Intel Core 2 Extreme QX6700 Quad-Core Benchmarks

Slimpickin writes "Intel gave access to quad-core Kentsfield-based systems to select members of the press at IDF. The embargo has been lifted on a preview of performance numbers with the new 2.66GHz Core 2 Extreme QX6700 processor. HotHardware showcases Intel quad-core performance from a few different angles, from digital video processing and encoding, to 3D modeling and rendering, along with a few of the more standard benchmarks. the new Intel quad-core puts up performance numbers, depending on the application, at nearly double the performance of a 2.93GHz Core 2 Duo processor based system. Core 2 Quad will also drop right into existing motherboards that are compatible with the Core 2 processor line."

Comments

Already tested: Two Quad-Cores in a Mac Pro, makin

[chriss]

A few weeks ago Anandtech already tried to plug two 2.4 GHz Quad-Core Clovertons (Xeons) samples into the new Mac Pro featuring two LGA-771 sockets. Worked like a charm, a nice eight core machine. And since dual socket motherboards are quite expensive, the Mac Pro might even be a cheap version.

Well, All I have to say is...

[HuckleCom]

Quake 3 Voice: QUAD DAMAGE!

Intel FSB vs. AMD Hypertransport?

[ricky-road-flats]

If the Intel Core 2 Duos are good (generally agreed I think) , and this chip - their first quad-core - is looking so good, at what point will the apparent advantage of the AMD platform (no FSB, just Hypertransport links) kick in? If not at four cores, at eight? 16? More to the point, before Intel gets round to releasing CSI?

I know on the face of it this chip is a kludge (two dual-cores connected to one FSB in a single-socket package, as opposed to AMD's forthcoming 'true' quad-core CPU), but if it performs well, so what?

Re:Intel FSB vs. AMD Hypertransport?

[jiushao]

It can equally well be argued that AMD's solution is a "kludge". Intel has four processors arranged in two pairs, within each pair the processors are connected by shared L2 cache, but the pairs are connected by the FSB. AMD on the other hand have all four processors communicating over HyperTransport links. Shared L2 is clearly better than HyperTransport links, and the HyperTransport links are better than Intel's current FSB.

The physical packaging simply doesn't tell much about the quality of the interconnect. Sure it is harder to make a truly great interconnect with separate packages, but looking directly at the interconnect tells the much more accurate story.

Either way, it is not an all that great suprise that the dual-FSB design of modern Intel platforms manages four cores decently, but yes, AMD probably still has a clear edge on 8 core systems.

Re:Intel FSB vs. AMD Hypertransport?

[Anonymous Coward]

Calling Hypertransport a "kludge" is a stretch. The advantage of Hypertransport with multi-core comes into play regarding access to main memory. All processors can access main memory simultaneously, with theoretical bandwidth actually scaling with the number of processors, rather than being divided among them. It really is better than the FSB, but AMD needs to get kicking with 65 nm and below.

Re:Intel FSB vs. AMD Hypertransport?

[jiushao]

Calling either solution a "kludge" is of course wrong. However, just running everything across HyperTransport is an obviously worse approach for core-to-core communication than shared L2. The trick about sharing cache though is that it stops making sense to talk about the cores "sharing access to main memory", since any memory fetches go into the shared cache. Plus that Intel isn't stupid, their current platform has two separate front-side buses, so there is quite a bit of bandwidth to work with.

On the other hand it is of course also true that the HyperTransport approach is perfectly symmetrical and scales to almost any number of processors, so AMD has a good hold on the 8 core and above market.

Re:Intel FSB vs. AMD Hypertransport?

[Jherek Carnelian]

However, just running everything across HyperTransport is an obviously worse approach for core-to-core communication than shared L2.

The real question is how important is core-to-core communication versus core-to-memory for "regular" workloads?

My gut says that for consumer-level workloads, memory is more important than inter-core communication because most consumer-level parallel processing is of the "embarassingly parallel" type - specifically codec processing - video, audio and "photoshop plugin" types.

My imagination may be lacking, but I can't think of any consumer-level compute-intensive workloads that are also fine-grained enough where inter-core communication is critical. Perhaps some of the distributed-processing/seti-at-home type jobs could make use of fine-grained parallelism within each "chunk" of processing, I don't know.

So, it is entirely possible that the downsides of shared-cache, like bus contention and false cache-line sharing, could be a hindrence in comparison to the sort-of "shared-nothing" approach that AMD has taken with their designs.

Re:Should have wait...

[EatHam]

Well, for sure now is the time to buy a quad core. I doubt there will be any new chips released for the next several years at least.

Time to refine operating systems...

[b0r1s]

For years, most operating systems have been designed for 2-4 processors, with some handling more, and others doing better with less (I'm sorry, FreeBSD fans, I use it myself, but let's be honest, SMP was horrible until 5-REL, and it still isn't up there with Linux and *ugh* Microsoft).

