Talk:Intel Core i7

is Core i7 gonna be in laptop? Lf2074 (talk) 21:04, 21 August 2008 (UTC)

While I don't think anyone but Intel knows the answer to that right now, given how the Nehalem product line is structured, I doubt there will be Core i7 laptop parts. Since Core i7 represents Bloomfield and there's no mobile analogue to Bloomfield, I expect that branding to be used for desktop processors only.

Now if Lynnfield (LGA 1160 desktop quad-core processors with two channels of DDR3 and integrated PCI Express controller) turns out to be, say, Core i5 and Havendale (LGA 1160 desktop dual-core processors with integrated graphics and PCI Express) turns out to be Core i3, given Intel's precedent in using the same branding for analogous desktop and mobile processors (Core 2 Duo Exxxx vs Txxxx, Pentium Dual-core Exxxx vs Txxxx, Celeron 4xx vs Celeron M 5xx), the corresponding mobile parts (Clarksfield quad-core and Auburndale dual-core respectively) might carry the same branding (in this example, Core i5 and Core i3 respectively). —Observ (talk) 05:59, 1 September 2008 (UTC)

I think that the mobile chips will be some variant of the Core i7 brand name, but like Observ said above, it most likely won't be i7 because that name is for Bloomfield. -- Imperator3733 (talk) 19:11, 2 September 2008 (UTC)

---

I think there is an error, the 2 basic models (800/900) officially only support DDR3 up to 1066 MHz, and the Extreme Edition will support up to 1333 MHz memory, check the table here: http://www.techarp.com/showarticle.aspx?artno=556&pgno=2 —Preceding unsigned comment added by 79.180.238.44 (talk) 19:38, 2 September 2008 (UTC)

Transistor count

• 731M transistors for the quad core version.

With stupid: i7 is quadcore-only until now (april 2009). —Preceding unsigned comment added by 62.194.207.186 (talk) 20:38, 4 April 2009 (UTC)

FSB

In the infobox, it says "FSB speed: 4.8 GT/s", this is a mistake, i7 will discontinue the use of FSB. I think we should come up with a new infobox. Greetings --201.212.140.93 (talk) 04:20, 9 September 2008 (UTC)

I tried to edit it to QPI, but now it won't display, I guess it's predefined info box. Still, I am not changing it back, no info is better than wrong info as i7 has no FSB —Preceding unsigned comment added by 58.35.10.117 (talk) 13:10, 24 September 2008 (UTC)

Don't wory, I have created a new infobox - CPU QPI. It displays, but unfortunately I believe I broke something, as GHz and GT/s won't link. If anyone can fix this, please go right ahead. Nintendo 07 (talk) 00:02, 26 September 2008 (UTC)

System bus speeds and CPU clock multipliers

Even chips without a front side bus still have a system bus and a multiplier. I'm currently trying to figure out what they are for these chips. They should be in the same approximate range as Intel's others, though. 2933 is not evenly divisible by 333 or 400, so I figure that it must run on a 266mhz bus with a multiplier of 11. The lower-clocked chip is not likely to run on a faster bus, so it's probably 266mhz x10. The 3.2ghz part is harder to pin down, as it could be 266mhz x12 or 400mhz x8, but the smart money is on the former. I sincerely doubt that Intel would skip straight to a 533mhz bus without going through 466mhz first. —Preceding unsigned comment added by 68.124.70.221 (talk) 01:25, 27 November 2008 (UTC)

The CPU-multiplier is not to the system bus in the Core i7 CPUs anymore, but it is to a genereal reference clock, which is 133MHz on current Core i7 CPUs. So the multipliers are 20, 22 and 24 for the model numbers 920, 940 and 965. --MrBurns (talk) 19:03, 4 December 2008 (UTC)

Pricing

This website reports different pricing than what is shown here. 69.115.150.3 (talk) 01:02, 17 September 2008 (UTC)

I notice that website's source is Hardspell. If you follow the link, the Hardspelll website only has prices in Euros. Neoseeker has seemed to just convert euros to US$. This is probably a flawed way to go about it: it is likely the European prices are more than the American.219.90.147.95 (talk) 10:26, 25 September 2008 (UTC)

