Intel 3.40EE & 3.60E - LGA Arrives

MBR writes "MBReview has taken a quick look at Intel's new high-end LGA775 processors, the 3.40GHz Extreme Edition, and the 3.60GHz 'E,' now known as the 560. They've covered some of the questions about pin frailty of the new LGA socket, as well as cooling issues that might arise from these new processors." ("LGA" stands for Land Grid Array, which moves pins from the processor to the socket it sits in.) Update: 06/19 20:50 GMT by T : Reader Chi-Energy points out that besides the new processor packaging, Intel has also just released its i925X and i915 chipsets, PCI Express and DDR2 DRAM for the desktop, and links to this review showcase with benchmarks at HotHardware.

Comments

New pins

[Deltawolf]

Oh wow! Now if your pins snap you have to replace your mobo instead of your processor. Sounds like its begging for trouble.

Re:New pins

[Naffer]

Most home motherboards are cheaper then the processor. Motherboards run around $150 while newer processers run above $200

Re:New pins

[Anonymous Coward]

I'd rather replace a CPU with a broken pin than tear apart my case and pull out the tray to replace a motherboard.

A CPU can be replaced in just a couple of minutes. A motherboard would take much longer, depending on your case type, how many cards you have, and all the various types of things you're going to have to unplug from it and plug back in.

Re:New pins

[JPriest]

A few points: $150 is reasonable for a motherboard, the above listed processors are likely to run closer to $450 and $900.

If 6 minutes of your time is worth $300 - $750 then you obviously make way more than I do.

AMD is going to start using the same technology. When Intel does it, it is a pain in the ass, when AMD does it, it's innovation.

Besides, you have to be pretty careless with your hardware to break a pin.

Re:New pins

[Too Much Noise]

There are some points you're missing about this.

Who's paying for the RMA? if the natural life span of the pins is about 8 insertions (as the mobo producers seem to claim), then there would be a large number of legitimate breakages that get sent back to the mobo manufacturer. Now, they can either replace the CPU socket (not very funny, I think) or throw away the whole mobo, including the rest of the perfectly good components on it[*]. As oppose do just discarding a defective CPU if its pins break.

[*]like the spankin' new and expensive Intel chipsets. I doubt $150 will happen anytime soon as a mobo price, as even the chipset estimated price seems to be above that. I also doubt mobo manufacturers getting too many returns due to bent socket pins will be very happy about all this - remember, their margins are quite slim these days. The least hurt by this is probably going to be Intel itself.

Your AMD jab is a troll. As far as they stated so far, the Opteron socket stays put for the foreseeable future (meaning at least one year). They will have no incentive to move to a pinless package unless it shows some solid advantage. Even Intel might have to back down on this if the hw producers get to unhappy (and they already have enough grief with the BTX form factor).

Finally - pins break. It's called mechanical stress. How many times do you think you can 'carefully' insert and remove a CPU in its socket before some pin gives in? At least, for the old sockets, all you had to do is match pins and holes ; now, with only point contacts, bending can come so much easier.

What should we be concerned about?

[irokitt]

I'm not as worried about frail pins as I am about the amount of heat these things push out, the size of the new heatsink/fan assemblies, and the noise they put out. I thought Socket 478 processors were hot, but LGA Prescott processors run even hotter, which makes me think Intel has a point when it says we should switch away from the ATX case factor and adopt BTX for Intel chips..

That said, are the Extreme Edition processors still selling for $900 USD a pop? Hardly seems worth the extra money for gaming, although a server that wants to survive Slashdottings could probably use one...

BTX (which includes LGA) standard resisted

[Synkronos]

Apparently the BTX form factor (of which LGA is a part) has been heavily resisted by many Taiwanese chassis, mainboard and heatsink manufacturers.

But what's new here? Word has it that this time round, the Taiwanese heastink, mainboard and PSU manufacturers - and quite a lot of them it would seem - are being rather less than enthusiastic or co-operative, about the sweeping changes and support that Intel is asking, nay demanding, of them.

I'd be interested to see if Intel can actually strong-arm them into it

Wow, look at the length of the 3.40 EE bar!

[Jeremy Erwin]

Oh, wait. It seems that none of those bar graphs include an origin. Never mind.

