Dear mr. Lee,

have you ever considered to end up your review with a lapidary "not meant for greener and silencers"?

I may have missed the point here, but what were the noise levels of the cooler? Is the pump speed adjustable?

I guess the cooler hasn't been thoroughly reviewed. At anyway, it's a run of the mill closed loop kit, I think it cannot be that good (for a silencer, at least).

Thanks for the review. I think the Photoshop bump is the amount of RAM and not the BW.

Some errata:

Page 1: caption for "Nehalem (LGA1366)vs. Sandy Bridge Extreme (LGA2011)." should be reversed.

Page 6: bottom 'discuss this article' link is bad.

I think Intel are just mocking AMD now. Bulldozer is such an epic failure.

While I still can't figure out what the extra silicon in Bulldozer (compared with Phenom II) is doing, since it's not contributing much to performance,
the visible extra silicon that's not used in the Sandy Bridge E makes me call it a 'management flop'.
I doubt that the Intel 32 nm process is still not mature enough to require 2 out of 8 cores to be disabled, and maybe also some cache disabled.
But using all 8 cores would have increased the TDP too much, if the frequencies wouldn't have been watered down.
Intel's management should have decided to have only six cores native, and also to keep it on the Socket 1366,
for compatibility, in which case I would have been happy to buy one.

Well there is a hint in the price tag. It suggests that their yields are low. In that light it makes perfect sense to write off two of six cores if your process can barely turn out that many. It will be interesting to see what the 4 core version using the same silicone will cost.

will be nice to see 6 or 8 dimm packed matx board, or some mini-itx x79

@MoJo: I wrote 2 out of 8, not 2 out of 6. And if you look again at the die image in the first page of the review, you'll understand why.
My point was that they should have aimed for 6 cores, and probably the management wanted 8, but it wasn't really possible with 32 nm.
With the Ivy Bridge die shrink, it sure will be possible. And could be just half the price of SB-E.

I doubt there will be any 8 dimm micro-atx mobos for it, I have only seen full ATX ones. A friend of mine wants to get one for a virtualization server with 32 GB.

Have you looked at Intel's profits? It suggests some of their prices are decoupled from their costs.

I think they dropped 2 cores because they can and because it maximizes whatever their yields are. It's not necessarily an indication of yields though.

I imagine a 6 core Xeon based on this will show up for lotta money.

It's the same die that will be used for the upcoming Xeon 8 core products. They just turned off 2 cores and some cache to make this "enthusiast" product. Nothing more, nothing less. While it's a huge die, I'm sure they are priced for Intel style margins.

Would you mind sharing the info, which upcoming Xeon products? I couldn't find any with 8 cores.
Usually upcoming Xeons are listed on a page at Wikipedia, but there are none with 8 cores, Sandy Bridge based:
http://en.wikipedia.org/wiki/List_of_fu ... processors
Intel have however 10 core Xeons, 32 nm based on Westmere architecture:
http://ark.intel.com/products/53580

There is list of what is claimed to be future 8 core processors detailed at http://en.wikipedia.org/wiki/Sandy_Bridge under the 'Server Platform' heading.

Thank you for the link, either I missed it or it was updated recently.
The top 8 core Xeon, at 3.1 GHz and 20 MB cache, would have a TDP of 150W. So the current SB-E also have 5-10 MB L3 cache disabled.
If Intel would push it to 3.3 GHz, it would rise the TDP to 170-180W, maybe more, depending on voltage requirements.

Something doesn't check out. Tell me what I'm missing:
The power consumption at load is very high compared to the SB quad-cores (not a bit more than 50% higher). It's apparently not some terrible inefficiency or a mistake because the Handbrake score is better than I would have expected as well. The power is apparently not being wasted (check the ratios for yourself).
It's as if the SB quad-cores were partially idle during the load tests, maybe because they were frequently waiting for data during the test (the 3960X has more cache) or maybe because the load was not properly threaded. In that case, SPCR's power consumption measurements would be flawed.
But maybe there's a simpler explanation I'm missing. Aren't the SB and SBE basically the same cores?

Well, the CPU cores are the same, but there are some implementation differences - like the memory controller. All that "available" bandwidth, but hardly any benefit in the benchmarks. So, it is the part, or the immature bios/drivers, or s/w? <shrugs> Wait a month and see what's changed/matured

There seems to be a general misunderstanding as to where this platform fits into Intel’s strategy. It stems from the workstation platform that has also been ‘relegated’ to serve as a high end desktop platform. It’s a niche market so it’s not practical or financially viable for them to develop a native 6 core CPU just for this market sector. It makes a lot more sense to use the strategy that they are using and they like to keep clock speeds high especially for desktops parts hence the 6 core strategy for now.
This was their strategy from the beginning so there is no implication that it has anything to do with poor yields. They’ve had major issues with the chipset but I haven’t heard any rumours about yields being poor.
They could possibly release an 8 core version in 6 months or so with a new revision but at least there is IB-E in about 12 months which is where the fun really starts.


They already have a native 6 core 32nm CPU for LGA1366.

The Westmere 6 core is 32 nm but is just a die-shrink of Nehalem (I currently have an i7 920-D0), so it doesn't have higher IPC.
I wanted a Socket 1366 6-core Sandy Bridge, so this is kind of my personal rant against Intel's strategy.
If I were to buy today a Core i7-980, it would be slower than a Core i7-2500 in applications that don't use more than 4 cores.

