Source: Tom's Hardware – Keywords: Intel, Core-i7, Nehalem
Categories: Hardware
Size Comparison--Bigger Die, Fewer Transistors
The Core i7 Processor
Like the Core 2 before it, the Core i7 processor (code name "Bloomfield") will be produced on a 45 nm process using high-K dielectric and metal gate technologies. The Core i7 also sports four cores. In a departure from Intel’s previous quad-core designs that were comprised of two dual-core processors sharing a processor package, Nehalem’s four cores all reside on one die. A look at the die shots reveals that Intel has designed its CPU to give the company an option of producing it as a dual-core part as well. Thus, it is feasible that we will see dual-core versions of Nehalem at some point.
| Model | Process | Size | Transistors |
|---|---|---|---|
| Core i7 | 45 nm | 263 mm² | 731 M |
| Core 2 | 45 nm | 2x 107 mm² | 2x 410 M |
| Core 2 | 65 nm | 2x 143 mm² | 2x 291 M |
Putting all four cores on a single die doesn’t really save a lot of space on the processor package since the die size has increased from 214 mm² (2 x 107 mm²) for the Core 2, to 263 mm² for Core i7. Since we know that 32 nm versions of the chips are already in development—thanks to Intel’s tick-tock cycle—we can assume that the Core i7’s larger heat spreader will be able to house six or even eight cores when the die shrink happens. Perhaps that is even the reason for the larger processor package in the first place.
In the new design, the number of transistors has dropped from 810 million to 731 million. The Core i7 CPU also has some contact pins on the top of the die, but it is obvious that Intel only uses these during manufacturing.
Core i7 Models
Intel is initially introducing three different Core i7 models: the Core i7 920 running at 2.66 GHz, the Core i7 940 at 2.93 GHz, and the Core i7 965 Extreme, which is clocked at 3.20 GHz.
| Model | Clock Speed | QPI | OPR | L3 Cache |
|---|---|---|---|---|
| Core i7 965 Extreme | 3.20 GHz | 6.4 GT/s | yes | 8 MB |
| Core i7 940 | 2.93 GHz | 4.8 GT/s | no | 8 MB |
| Core i7 920 | 2.66 GHz | 4.8 GT/s | no | 8 MB |
OPR = Overspeed Protection Removed
Quick Path Interconnect
With the Nehalem architecture, Intel is finally saying farewell to the classic front side bus interface we have known for so many years. Instead, Core i7 processors will utilize the new QuickPath Interconnect (QPI) to communicate with the northbridge.
On the lower-end Core i7 models—i.e., the 920 and 940—this interface offers a bandwidth of 4.8 GT/s, which equals a bidirectional bandwidth of 9 GB/s. The Core i7 965 Extreme, on the other hand, comes with a faster QPI connection of 6.4 GT/s, or 12.8 GB/s. That’s the exact same bandwidth that the classic FSB of the Core 2 offered at 400 MHz.
AMD’s HyperTransport protocol can transmit up to 25.6 GB/s at 3.20 GHz. However, since the bandwidth-consuming memory interface is no longer part of the northbridge—having instead been integrated directly into the CPU—the advantage of switching to QPI will not have too much of an impact in the desktop segment. Here, the QuickPath Interconnect will only have to handle the data from the PCI Express connections and the southbridge, which is connected to the northbridge via PCI Express as well.
In contrast to the classic FSB, QuickPath Interconnect enables direct communication with another processor. The result is improved performance, since using the FSB design, such communication was previously only possible via a detour through the comparatively slow northbridge.
QPI brings a fascinating possibility to the table: in order to create a dual-CPU motherboard, a company simply has to solder a second processor socket onto its PCB. Since the processors can communicate with each other directly, this option is independent of the chipset, making it very simple, not to mention inexpensive. A good portion of today’s software shows marked performance gains thanks to the four additional virtual processors provided through Hyper-Threading. When adding another CPU with real cores, the performance improvement should be even more pronounced.
