The Lynnfield Preview: Rumblings of Revenge
by Anand Lal Shimpi on May 29, 2009 1:00 PM EST- Posted in
- CPUs
Lynnfield’s Secret? Turbo Mode
The current Core i7 runs at a multiple of its BCLK, which is 133MHz. The Core i7-920 runs at 20x BCLK (2.66GHz) while the i7-965 runs at 24x BCLK (3.2GHz). If the chip isn’t running anywhere near its max TDP, the i7 will overclock itself by one speed bin (133MHz). Now say you’re running a single-threaded application that’s got only one core active, if the chip is cool enough the i7 will overclock itself by two speed bins (266MHz).
Turbo mode on the i7 works well and as I found is generally responsible for a 3 - 4% increase in performance. Despite its benefits, the i7’s Turbo mode is very conservative. Our own overclocking tests show that i7s have no problems running at up to 3.8GHz without any additional cooling or voltage, even with all four cores under full load. Given that it was Intel’s first Nehalem architecture, I can understand the hesitation not to go crazy with turbo.
The second version of Intel’s Nehalem turbo mode came with the high end Nehalem Xeon processors. While the single-socket processors worked just like the desktop i7s, the higher end dual-socket Xeons can turbo up more aggressively.
The Xeon E5520, E5530 and E5540 can all boost their clocks by one bin if 3 or 4 cores are active, but two bins if 1 or 2 are active. The X5000 series gets even more aggressive; with 3 or 4 cores active the chips can overclock themselves by up to 266MHz, and if 1 or 2 cores are active they can turbo up an additional 400MHz.
Lynnfield marks Intel’s third generation Nehalem turbo and is correspondingly more awesome.
The leaked roadmaps show that the two higher end Lynnfields can turbo up to five bins, or 667MHz while the entry level Lynnfield can turbo up to four bins (533MHz). I’m guessing this is for a single active core, but what about when more than one core is active? The table below has the turbo specs for the Xeon X5570 (95W TDP), the Core i7-940 (130W TDP) and my guesstimates for the 2.93GHz Lynnfield (95W TDP):
| Processor | Clock Speed | Max Turbo | |||
| 4 Cores Active | 3 Cores Active | 2 Cores Active | 1 Core Active | ||
| Intel Xeon X5570 | 2.93GHz | 3.2GHz | 3.2GHz | 3.33GHz | 3.33GHz |
| Intel Core i7-940 | 2.93GHz | 3.06GHz | 3.06GHz | 3.06GHz | 3.2GHz |
| Intel Lynnfield ESTIMATE | 2.93GHz | 3.2GHz | 3.2GHz | 3.60GHz | 3.60GHz |
Given the similarity in clock speed and TDP to Intel's Xeon X5570, I'm guessing the 2.93GHz Lynnfield will follow the same 2/2/3/3 turbo pattern as the Xeon. The higher max turbo frequency means that we'll at least see 3.60GHz with only 1 core active and I'm not really sure what will happen if two cores are active; if Intel follows the Xeon pattern then we'll see 3.6GHz as well, but we may very well see 3.33GHz or 3.46GHz instead when only two cores are active.
Either way a quad-core Lynnfield, thanks to its aggressive turbo mode, will end up delivering good performance regardless of the number of concurrent threads. Forget about it being a quad-core CPU and just think of it as a CPU that will perform as best as possible given its 95W thermal envelope. This, my friends, is the future of multi-core processors. It doesn't matter how many cores you have, just view them as execution resources; if you only need two powerful cores, that's what you get, and if you need to run 8 threads then that's what you'll get. Imagine what we'll get on the 4th or 5th generation of turbo modes.
To those who are wondering why Lynnfield even makes sense, I believe its turbo mode will be its saving grace. I'm more puzzled by the i7-920 at this point.
Facebook
Twitter
RSS
95 Comments
View All Comments
fpaat - Tuesday, August 4, 2009 - link
Cheap as*es who keep buying these lower spec broken underclocked BS parts contribute to the decline of PC gaming.There is exactly zero reasons to release so many different chips, except for crap manufacturing.