I’m sure it does have a slightly higher IPC which is usual even for so called die shrinks. It gains extra instructions over your chip as well.


That’s just totally unrealistic. Intel are well known for wanting to push the overall platform performance which means they have little concern for backward compatibility.
I know people like to moan about backward compatibility but Intel are targeting the mainstream market not the upgrading market and this is a Workstation/Server derivative where upgrades are in single digits apparently.
Focusing on backward compatibility means designing for <10% at the expensive of the >90% which is madness.


The whole point of buying a six core CPU is surely because you need more performance in the heavily threaded (>4 cores) applications that you use. If that means that more lightly threaded applications take a slight hit then surely that is worth it if you are getting a large gain for heavily threaded applications?
Every situation varies but you sound like you are whinging just for the sake of it.

I've always felt very high end expensive processors (and we can go back many years on this) were/are very poor value for money esp as in the longer term they get beaten out by much cheaper alternatives as time goes by. The review was interesting but rather highlighted this critical point.

One for the pc boy racers with deep wallets. You can build a very good entire pc for less than the cost of this CPU

I think the situation was worse in the past when the $1,000 CPUs were merely a speed bump whereas at least this offers 50% more cores than the mainstream processors.


The $1,000 Extreme version does seem an indulgence when you can get 95% of the performance with the i7-3920K for ~56% of the price.


You are paying a premium for having tomorrow’s performance today.
If I could regularly utilise all 12 threads I would buy an i7-3920K as when you look at the overall cost of a PC including the screen, software etc the difference in price between a system with an i7-2600K and an i7-3920K seems reasonable considering the performance difference. Roughly an extra $240 for the CPU and $120 for the mobo so ~$360 in total. So for even a $1,000 system this equates to an extra third on the cost for up to an extra 50% in performance. If the percentage gain for performance is higher than the percentage gain in price I consider that a good deal.

I'm not even sure if the primary purpose of the Extreme models is to actually be sold to someone.

I imagine it's mostly there for marketing purposes: To throw glory at its lesser siblings, and to make the rest of the lineup seem cheap in comparison.

While i don't disagree with your statement, there other things to consider, like Ivy Bridge is like 4 months away, with rumors saying it will be 10-20% faster than Sandy Bridge, that should put the i7 3700K right next to a i7 3920K, for $300 less. Personally i would take the ivy bridge more because there is not that many apps that will take the advantage of the extra cores, so i would prefer 4 cores with similar cpu power as 6 cores, specially in games i dont know even 1 that uses more than 4. Now there might be apps specially design for multithreaded and takes the advantage of 6 cores 12 threads... but to justify the $300 extra... and to justify a much more expensive mobo, not to say power consumption.... its just not worth it for the average user.

Once Intel change their release of their two platforms the lower end becomes much more attractive. What i mean is that when the x58/1366 i7 920 was release was a big jump over 775, there were no 1156 on the market to compete, like a year later we saw the 1156 which offer good value, performance and consumption. but the i7 920 still hold its own, Intel shoulda released Sandy Bridge E X79/LGA2011 to replace the X58/1366, but instead they released Xtreme CPU with 6cores for the X58/1366 and they went main stream with Sandy Bridge 1155 and we got really cheap cpus that could compete with $1k i7 980x, thus the need of new platform, but with them changing their order releases, with mainstream before high end, they are imo just killing their high end market, i would have gladly payed for Sandy Bridge E i7 3930K $600 at the begging of the year (and intel delaying the mainstream 1155 for November, like right now, and ivy bridge none E for Nov 2012), as i would have gotten a cpu that would perform as Ivy Bridge two years in advance for the extra $300, but now its backward, with Sandy Bridge E you get CPU that cost almost twice as much that in less than six months time there will be a similar with Ivy Bridge, with much less cost on the other hardware needed to run.

Intel needs to revert their releases for their E line, being before the none E, to make sense into spending that much. Either way for me its Sandy Bridge for now, probably CPU upgrade on Ivy Bridge and waiting for Haswell (not E), ill skip the full E line at least for a couple of years.

Abula, do you know SPCR's numbers are wrong or misleading?
You talk about 10-20% as if it was going to matter. But the number of cores is not this chips' only advantage. SPCR claims this is 59% faster than a 2600K for a real-world task...
Of course, the 2600K has never been the CPU to beat for this type of load anyway.

Hello,

Are there any good articles discussing quite CPU coolers for the 3960x? I usually use the Scythe ninja but I'm not sure what is the right CPU cooler for the LGA 2011 socket.

thanks

I haven't seen any articles, but frozencpu.com has a Noctua NH-D14 listed as being lga 2011 compatible, and they have socket adapters for a number of existing coolers, including notcua, but they're out of stock of the prolimatech adapters. Since the 2011 has the same socket holes as the lga 1366, it looks like the adapters are mostly just screw differences.

Yup. The mounting hole locations are identical, but rather than go through the board, the screws mount into the metal frame surrounding the socket on the top side of the board. A lot more convenient than pop-in plastic or bolts w/ backplate, and probably as secure as the latter.