There are two dollar amounts given for each CPU - what's the difference? --Vossanova ^o< 14:20, 21 October 2008 (UTC)

Please read the note immediately after the table. There are no official prices from Intel yet. the three prioces given are from three different rumors reported in the trade press. look at the references for details. -Arch dude (talk) 14:31, 21 October 2008 (UTC)

Okay, let's just go ahead and remove them then. We shouldn't have any rumors or speculation here, even if it comes from a "reputable" site. --Vossanova ^o< 14:35, 21 October 2008 (UTC)

Much of this article is speculative: that's why it has the "Future chip" box at the top. As we get nearer and nearer to the actual launch date things become more clear, but until Intel actually tells us we do not really know. By now, the only major remaining unknowns are the actual launch date and the actual pricing. I did not put the prices in the article. I did try to make it a bit more clear that these are estimates. The price situation was not as out of control as the date situation, which I addressed by adding a paragrapth. Perhaps we need a paragraph for the price situation? It's not clear to me on what grounds we can remove a rumor: the existance of the rumor meets the WP standard for verifiability. -Arch dude (talk) 17:27, 21 October 2008 (UTC)

While having prices at all is a little questionable (see WP:NOPRICES), most of the "Lists of ... processors" articles do have launch prices (wholesale, not retail), so I've accepted it. But the way they're given in the table now is outright confusing - two different US prices, and a UK price for a retail pre-order that was up for a limited time? It's a little out of hand. I would suggest moving all that info into a new paragraph (e.g. X listed the prices at $$$, Y listed the prices at $$$), and let people decide if that paragraph is worth keeping or not. --Vossanova ^o< 15:45, 23 October 2008 (UTC)

Looks like they're out, but they're priced a bit differently than on the page. Newegg has them at $319.99, $599.99 and $1069.99. Not sure how you cite prices that can vary from place to place. -RandomEngy (talk) 23:13, 17 November 2008 (UTC)

Ordering

I've changed the processor order to go from lowest to highest to match all the other "list of processor" articles. --24.82.242.132 (talk) 08:23, 23 October 2008 (UTC)

Clocking

Just my two cents clocking wise with Super Pi is that the program was written about 13 years ago and therefore does not take into consideration any new instructions since then. If a benchmarking program was used that could make full use of the new instruction set, I think we would see a larger, positive, gap. —Preceding unsigned comment added by 24.57.133.8 (talk) 01:47, 27 October 2008 (UTC)

Excellent point. we need to fix the Super PI page. and then put caveats on any modern page that refrences this benchmark. -Arch dude (talk) 03:19, 27 October 2008 (UTC)

Core i7 is out in the wild

I don't know if someone at the store had slippery hands, but an i7 has been sold to a lucky guy over at OC.net

http://www.overclock.net/intel-cpus/406091-i7-out-wild.html

Chris122990 (talk) 02:49, 3 November 2008 (UTC)

Handling the 3 November articles

Collegues: There are a flurry of new reviews: more than twenty new articles in the last few hours. We need to avoid chaos in our editing. Any thoughts on how to do this? I think we need to remove the older review refrences and then pick two or three of the most reliable of the new ones. Thre is also a chance that Intel will make some official announcements or even actually release the product. I slapped a "current event" header on the article. I also re-insteded the "future chip" header until we see an official announcement. -Arch dude (talk) 11:03, 3 November 2008 (UTC)

This was removed by someone (not me) because there didn't seem to be a flurry of edits. I agree with the removal. --XaXXon (talk) 20:50, 12 November 2008 (UTC)

Regarding Hyper-threading

With regards to HT being only useful when there's more than 4 threads, in a normal HT processor like Atom or Netburst this statement would be true. However, this is not always true with the i7 due to its ability to overclock itself when fewer than four cores are in use. It is conceivable that two tasks might be able to execute faster on a single, overclocked core via HT compared to having them run on two separate cores running at a lower clock speed, particularly in cases where one task is memory bound and the other not. Even if this functionality is not immediately obvious with current benchmarks, this may change in the future as Intel releases updates to the power management firmware and Microsoft updates their thread scheduler to take advantage of this new capability. --JLTate22 (talk) 15:03, 4 November 2008 (UTC)

