Intel Core i7 (Nehalem) benchmarks are out

[phongn]

Intel's NDA on performance for their next processor revision, Core i7 (code-named Nehalem), has expired and now the usual suspects online have benchmarks up. In short, it does very well at multithreaded and bandwidth-intensive tasks (especially the "Extreme" edition, which runs its interconnect at full speed) but the gains in things like single-threaded applications are much more modest.

Reviews: Tech-Report, Lost Circuits, AnandTech.

[Ace Pace]

Guru3d has some very interesting Multi-GPU scaling results. Their thesis is that SLI/CF were infact CPU bound for this generation and are showing some very impressive gains with Nehalem compared to Peryn.

Thoughts?

[Fingolfin_Noldor]

The chip is fast I will give Intel credit, but because of the need to upgrade the motherboard, I think I have to hold off chucking one of these things into my box by over a year.

[phongn]

The chip is fast I will give Intel credit, but because of the need to upgrade the motherboard, I think I have to hold off chucking one of these things into my box by over a year.

By that point the "mainstream" Nehalem line will be out (with a different socket!) as well.

[DaveJB]

With this release, Intel have pretty much taken away the few remaining advantages that AMD might have had. They're going to have one hell of a task designing anything to compete with this thing.

[Starglider]

For desktop applications, performance falls well short of the hype. To be fair most of that was idiot fanboys, but even Intel claimed that the performance increase would be 10 to 25% per clock, and it's more like -2 to 10% for everything except some encoding and rendering apps. However the architecture should scale to (somewhat) higher clockspeeds than its immediate predecessor and the performance benefit will go up a bit over time particularly as more apps become multithreaded. If this had been another Penyrn style jump AMD would be instantly out of the game - as it is they have a chance, albeit a slim one.

The server chips should be more exciting. Finally I really like the fact that latency on the XCHG instruction has nearly halved; my own code makes very heavy use of that.

[DaveJB]

Rendering and encoding are the two most CPU intensive areas though, and it's where Core i7 has colossal advantages over Phenom. I can't see AMD putting out enough of an improvement in those areas, and others are likely to still be bottlenecked by I/O and GPU performance.

[Mr Bean]

For the love of Xenu yes I love the improvement but motherfuckers AMD and Intel I said I would not upgrade until there was a sub 400$ 4 Ghtz out there and your making it damn hard. Yes I know the race to 3 Ghtz was by putting clock speed over efficiency, but for bob sake we've been at or around 3Ghtz for over three years now. Can I get some extra speed here?

[Starglider]

For the love of Xenu yes I love the improvement but motherfuckers AMD and Intel I said I would not upgrade until there was a sub 400$ 4 Ghtz out there and your making it damn hard. Yes I know the race to 3 Ghtz was by putting clock speed over efficiency, but for bob sake we've been at or around 3Ghtz for over three years now. Can I get some extra speed here?

Any particular reason why you what raw clockspeed instead of actual computing power?

The P4 only managed to clock so high by being having high inter-instruction and pipeline refill latency, and relatively poor instruction-level parallelism.

[Mr Bean]

Any particular reason why you what raw clockspeed instead of actual computing power?

The P4 only managed to clock so high by being having high inter-instruction and pipeline refill latency, and relatively poor instruction-level parallelism.

Because there's only so much improvement that can be done. That and because of the strong belief that Intel is sitting on the extra clock-speed. It make sense from an economic standpoint. (What would it be like today if they were releasing 4.2 Gthz Nehalem which pound for pound already kicked AMD's arse? AMD would have been put away) but not so much from a consumer standpoint.

What I see is, and the feeling I'm getting is that Intel keeps AMD around least some actual competition come from somewhere. I feel like Intel is carefully pacing itself, rather and push hard ahead as they were doing when AMD was an actual contendor and there was a race to be won.

But seriously you can throw on 64 MB's of on-die Level 3 cache, make it .32 micron sized and have twenty four cores. But there's only so far you can push before you have to start pushing up the clockspeed as well. And that's what I hate to think we are going to be sitting here two years from now doing. Because twenty four cores do not equal 72 Ghtz, no more than four cores equals 12.8 Ghtz. Tom's Hardware ran that test last year demonstrating for what I use computers for(Gaming, downloading, not encounding movies or ripping thousands of songs at once) what affects my needs most is clock speed. My GPU not being fed fast enough and in fact preforming going UP when all but one core is turned off since current cores still don't hand jobs out equally between cores.

