's going on, and I don't think you do either; what's going on is a fundamental shift in the overall computing market. Suddenly, a hundred million (or more) affluent customers are finding that they can watch movies, browse the Web and edit documents on an iOS or Android tablet, even more conveniently than any laptop. Those hundred million + CPUs in those tablets are not being sold by Intel. This phenomenon has stung Intel's bottom line, and has gone a long ways to bankrupt AMD. Will Haswell make its way into the nex-gen iPad 6? IPhone 8? Samsung Galaxy S6? I doubt it. Intel's performance is second-to-none, but they are the dominant player in a market that is suddenly shrinking. People want battery life, not 200 FPS of WoW. Things just got a lot more complicated for the big I.

Originally Posted by Idontcare

His sig is not make believe.

I think Adam can run Defense Department-quality flight simulators on his rig.

[Lepton87]

Originally Posted by boxleitnerb

I believe Haswell will challenge Ivy-E in the same power envelope (with or without OC, your pick). Singlethreading - no contest, Haswell will win hands down.

Multithreading...let's see, I'll make some guesses:
10% more IPC, 10% improved HT, 15% higher clocks (again, at the same wattage). Combined this gives a plus of nearly 40%. Close to the benefit from having 50% moar coars.

However, if Haswell is a stinker too at OC, then all bets are off. Personally I hope for an easy 5 GHz.

What makes you think that haswell will OC better? It's a wider core, why would a wider core OC better unless they deepen the pipeline? Besides, I think IVY-E would be 8 core even for a consumer version, not 6 cores like cut-down 1000$ consumer CPU. It's atrocious that they cut-down 1000$ EE CPUs. Don't forget that SB-E is already 8 core CPU.
UPDATE: I forgot to mention on-die VRMs circuitry, they can make those circuitry handle only 20-40% more power than a stock haswell needs effectively hampering OC to a massive degree. It's not to make our life harder, we're too small a market for them to care. The reason for that would be to cut cost. See poor Ivy-bridge TIM. Does it work for 99.5% of their market? Yes it does. Does it save them money? Yes, it does. So why would they care about us overclockers?