Sorry, no. Maxium "turbo mode" increase adds 3 to the multiplier, which is a max gain of 400Mhz. even on the 920 (that starts with a multiplier of 20) the "turbo" gain is at most 15%. For higher multipliers (i.e., the 940 and the 965) the turbo gain is a lower percentage. The best possible HT packing efficiency as reported by Intel is about 30%. There is no way this can match the 100% gain you get by using a second processor. And this is assuming that neither of the two threads suffers from the resource-sharing penalty imposed by HT. A typical HT gain is closer to 15% that to the max theoretical 30%. Furthremore, as compilers become more aware of the Nehalem microarchitecture, the HT gains will be lower and the HT penalty will be higher. -Arch dude (talk) 15:18, 4 November 2008 (UTC)

I amended the article to read "when all cores are fully powered" for both the reason you state as well as low-power mode. It's possible that in some applications requiring careful power management it is a much better choice to leave 2 cores powered down during many operations and run 3 threads through the remaining 2 cores. Hyperthreading will offer improvements in this case. And Arch dude, while what you say is true it's beside the point-- if you consider that the processor can find itself in a situation where some cores are powered down, the overclocking and hyperthreading together WILL help performance. Perhaps not as much as starting up the next core, but it will still make a difference. --Sam (talk) 19:25, 4 November 2008 (UTC)

I'm sure we could construct an application that could in theory benefit from HT with less than 4 cores active. I'm also sure that there is no way that Vista or the Linux kernel would be able to identify this optimization. Based on the flood of new benchmarks, Vista cannot even manage to keep a thread on the same processor and it does not know that which "virtual" CPUs are share a core, so it is making horrible choices when it assigns threads to CPUs. Unless you can cite reliable source, I think you should remove your insertion, although I will not remove it or flag it myself. Note also that Core i7 is not intended for systems that need tight power control. later Nehalems will address laptops, but a 130W TDP processor will not. -Arch dude (talk) 02:09, 6 November 2008 (UTC)

ECC?

A new editor added the lack of ECC support as a drawback, but we need a better cite if this is indeed the case. I removed the statement as a WP:SYNTH. Please discuss here. -Arch dude (talk) 04:10, 11 November 2008 (UTC)

That is me (talk), and I added a pointer to the ECC memory page, where the usual rule of thumb is one bit error per month per gigabyte. I and my colleagues have thousands of gigabytes in running computers, and the lack of ECC is a drawback for scientific computing. All AMD CPUs have memory controllers that has ECC. Core 2's can be supported by the X38 and X48 chipsets. But not the Core i7. —Preceding undated comment was added at 04:14, 11 November 2008 (UTC).

I'm not arguing with your conclusion, but you will need to cite a "reliable source". Please see WP:V. I also have not yet seen a source that says that core i7 cannot support ECC. Please cite a source. If you need help with formatting the cite, then just cram it in any old way and we can fix it later, but we need a source. Thanks for your understanding: this "reliable source" thing is a PITA when you have personal knowledge of a subject. (BTW, please sign your posts, since it makes these discussions easier to follow. to sign, jut type in four tildes, and the software will sign for you. Thanks.) -Arch dude (talk) 04:26, 11 November 2008 (UTC)

I don't think this should be posted until it is proved. And I don't even see it as a "drawback" for a desktop chip even assuming it's true. Nehalem-EP socket 1366 based (server) motherboards requiring ECC DDR3 memory have been announced, so at least the server grade Intel chips in this family will support ECC. And ECC is almost exclusively used in servers, very rarely in desktop PCs; ECC is also slower due to the error checking overhead, so it's less desireable for typical desktop applications, like gaming, video processing, etc. anyhow. 76.10.146.131 (talk) 19:07, 11 November 2008 (UTC)

At some point my edits said "no current core i7 boards supports ECC" and it was changed to what it was now. I am not sure by whom at the moment. I have been double-checking all the motherboards I can see: the Intel DX58SO, the two ASUS boards, the gigabyte board, none have ECC. I have not seen supermicro and tyan boards yet. If anyone can point to anything from Intel, I will be thankful. Moscito (talk) 01:38, 13 November 2008 (UTC)