[phongn]

If this had been another Penyrn style jump AMD would be instantly out of the game - as it is they have a chance, albeit a slim one.

I'm not sure how well Shanghai can do to get AMD back in the game, especially as Intel can put significant price pressure on AMD. There's the further problem that there's no tight coupling between design and fabricator for AMD anymore.

The server chips should be more exciting. Finally I really like the fact that latency on the XCHG instruction has nearly halved; my own code makes very heavy use of that.

Well, Intel has so many transistors and now they have tons of memory bandwidth that it looks like they're improving all sorts of things where they can.

Rendering and encoding are the two most CPU intensive areas though, and it's where Core i7 has colossal advantages over Phenom. I can't see AMD putting out enough of an improvement in those areas, and others are likely to still be bottlenecked by I/O and GPU performance.

It's not so much that rendering and encoding are so CPU intensive (though they certainly are) but the fact that they are trivially parallelizable problems. There are lot of very CPU-intensive problems that are strongly serial in nature.

For the love of Xenu yes I love the improvement but motherfuckers AMD and Intel I said I would not upgrade until there was a sub 400$ 4 Ghtz out there and your making it damn hard. Yes I know the race to 3 Ghtz was by putting clock speed over efficiency, but for bob sake we've been at or around 3Ghtz for over three years now. Can I get some extra speed here?

Not anytime soon.

Because there's only so much improvement that can be done. That and because of the strong belief that Intel is sitting on the extra clock-speed. It make sense from an economic standpoint. (What would it be like today if they were releasing 4.2 Gthz Nehalem which pound for pound already kicked AMD's arse? AMD would have been put away) but not so much from a consumer standpoint.

The average consumer has more CPU power than he knows what to do with. It's only a relatively small subset of the market that really cares about the bleeding edge of performance. I also doubt they're sitting on clock scaling.

But seriously you can throw on 64 MB's of on-die Level 3 cache, make it .32 micron sized and have twenty four cores. But there's only so far you can push before you have to start pushing up the clockspeed as well. And that's what I hate to think we are going to be sitting here two years from now doing. Because twenty four cores do not equal 72 Ghtz, no more than four cores equals 12.8 Ghtz. Tom's Hardware ran that test last year demonstrating for what I use computers for(Gaming, downloading, not encounding movies or ripping thousands of songs at once) what affects my needs most is clock speed.

The CPU industry has, by and large, run into enormous problems trying to get clockspeed up higher and higher. We're not just going to be sitting here for a couple more years with only small increases in clockspeed per generation (if any increase at all), we're going to be sitting here for a good while. The only processor commercially clocking past 4GHz I can think of is POWER6, and it had to strip out a lot of goodies to do it.

My GPU not being fed fast enough and in fact preforming going UP when all but one core is turned off since current cores still don't hand jobs out equally between cores.

That's an operating system (and possibly game engine design) issue, not the processor itself.

[Fingolfin_Noldor]

Intel pretty much declared they would push the envelope as far as they could. Technology overmatch, in the sense. AMD just doesn't have the engineering resources to compete. They will have to play second fiddle, or work doubly hard to compete against Intel.

[Seggybop]

For the love of Xenu yes I love the improvement but motherfuckers AMD and Intel I said I would not upgrade until there was a sub 400$ 4 Ghtz out there and your making it damn hard. Yes I know the race to 3 Ghtz was by putting clock speed over efficiency, but for bob sake we've been at or around 3Ghtz for over three years now. Can I get some extra speed here?

Do you have anything against overclocking? It's true that many recent Intel CPUs can operate at > 4ghz and Intel could release some at that speed if they desired. If that's all you want, then instead of waiting for them to give you an opportunity to feed them more money, you could get an E7200 or E8400, change your BIOS settings, and be happy.

[Stark]

He decided not to buy a chip until <irrelevant performance measure> reached <arbitrary integer value> below <random price point>. It's not like C2Ds don't kick the shit out of the stupid 3ghz+ P4s that used to fly around and are ludicrously cheap or anything.

[Mr Bean]

He decided not to buy a chip until <irrelevant performance measure> reached <arbitrary integer value> below <random price point>. It's not like C2Ds don't kick the shit out of the stupid 3ghz+ P4s that used to fly around and are ludicrously cheap or anything.