Inside the Core i7 processor, the base clock with its multiplier for setting the correct clock speed still lives on. It runs at 133 MHz, meaning the multipliers are higher than those found on the Core 2 models.
- Previous page Eight Virtual Cores Through...
- Next page SSE 4.2 And The Technical Nitty-Gritty
- 1 / 2
- Next
-
Actually, Ci7 OCs quite well...
http://www.custompc.co.uk/news/605 [...] n-air.html
http://www.custompc.co.uk/handsong [...] cking.html
custompc.co.uk...ehhh...it's quite lame and unprofessional
mi1ez obviously not. Considering you can only over clock the $1000 pieces of crap E8400 will do me well for the next 3-4 years until Amd catches up or intel allow end users to dowhat they will with their cpus.
Ah well, the O/C limit or a similar barrier was inevitable, while I really enjoyed being able to get such a great O/C (and will continue to now that Penryn's are going to be coming right down in price
) I can fully understand this move by Intel.
Nice chip to btw, but I don't think I'll be looking at getting one for a loooong time
Screaming fast - obviously much much faster than the 9950 B.E. from AMD.
But the AMD retails for £140 - I could feasibly build a whole PC based on the 9950 for the price of one of these uber Intels.
I don't want to come across as a fanboi for either side - I would like to come across as someone interested in know which CPU has the best performance for any given price point - y'know like you used to do and still do to some extent in the GPU monthly round up.
I just hope this pushes down the price of Core Duo and current Quad Core cpu's to a more affordable price point, AMD will have to practically give their CPU's away now
As always - No software = no point.
No games will be made for this - Most will all held back for consoles.
Good for small art productions houses. A little ILM render farm on the desktop for $1000.
I'd like to see a banch mark for,
"This time 5 years ago you would have needed 10 computers to get this work load done". Not a needed benchmark; but as this is a new cpu it would be great to know how far we've come over the years.
ALSO,
Intel bought Project Offset http://www.projectoffset.com/ a while back. I would hope they made them write the game to show off the i7.
Intel would to well to fund its own games division just so people have something to use when the new system comes out.
Any thoughts or inside knowledge about this being true?
Intel has in effect laid down the Gauntlet for AMD. And let's not forget AMD's popularity in the GPU market. I don't think AMD is in that bad a position.
The Forged Alliance benchmark table seems to be written in thousands with a random comma, and indicates a core i7 extreme 965 @ 3.84ghz gives 3635fps.
Compared to Core 2 technology, AMD's inter-core bandwidth is still superior especially in clusters. If I could have the choice between many Phenom's or a new, expensive and potentially unstable architecture like this, I know what I'd choose. Of course, it is different if one is building workstations where money is not an object. But I do not think AMD is suddenly bad just because Intel brings out yet another revision of it's own product range. In my opinion the only bad thing AMD did was appear to become complacent with their product portfolio. But on the bright side they have done some great work in the GPU field.
Do I smell the chemical tang of toasted NetBurst in the air? Oh, and thanks for the unbiased article Toms! XD
Sorry, but Ci7 stinks of the same smell of fail as the dusk of the P4 architecture, for the same reason - Intel's creating processors that are simply too high-strung. A note of irony (particularly distasteful to the Nintel fanbois) is that the Ci7 seems to be suffering the same issues as AMDs K10/Agena quads...
18 months after we're flabbergasted by €59 E2180s hitting ~110% OCs on air Tom tries to be proud of OCing a €1049 Ci7 by a whopping 20%. One particularly amusing setence tries to put things in perspective by saying the limiter had been set to 300A. Did you really believe you could finally hit 3.85GHz by dissipating ~420W through the chip? XD And I'm not trying to compare a dualcore to a quad, but how big an advantage will you expect from the 3.84GHz Ci7 when gaming when compared to the OCd 3.7GHz E2180? More importantly, would it justify a CPU 17.5 times the price? At this point I can't even tell if Intel put in the limiter through greed, embarrasement or fear over the Ci7s tight, tight tolerances.