Are these really drawbacks considering, today is Nov 14, and the chip comes out on the 17th? is it a drawback on the chip if its something that WILL be corrected with time? Is it really a drawback for the architecture if in 10 years none of it will be true? By then, the chip will be dead, and presumably there will have been ecc. Also can the natural pricing of new technology being more expensive than the previous generation be considered a drawback? Why not list the chips own base price as a drawback then? --67.241.177.245 (talk) 07:44, 14 November 2008 (UTC)

Speaking as just one editor, I decided not to delete this drawback yet. Once we actually get to see the Intel data sheet for the Core i7 processors, we will know whether or not the integrated memory controllers have ECC. If they do, we remove this objection from this article. (It can be moved into an article on motherboards.) If the device cannot do ECC, then we re-phrase the sentence to refer to the Core i7. We do not yet know about later instances of the Nehalem archictecture (Gainstown, Becton, Lynnfield, etc.), or even about possible new members of the Core i7 family (975? 985?) but we will know about the processors that are the subject of this article. -Arch dude (talk) 08:36, 14 November 2008 (UTC)

It seems like a silly drawback to me. Obviously if one is involved in scientific computing and also listens to Daniel Bernstein, then this is relevant. I doubt that many people will fall into that category. You might as well say that its lack of ISA slots or its not having 16 cores is a drawback. Chaotic42 (talk) 23:33, 14 November 2008 (UTC)

Ask for opinons from other editors: I'm neutral of whether or not it's a drawback, but the ability or inability of the on-board memory controller to support ECC needs to be listed in this article once we know which it is. The reason: unlike all earlier Intel x86 processors since 1994, this is the first one with its own memory controller. if missing, ECC abilility therefore cannot be added by selecting a different motherboard. -Arch dude (talk) 00:28, 15 November 2008 (UTC)

The spec sheets for ASUS bloomfield motherboards http://au.asus.com/products.aspx?l1=3&l2=179&l3=815&l4=0&model=2588&modelmenu=2. I'll try to collect all authentic ones from the manufactuers

They are releasing i7 based xeons soon, and ECC is basically a requirement with regards to servers, so there must be the ability for it. Maybe intel will add it in later revisions? —Preceding unsigned comment added by 134.36.92.18 (talk) 12:15, 29 November 2008 (UTC)

I think that the i7 has ECC capability, just disabled as market segmentation. They will likely sell you a Xeon at twice the price, identical in every way except that the ECC is enabled if you have a top-end motherboard from supermicro. Moscito (talk) 02:39, 5 December 2008 (UTC)

Intel said in their FAQ on the I7 that "typically ECC memory is used on servers and workstations rather than on desktop platforms.” However Microsoft published a report in 2007 rumored to say that single-bit error rates in DRAM may be among the top ten causes of systems failures, suggesting Windows users may need ECC: ‘Microsoft says PCs may need DRAM upgrade', http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=199601761 72.73.92.130 (talk) 06:48, 27 March 2009 (UTC)

I talked with a PogoLinux representative on April 25/2009 and he said that they ship workstations with ECC-supported Core i7 configurations. For example, here's their Core i7 standard system page: http://www.pogolinux.com/quotes/editsys?sys_id=28712 70.68.70.186 (talk) 19:33, 27 April 2009 (UTC)

You actually don't really need ECC for desktop systems, if your RAM is working correctly (it should pass at least a few hours on memtest86 and goldmemory). The problem ist that most people don't test there RAM at all or rely on the BIOS-test allone, which is unsufficient. From what I remember, the error rate of correct working non-ECC memory is about 10^(-10) errors/(bit*hour) (which would be about 100 times higher then ECC), which is about 3 errors in 24 hours, if you have 1GB RAM used. Also ECC has a lot of drawbacks, i.E. higher price, lower performance, harder to archiev high clock speeds/overclock and ECC memory-controllers are more expensive to build and need a more expensive motherboard design. --MrBurns (talk) 01:19, 15 May 2009 (UTC)

DDR3 price

The DDR3 price "drawback" is silly, but it is asserted multiple times in the trade press, so I guess we must keep it. The actual situation is that DRAM production follows a boom-and-bust cycle, with periods of massive overproduction causing market gluts and consequent price collapses. We are currently in a market glut. During a glut, slower speed grades and older technology inventory is sold below cost, and even newer stuff is sold a break-even. Manufacturers eventually cut production and balence is restored, followed by a DRAM drought and a price spike, and the cycle repeats. In the current situation, You can buy three 1GB DDR3 DIMMs for a total of about $110 US: that's less than half the price of the cheapest processor. This objection is therefore silly. -Arch dude (talk) 08:36, 14 November 2008 (UTC)