With 4 core out this year, and 80 cores out in 5 years, it's time to rethink multiprocessor operating systems. There needs to be a significant change in the locking and threading metaphors, because 4 and 8 way will be obsolete by this time next year.

Re:Time to refine operating systems...

[Jeff DeMaagd]

So why does an app or a library have to care how many CPUs or cores the PC has? Surely that's the job of the OS?

Yes and no. Programs can split the hard work across several threads and all of those threads will be managed by the task scheduler regardless of how many cores there are. The hard part is making an algorithm that can split the heavy processing work to multiple threads, that threading has to be programmed. If the program has all the hard work in one thread, then it's not going to use more than 100% of one CPU, 50% each of two CPUs, etc.

Re:Time to refine operating systems...

[Elladan]

To elaborate:

Usually, if an application can split its work up into 2 threads, it can split its work up onto n threads (if it's well designed). This isn't
always the case, but it tends to be. The hard part is breaking an algorithm up into pieces, usually not the number of pieces in particular.

So, for example, it could spread its work over 100 threads on a 1-CPU machine. This would be an inefficient use of threads, if they're all doing work. Usually 1 or 2 threads is ideal for a 1-core machine.

Similarly, it could spread over 2 threads on a 4-core machine. This would also be inefficient, since 2 of the cores would be idle.

So, ideally, the program usually should have some way of asking the OS, "how many threads should I run for best efficiency?" In general, the question it really asks is, "how many execution cores are there?" and then it just uses that, or multiplies by 2, or something.

So, the application needs to know how many cores there are as part of its auto-configuration, for performance. Often apps don't auto-configure this sensibly, and just spawn a fixed number. This isn't ideal, since it requires the user to twiddle something for no particular reason.

Before the naysayers come out

[Kelz]

I seem to remember a particular article in which everyone seemed to decry the chip before it came out, citing "wait for the independant reviews". True, do that, but this time maybe people will think of Intel with a little more credibility? But then again, this is slashdot!

Re:Should have wait...

[gbulmash]

Why did everyone told me to rush on Core2 Duo when it got released saying it's the perfect time buying CPU..... now quad core get released a few months after.

I'm reminded of a cartoon I saw years back, where a computer salesman is showing a customer a selection of computers: "Here we have the ones that will be obsolete in 6 months, and over here are the ones that will be obsolete in 9 months."

Thing is, that though Intel is releasing a consumer grade quad soon, they're only releasing the "Extreme" version, which will be their highest-priced CPU. You won't see reasonably priced quads for 4-6 months after the Extreme version hits the supply chain. Unless you're doing hardcore 3d or video, a moderately-priced Core 2 Duo *should* be good enough to handle most of your tasks for the next 2-3 years before you start feeling the irresistable upgrade itch.

- Greg

Re:How about quad memory capacity?

[Travoltus]

That's what I meant by average... I know cpu's don't "demand" memory, but having enough memory for each CPU is a good idea.

The apps I run at home (video conversion, maybe a VMWare instance) would each use very close to 512mb apiece. I might even run Oblivion in one cpu while turning a DVD converter loose on another process; AFAIK Oblivion will grab whatever it can, so 1gb for that cpu isn't unfeasible.

I can imagine other, more memory intensive apps trying to run in tandem and running into problems if you have under 2gb for four cores, if you're using all four cores to the max.

Re:ExtremeTech has more benchmarks

[mikemuch]

Well that linking format didnt' work: http://www.extremetech.com/article2/0,1697,2021888 ,00.asp

Quad Core Gaming

[Psiven]

Currently the only game in the near future that will take advantage of multithreading is Crysis, shortly followed thereafter by HL2:EP2. In the case of Crysis, lead designer Cevat Yerli is quoted as saying that they are "scaling the individual modules, like animation, physics and parts of the graphics with the cpu, depending how many threads the hardware has to offer" (incrysis.com). But he has also stated that the game will get a 10-15% boost per thread in a 64bit environment compared to 32bit. If this is true, then what are the implications on performance when operated in AMD's upcoming 4x4 processer?

The AMD 4x4 is pure 64bit, so does this mean that when compared to Kentsfield, a quad 32bit processor, Crysis would behave 40-60% (4x10% or 4x15%) faster?

Re:Quad Core Gaming

[neilyos]

FYI, Kentsfield has 64 bit instructions, and therefore is 64 bit capable. Please stop spouting this "AMD is the only pure 64 bit processor" garbage.