And as for AMD... yes the K10 was tits, yes it has no high-end CPUs at the moment, but such blatant bias! Let's see... 2.65 times faster than the 9950? Okay, I read the article too, and the only tests that support that are some synthetics. At one point you make this claim just one line above a benchmark table that puts the lead at 62%! So does Toms have a serious issue with simple maths or are they a subsidiary of Nintel? And can we really trust their judgement in either case?! AMDs fastest CPU may be relatively weak, but its cheaper AND faster than Intel's slowest 45nm quad. And that quad that sucks for being 62% slower than the fastest commercial CPU is 1/6th the price of said chip. Bang for the buck? Hmm, hard choice! Lets not forget platform costs and the fact that it can probably be OCd a few percentage points higher than a production 965 (let alone the sample!)
Not to mention that IF Ci7 is a success it'll cause mass job losses and a lot of smaller companies to go bust. Because a big part of the C2s enthusiast appeal was due to the fact that anyone with a few quid could be an overclocker. On the other hand the Ci7 puts OC back in the hands of the ultra-rich. Bear in mind that the worst is yet to come... these are the highest-binned, most efficient chips in the initial stepping, and we might not see another stepping before the die shrink. Mid-range chips will suffer greater variance in efficiency and potential OCs of a mere few percentage points. We might not even see budget 45nm Ci7s as many chips might be so wasteful they'll be throttling at stock clocks! So what happens when the box cooler is effectively the state of the art as there's no real demand for aftermarket mods except the barking-mad, ultra-mega-high-end €7000 rigs?
[/rant]
These are great chips. They are the state of the art, and I hope they rule the server market for some time to come. But like Skulltrail they have no place among 99% of PCs or even 90% of enthusiast rigs. The high end is ultra-overkill and the low end is too limiting and a step BACKWARD from the current outgoing generation. If Intel doesn't sort out the limiter I hope someone does it for them. Because if the non-ultra-high-end chips remain such bad value for us mere peasants we'll have to hope that AMD or Via launch a strong challenge to Ci7. Else mid-range gaming, not to mention large chunks of the enthusiast industry, are in jeopardy of being pushed back underground. We're not all protein folders out here Intel!
This new strategy of Intel's, to prevent over-clocking on their low end processors, is a direct consequence of their domination over the CPU market; now that AMD is no longer a real threat, Intel can sacrifice the quality of its product for greater profit. This is a shame. I hope AMD are able to come back on even terms with Intel and offer some CPUs with real performance.
Lets not forget that Intel can do whatever the #%@$ they want because if you don't want these then, hey, don't buy them! blaming intel for the choice just to think they want to irritate some guy that likes ocing is ridiculous and short minded.
That said it IS a shame the low ends cant be properly oced when the tdp is high. But lets not forget that AMD is put two steps behind now!
I really hope they can get another amd athon x2 revolution like 4 years ago with a new cpu. But they have the advantage of being very strong in gpu land with the hd 4xxx series... I would like to see what happens in one year or so.
as for the article, well it could be better.
U guys all concerned over the overclocking on core i7s dont have anything to worry about. Toms was wrong. They all overclock. The X58 motherboards will be shipping with a option in the BIOS that tells the cpu to ignore the tdp/current limits. Read here
//www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3453&p=2
I guess Toms didnt spend enough time in the BIOS.
In the last paragraph when it states that a disadvantage of the Core i7 was that you would need a new cooler and mainboard because of the physical size difference is that not what you would expect when moving up to a new technology? The same can be said when having moved from the P2 to a P3 or P4. So I cant see this as a disadvantage unless you have spent hundreds of dollars buying a particular cooling fan.