The drawback is not silly: DDr3-prices may be cheap compared to RAM-prices a few years ago, but they are still about double as expensive as DDR2. This is, because DDr3 is a new technology and most users still buyy DDR2, so DDR2 is produced and traded in larger quantities, which makes the price cheaper. I also don't think, that DDr3 is sold below production costs. this is maybe true for DDR2, but not for DDR3. Of course the price situation of DDR2 and DDR3 will change, I think about the end of 2009, when there will be Nehalems, which are not high end or at least upper middleclasse available in great quantities. At this time, there shoul be not a lpot difference between DDR2 and DDR3 anymore and in about 2 years DDR2 will be considerable more expensive then DDR3. I have a lot experience w/ DRAM-productioncycles (as a consumer), it was similar w/ EDO/SDRAM, SDRAM/DDR1 and DDR1/DDR2. --MrBurns (talk) 23:28, 19 November 2008 (UTC)

Question to arch dude: given that people don't just buy 1GBx3 -- i almost never set up a workstation without 4GB of RAM (I run Linux, many people now run 64-bit versions of vista), and that if DDR3 1333 2GBx3 is about as much money as DDR2-800 2GBx4 ECC, shouldn't we consider the difference to be non-trivial? (This are both Kingston branded memory in downtown Taipei.)

Hey, I'm just one guy, with just one opinion. Why not propose a nominal AMD setup versus a nominal Core i7 setup and compare memory costs? Here: let's use DDR3 1333 2GBx3 @ $74 = $222 for the Core i7, and DDR2 667 (1GBx2 @$12 + 2GBx2 @$21) = $64 for the AMD. The percentage amount is not interesting: the dollar amount is. (prices from Pricewatch) The 6GB costs $158 more for the Core i7. This is a trivial amount compared to the cost of the processor. -Arch dude (talk) 03:12, 16 November 2008 (UTC)

If you are using Windows x32, you can use only about 2.75-3.25GB, so it doesn't make sense to buy more then 3GB. And most users still use x32, not x64. And also most users don't need more then 3GB, you don't even need that much memory for very memory hungy games like Far Cry 2. --MrBurns (talk) 23:33, 19 November 2008 (UTC)

If only that could've been stated 1 month later after GTA IV was released for the PC... ⒺⓋⒾⓁⒼⓄⒽⒶⓃ② ^talk 05:44, 10 January 2009 (UTC)

Hey, I agree with some of your point, certainly for some of people it may be miniscule but if you measure it like that, then I say wow, Arch Dude, that's the price diff between a Q9650 and a Q9550. And I would venture a guess that most people who have high-end processors also tend to have more RAM. Moscito (talk) 03:22, 16 November 2008 (UTC)

DDR3 can be gotten for $30 Canadian a Gigabyte. At this point it's not expensive. —Preceding unsigned comment added by 64.230.86.52 (talk) 20:27, 14 January 2009 (UTC)

Conflicting information about turbo speed increases

The clock speed increases offered by the turbo mode are quoted as "133-266MHz" in one section and as "400MHz" in another. I've amended these to show that the step size is 133 MHz with no fixed maximum and provided a link to the Intel white paper on the technology. Juux 16:37, 25 November 2008 (UTC) —Preceding unsigned comment added by Juux (talk • contribs)

Early Test issues

I can't seem to access the cited source, but the article says "In the single-threaded Super PI Template:UnsignedIP -->

Please read the sentence carefully. The processors being compared did not have the same clock speed. Therefore the core i7 executed fewer clock cycles per second. Therefore, it completed the test in fewer clock cycles even though it took more time. To get a fair comparison, we need to replace this with an newer benchmark that compares a QX9770 with a Core i7 965, both running at 3.2Ghz: there are now hundreds of published results to choose from. -Arch dude (talk) 22:05, 16 December 2008 (UTC)