My favorite application will

[benhocking]

If I add more cores, it's so I can simulate bigger neural networks in close to real time. Depending on my level-of-detail, 10,000 neurons take anywhere from 250-750 MB. In order to run 100,000 neurons I need dozens of CPUs (currently done via Beowulf clusters), and each CPU needs a lot of RAM. (Up to about 100k neurons, synapses scale roughly as n^2, not n - at least in our model of the hippocampus.)

Names

[jeffy210]

2.66GHz Core 2 Extreme QX6700??

So now, now only have they gone back to pointing out the clock speed, they add the NVidia product name at the end? Surely there's got to be a simpler way to do this, without even taking into account AMD. I mean you have:

- Dual Processor Pentium
- Dual Core Pentium D
- Core 2 Solo
- Core 2 Duo
- Core 2 Quad
- Dual Processor Core 2 Quad

Seriously, that's some major word jumble and you haven't even specified anything like clock speed (I know it's not all about clock speed, but uniform naming to differentiate would help).

Re:Sure ..

[The Warlock]

If an employee isn't smart enough to manage the differences between MS Word and OOo Writer by himself, show him the fucking door, because he's clearly so goddamn stupid he'll generate problems in other areas. Seriously.

Re:How about quad memory capacity?

[Doctor Memory]

Oblivion will grab whatever it can, so 1gb for that cpu isn't unfeasible

Do you know that your Oblivion process will stay on whatever core it starts on? If it's any kind of a useful application, it's going to have to do I/O sometime, and at that point will give up the CPU. Once the I/O completes, it'll be placed on the run queue and should be picked up by the next available processor. It shouldn't matter how many processors you have, any program that takes a given amount of memory to do a given amount of work will take up that much memory. If you're planning on running more programs to take advantage of your additional CPU power then yes, you'll want more memory. But if all you want to do is execute your current software and workloads faster (assuming they run faster on a multicore system), then you won't need any more memory.

I can imagine other, more memory intensive apps trying to run in tandem

You've got a better imagination than I do, then. I can't see applications forking off copies of themselves and jockeying for position! If you meant "I can see running other, more memory-intensive apps in tandem" then duh, you'll use more memory. Exactly the same as you would on a single-core system. If you've got an app that scales well, it'll still take the same amount of memory no matter how hard it's exercising however many CPUs. Input set sizes are pretty much fixed, whether they're hard-coded or dynamically configured based on system size: your app will allocate the same amount of memory either because it always grabs 32M or because it always grabs 1/16th of total system memory. Number of CPUs has nothing to do with it. Unless there's some software that allocates one thread per CPU, and allocates some fixed buffer size per thread, which now that I mention it actually sounds reasonable for some classes of software, but I've never heard of it actually being done.

Did Intel learn *anything* from Java2?

[Kostya]

What the hell is with this Core2 Quad crap? It should be Core2 and Core4. You would have thought Intel would have learned from the nightmare Sun/Java went through with the whole "Java2 1.4" branding nightmare. Sun finally wized up and started calling everything Java 4, Java 5, and Java 6. Why would Intel start such a fiasco?

I get that they are trying to say "Hey look, it is a totally different architecture!" But calling it Core2 isn't going to do that. People will just end up calling them Dual Core or Quad Core anyways, not Dual Core2 and Quad Core2. It's just going to detract from their branding, not help it.

Re:Did Intel learn *anything* from Java2?

[nuzak]

> What the hell is with this Core2 Quad crap? It should be Core2 and Core4.

Core 2 is the second iteration of the "Core" line. There is Core 2 Solo and Core 2 Duo. It's the new "Pentium", stick with a single brand and append numbers to it. It doesn't help that there's *also* pentium-branded chips still being made.

I agree though, it's still a mess. I'm pretty experienced, and I get confused by it. Quick, which is newer, a Foofra QXV5024351GL or a Wibble RG188716912B?

Re:Already tested: Two Quad-Cores in a Mac Pro, ma

[nuzak]

You don't void your warranty for upgrading a Mac or any other computer. Your "friends" are wrong about this, at least in the US, because it would violate the Magnuson-Moss Warranty Act.

My interpretation of Magnuson-Moss is that it prohibits bundling, like say, Apple requiring you purchase Apple-branded CPU's to upgrade. Pulling out your own CPU is probably still a warranty killer. They just can't automatically call it void if the problem is obviously unrelated and a defect in the merchandise, like oh, the paint starts yellowing after you open it up (that's about the only thing I can think of that's TOTALLY implausible to connect to a cause of opening up the case).

However, I bet you anything Apple's warranty says Limited Warranty. At that point they're largely off the hook. State laws may still apply.

It's not Core 2. It's Core 2 Duo

[WoTG]

But yeah, it's got to be one of the stupidest naming schemes from a major vendor of anything that I've seen in a long time. I was looking forward to a Pentium V. Sure it would have seemed a bit redundant, but so is the Core 2 Duo chip that's in the new laptop I've been eyeing...