Screaming fast - obviously much much faster than the 9950 B.E. from AMD.But the AMD retails for £140 - I could feasibly build a whole PC based on the 9950 for the price of one of these uber Intels.I don't want to come across as a fanboi for either side - I would like to come across as someone interested in know which CPU has the best performance for any given price point - y'know like you used to do and still do to some extent in the GPU monthly round up.
acoording to wikipedia the 2.66ghz intel core i7 only costs £150 very cheap
Where's the credibility?
"For our overclocking experiments, we set the Overspeed Protection features to 300 amperes and 300 watts."
The power limit will be reached well before the current limit (300 Watts corresponds to ~215 Amperes @ 1.4 Volts)
That's some serious current draw!
"While Intel’s "Desktop Control Center" utility ... showed us the correct clock speed we expected from the Turbo Boost feature kicking in, we didn’t register any performance improvements in our benchmarks"
I thought it reduced the clock speeds (via multiplier) of idle cores? It's differences in the power usage / temperatures that they should have been looking for.
"this new performance heavyweight is an incredible 165% faster on average than AMD’s flagship model."
"Analysis: Core i7 965 Extreme vs. Phenom X4 9950 BE vs. Core 2 Extreme QX9770"
"Overall: -62.2% -20.2%"
What? How does 62.2 % slower suddenly become 165 % faster in the other direction? The latter implies that the intel chip operates at 2.65 times the "speed" of the AMD, whereas the figures actually show it operates at 1.608 times the "speed." (100 (%) / 62.2 (%) = 1.608 --> 60.8 % faster.) Sort it out.
Also, the performance table entry for the 920 vs. 940 (pg. 12) seems to have the opposite sign convention to the other entries in all other tables. I.e. it seems as though the 940 is 38.2 % slower than the 920!
Highly impressive benchmark scores, on the whole, though. Too pricey for me, mind: only just got a Q6600, happily running @3G stock.
Interesting: http://www.tomshardware.co.uk/Inte [...] 31375.html
Where's the credibility?
For our overclocking experiments, we set the Overspeed Protection features to 300 amperes and 300 watts.
The power limit will be reached well before the current limit (300 Watts corresponds to ~215 Amperes @ 1.4 Volts)
That's some serious current draw!
While Intel’s "Desktop Control Center" utility ... showed us the correct clock speed we expected from the Turbo Boost feature kicking in, we didn’t register any performance improvements in our benchmarks
I thought it reduced the clock speeds (via multiplier) of idle cores? It's differences in the power usage / temperatures that they should have been looking at.
this new performance heavyweight is an incredible 165% faster on average than AMD’s flagship model.
Analysis: Core i7 965 Extreme vs. Phenom X4 9950 BE vs. Core 2 Extreme QX9770
Overall: -62.2% -20.2%
What? How does 62.2 % slower suddenly become 165 % faster in the other direction? The latter implies that the intel chip operates at 2.65 times the "speed" of the AMD, whereas the figures actually show it operates at 1.608 times the "speed." (100 (%) / 62.2 (%) = 1.608 --> 60.8 % faster.) Sort it out.
Also, the performance table entry for the 920 vs. 940 (pg. 12) seems to have the opposite sign convention to the other entries in all other tables. I.e. it seems as though the 940 is 38.2 % slower than the 920! Confusing.
Highly impressive benchmark scores, on the whole, though. Too pricey for me, mind: only just got a Q6600, happily running @3G stock.
Interesting
Sorry if already mentioned.....
From article page 9
"No matter what sort of current is drawn, ASUS’ configuration always gives you one "tick" with Turbo turned on. We were never able to get more than 133 MHz out of the setting, even at idle, and there is no BIOS option to adjust the sensitivity of Turbo mode on the board."
Looking at the rampage manual there turbo settings for the 965 in the turbov software which default to a x25 at all times. Maybe if you change it to a x 26, you will get two ticks. Looks like it can only be set in the software rather than the bios....