Buy an i7, be done. i5 = celeron = piss off Intel. ATI and NVIDIA can piss off as well for release 10 different versions of video cards per generation.
fpaat - Tuesday, August 4, 2009 - link
I LoL at people who still buy the "celeron" equivalent of computer parts.It's fine if your a teenager on a budget, but if you're an adult still buying gimp'd parts because you can't afford proper components, well, fail.
grimpr - Thursday, July 16, 2009 - link
Smart users would simply buy a fast SSD disk and a Windows 7 Home premium 64bit pack. Anand may glorify to heavens Lynnfield i5 but there is not a single excuse for dumping perfectly good hardware from 2006 upward for Intels new mainstream platform. Best upgrade would be in 2011 with a brand new platform from AMD, new CPU's from both companys that carry the AVX instruction sets, cheap low latency and fast DDR3-1600 ram as a standard, mature,cheap and fast DX11 cards, PCI-EX 3.0 and many more things.Hrel - Saturday, June 6, 2009 - link
I HATE their stupid price fixing screw the consumer crap. CPU's over 200 dollars are useless, that's simply too much for a CPU. They won't give us dual cores based on nahaylem cause they'd have to charge 190 and less; OH NO! Greedy bastards. I REALLY want AMD to get back on top so prices get driven down again.It's INFINETELY infuriating that the ONLY way that intel charges fair prices is if someone else MAKES them.
Hrel - Saturday, June 6, 2009 - link
Seriously??? E8400 anyone? I REALLY don't care about having 8 threads, like even you said in this article, doesn't really help consumers. I'm far more interested in highly clocked dual cores, and wish Intel would release dual core CPU's based on the new architexture, throw HT onto a solid dual core and I can't see any games in the near future even fully utilize it.I'd MUCH rather see a 4GHz dual core cpu with HT than some more stupid quad cores at sub 3GHZ clock speeds. This is 2009, we need to leave the sub-3GHz range in the history books.HexiumVII - Tuesday, June 2, 2009 - link
Lynnfield looks mighty good. Unfortunetly USB3 and SATA3 are looming. Since they won't be integrated into mobos for a while, high bandwidth on the PCIe slots for add-in cards will be crucial for the transition. I don't know about you buys, but USB3.0 will be the most exciting upgrade we had in a while.MasterShogo - Monday, June 1, 2009 - link
I'm not really sure if this has been talked about at all, but it seems to me that the option for an Nvidia-supplied chipset providing integrated graphics, and it still be good, isn't completely lost.I think you have to ask, why does a person want Nvidia integrated graphics in the first place? I see two reasons. For one, to provide a "business-like" or super-simple PC that has adequate graphics performance for normal stuff and simple games, but with a powerful processor. For this use, Nvidia could just make a chip that uses 8 of the PCIe lanes coming out of the CPU socket and leave the other 8 for the user/vendor to optionally provide a faster video card. Sure, that slot wouldn't have the awesomeness of a 16x PCIe slot, but it would be perfectly fine for this machine. Plus, the added video card could still use hybrid SLI or something similar for power-saving purposes...
Which leads me to the other reason people (like me) might want it. Power savings via hybrid-SLI. I would personally want all 16 lanes coming out of the CPU socket for my SLI setup or just one monster card. But in this case, I think it would be perfectly fine to have a 3rd party P55-like chip provide hybrid-SLI video for power savings. I wouldn't care about the starved DMI link, because I would never expect that video chip to do anything difficult, that's what I have the other video cards for! This is assuming the drivers work well enough to switch over very seamlessly when it needs to happen.
In other words, I think there is still enough input/output throughput for Nvidia to sell chips that use integrated graphics, but it is just going to be more spread over the board now, making their electrical design more difficult. Possibly requiring two different chips to provide for the two different scenarios. I don't see why they couldn't do it, though.
Or even better, make a 3rd party chip that can accept, as an interface, both the DMI and the PCI links coming out of the socket, but only using what is available in the particular board you are manufacturing at the moment. Maybe that would be too expensive, though. Any thoughts on how feasible that may be?
In any case, I don't need the three memory channels. It's not that I don't WANT them, it's that they don't help me at all. It's not like I'm running a major database server in my bedroom or anything. But having a direct PCI connection to the CPU would be cool, and I think this processor is just super-duper for a variety of reasons.
Kreed - Monday, June 1, 2009 - link
In terms of Photoshop CS4 performance, would a 2.8GHz i5 outperform 2.66GHz i7 920?Beno - Sunday, May 31, 2009 - link
if you test the i7 920 with HT off, lynfield will lose number 1 in those gaming benchmarkswhy didnt you bench it without HT?
genkk - Sunday, May 31, 2009 - link
is there any new mobos like the dragon platform to bench, anad is even using an old catalyst driver......... your 65% chipzilla and 35% chimpzilla