More Clock speed on the same processor means more FPS, actual accurate facts do not apply! Thus have I been taught.

Also more to the point as Seggybop mentioned OC's regularly take chips past 4Ghtz. For fuck sake Nehalem has built in over clocking as a FEATURE. They are low clocking these things in order to give themselves profit room.

[phongn]

More Clock speed on the same processor means more FPS, actual accurate facts do not apply! Thus have I been taught.

Yes, but the drive to a higher-clocked processor might also require other side effects that can reduce performance (taken in isolation).

Also more to the point as Seggybop mentioned OC's regularly take chips past 4Ghtz.

Which is completely irrelevant. Overclocking to that level generally requires expensive cooling systems and running far in excess of what Intel would probably feel comfortable selling. The increased voltage requirements on the processor may have negative long-term effects on the CPU as well.

For fuck sake Nehalem has built in over clocking as a FEATURE. They are low clocking these things in order to give themselves profit room.

Yeah, highly conservative overclocking. Why do you actually prove your assertion that they're deliberately low-balling things?

[Uraniun235]

Aren't the fastest Core 2 chips about as fast as a 4GHz Prescott anyway?

[phongn]

Aren't the fastest Core 2 chips about as fast as a 4GHz Prescott anyway?

Pretty much.

[Fingolfin_Noldor]

Bean, current processor design is about achieving a balance between heat output, power consumption and processing power. Too much clock speed either requires extremely good chip samples, or extremely good cooling systems. It's a toss up.

[Xon]

Also more to the point as Seggybop mentioned OC's regularly take chips past 4Ghtz. For fuck sake Nehalem has built in over clocking as a FEATURE. They are low clocking these things in order to give themselves profit room.

It is more accurate to describe it as aggressive underclocking when not in use and only running at peak rated clock rate under high load.

[DaveJB]

You're thinking of the Core 2. Core i7 can boost itself ether one full speed grade or half a grade under certain conditions, depending on the load, temperature and how many processor-intensive threads are being handled.

[The Kernel]

What's really striking is to look at the comparison clock for clock between the Core i7 and the Phenom. At video encoding tasks for example, the i7 920 is nearly TWICE AS FAST as the equivalently clocked Phenom X4 9950. That's just an insane level of difference.

I think it's clear that as applications become even more heavily threaded that Nehalem is going to really come into its own. It's an extremely forward looking design and its also amazingly power efficient compared to Penryn (which was one of the most power efficient chips ever built).

What AMD should really be worried about though is not Nehalem on the desktop, but the Beckton-EX server chips. Expanding Nehalem to 8-cores and turning it loose in the server space is going to absolutely slaughter all of the Opteron chips.

[Darth Wong]

The writing has been on the wall for AMD for some time; they're sliding back into their original position as a low-performance budget chipmaker. They had a nice run where they were actually considered competitive, but people like me still remember the days when no one took them seriously, and it looks like those days are going to return, if they have not done so already.

[The Kernel]

The writing has been on the wall for AMD for some time; they're sliding back into their original position as a low-performance budget chipmaker. They had a nice run where they were actually considered competitive, but people like me still remember the days when no one took them seriously, and it looks like those days are going to return, if they have not done so already.

What's really surprising about this is that there seems to be a lot of expectation in the enthusiast community that AMDs new "Deneb" line of processors is going to miraculously compete with Nehalem, despite the fact that all indications are that it is just a die shrink with some of the usual optimizations thrown in for good measure.

AMD has spent the last 2.5 years trying to find a competitor to the Core 2 chips and they STILL haven't managed to do it. Their latest top-of-the-line Phenom X4 9950 is still slower than the 65nm Core 2 Q6600, which is the oldest and lowest end quad core in Intel's stable. It's even worse when you look at dual cores where AMD is getting slapped silly by the hyperfast and cool running Penryn dualies.

Unfortunately, I think things are only going to get worse for AMD at this point. Sure there is the possibility of them pulling a rabbit out of their hat, but at some point having 1/6th the R&D budget as your competitor is going to catch up with you.

[Darth Wong]

I suppose they could hope for a colossal Intel fuckup again. Let's be brutally honest: they should never have been competitive. The only reason they became competitive in the first place was a series of Intel missteps: unreasonably high pricing, new CPUs that were only marginally better than their predecessors, a spectacularly failed and buggy motherboard chipset which shook consumer confidence in the brand, and all of this happening at the same time.