There must be some kind of typo or miscalculation in the Super PI results cited in the article at the time of this comment [QX9770, 3.2GHz, 14.42s; i7 920, 2.66GHz, 11.54s]. The conclusion that this indicates a 20% advantage per clock cycle for the i7 is wrong; it should be 50%. If, however, the time for the i7 should be 15.36s, as in the anonymous comment by 12.31.231.168 above, that would represent a 13% advantage for the i7 (which is a little easier to believe). I can't seem to access the website referenced in the footnote either, or I'd find the correct times myself. If yobeta.com is kaput anyway, perhaps a completely different study should take its place in the article. —SaxicolousOne (talk) 06:06, 30 December 2008 (UTC)

Where did the 11.54 seconds come from? The quoted sentence saye 15.36 seconds. -Arch dude (talk) 17:45, 30 December 2008 (UTC)

I don't know where 11.54 seconds came from, but that's been the figure in the article since a user named "Juux" put it there at 23:47 on Dec. 28, 2008. Apparently, that figure had been back and forth a bit prior to that. And since my last comment, someone has edited the article under the assumption that the 11.54s figure is correct, stating that the i7 was, in fact 50% faster than the QX9770.... —SaxicolousOne (talk) 01:53, 1 January 2009 (UTC)

L1 cache

We need to state how much L1 cache each core has. NSK Nikolaos S. Karastathis (talk) 03:43, 26 December 2008 (UTC)

Done.-Arch dude (talk) 02:35, 27 December 2008 (UTC)

Thermal Considerations

I haven't seen any mention of the temperature range that this processor runs in. I'm referring specifically to the processor temperature not the environment. —Preceding unsigned comment added by 64.230.86.52 (talk) 20:35, 14 January 2009 (UTC)

Image

Is there noone out there who bought an i7 and could make a High-Res, free image? —Preceding unsigned comment added by 78.52.132.128 (talk) 17:04, 31 January 2009 (UTC)

You don't see much more on aa high res image than on the one used now, when the heat spreader (IHS) isn't removed. But I somehow doubt that the image is a real ci7 image. I don't own a ci7, but all Intel CPUs using IHS form the first Pentium 4 CPUs to Core 2 Quad had more text on the IHS then just the name of the product line (including an 8-digit alphnumeric number (at least on Core 2 Wuad CPUs), which looks like a serial and is L738B526 on my Core 2 Quad Q6600). --MrBurns (talk) 22:09, 19 March 2009 (UTC)

Under the Intel Core 2 article you find a lot of photos of C2 CPUs w/ IHS, inclusing my Q6600 in the kentsfield section. So you can see much text on this heatspreaders and no big Core 2 logo, from what I know Ci7 heatspreaders should be printed similar. Even most Pentium 4 CPUs have similar things printed on them, see i.E. [1]. --MrBurns (talk) 23:07, 19 March 2009 (UTC)

Corei7's are "healthy"?

The drawbacks section contains this statement: "Core i7's are healthy for your motherboard (mainboard)." Really doesn't make sense to me. Spam? Errantkid (talk) 14:04, 29 April 2009 (UTC)

You probably could have deleted that without asking. I did so. --Vossanova ^o< 14:39, 29 April 2009 (UTC)

What have drivers to do with clock for clock performance?!

Under the drawbacks section is says:

more recent testing done on all clock rates of official hardware with final drivers and BIOS revisions show that Core i7 at the very least beats Yorkfield clock-for-clock...

How on earth can a driver (who is run by the processor) make the processor perform better?! A CPU have a number of instruction it can run pr clock and a good driver will maybe use fewer instructions so that you can run more code pr clock. So indeed we can get a performance increase by writing better software but the CPU itself is NOT able to execute more instructions pr clock (perhaps unless a microcode update is performed).

So my point is : with smarter tuned software you are able to do more stuff in a set timeslice but the processor itself can't execute more instructions pr clock than what it was designed for! —Preceding unsigned comment added by 80.213.182.76 (talk) 09:51, 2 May 2009 (UTC)

Furthermore, shouldn't this section just not be in Drawbacks? It basically says "tests from before the hardware was official were not so good, but now they're good" right? Really it's not a drawback then. 71.58.213.116 (talk) 17:54, 16 May 2009 (UTC)