Re:Wow, look at the length of the 3.40 EE bar!

[ameoba]

Yeah..

nothing like a graph that makes a 10% difference look like a 90% gap.

Re:Is not

[NanoGator]

"These chip make futiliity. Why make processors of like these new when you can improve on 64 bit? The battle is to will be lost to Athlon without 64 bit competition by."

Are you running 64-bit apps I'm not aware of?

Re:Pictures of the socket

[Aphrika]

here's a couple of pictures.

For a more comprehensive overview of the whole BTX, DDR2, Socket 775 and PCI Express malarkey, I'd recommend having a look here. Interesting stuff.

Re:Is not

[bobthemonkey13]

Not only does AMD have the only desktop 64-bit offering right now, but their chips are much faster than Intel's at the same clockspeed, even in 32-bit mode. Whereas Intel's engineers are just running their chips at insane clockspeeds, AMD's are actually designing better processors. For the price of a 3.4GHz "800"MHz FSB P4EE ($989 on pricewatch right now), you could buy two Opteron 246s ($441 each) with cash to spare. If you want to talk raw, meaningless numbers, the Opterons still beat the P4EE (4GHz and 2MB cache total). Of course, SMP isn't simply additive like that, but consider the advantages of 64-bit and multiprocessing, and the fact that AMD chips are /much/ faster than Intel's at the same clockspeed (even on 32-bit code), and there's no contest. All halfway-modern Windows versions and Linux kernels can support SMP, and the latest support amd64, too.

About time for them to use LGA...

[DraconPern]

It's about time they catched up and started to use the LGA connection! The NEC VR10000 MIPS chip had LGA in 1998, as well as LGA contacts on the motherboard. To connect the LGA on the proc against the LGA on the board, a plastic holder with wads of springy gold wires was used. There was no issues with bent pins, etc. The only problem was lossing those wads of gold...

another crap review

[phrasebook]

I'm sick of reading reviews that compare new products with other new products. Example on MBReview: comparing P4s that are all pretty much brand new, all expensive, hardly any difference between them. I want to see how it stacks up against my P3-866, not another P4 that I've never even seen. At least throw an older proc in there for comparison. Same with video card reviews. I don't give a hoot how the Radeon 9600 compares with the 9500... how does it compare with my GF3? FFS these reviews suck. At least throw in an older chip just for a relevant comparison. And stop mentioning how Quake 3 is getting old but is still useful: "this benchmark is slowly progressing towards an archaic stage". STFU. Who keeps regurgitating this crap.

Best for use only in winter

[Sivar]

According to Sandpile.org, the 3.4GHz Pentium IV Prescott can use up to 127W, and has a typical power usage of 103W (when browsing the web or reading email).
In my opinion, it is rediculous for a single processor to single-handedly run up your power bill. That's like having two light bulbs on 24/7 (assuming you keep your computer on), not to mention the power needed to cool your PC, let alone your house's air conditioner.

I would take a VIA chip for low-performance stuff, and an Athlon64 for performance computing. support 64-bit software including 64-bit Linux distributions, are faster than Intel's best even running 32-bit software, and they have a maximum power usage of 89W. Because of Cool'n'Quiet mode, they spend most of the time running at 800MHz consuming about 30-35W and generally not requiring a loud and abnoxious cooling fan.

It is actually impressive what the chips can do at 800MHz. You can play a full screen DVD at 1400x1050, and the CPU usage tops out at about 5% (at 800MHz). If, of course, you run something that requires more power, like a video game or a compiler, the processor instantly switches to full speed. Handy, that.

Re:Best for use only in winter

[Anonymous Coward]

That 103W figure of yours is bullshit - at least how you've understood it. That's the typical maximum power consumption, or what Intel call the "Thermal Design Power". While web browsing or reading email, power consumption is down around 30W. You need difficult-to-design test vectors to push the CPU beyond the TDP, and unless you know what you're doing, you're unlikely to get more than a couple of watts beyond it.

One way to prove this to yourself is to simply remove the heatsink from a running Pentium IV Prescott system. Shock, horror: it will continue to run, only slowly. That's the same slowly the system silently goes into and out of depending on the CPU load (it takes about 10000 CPU cycles, or about 3 millionths of a second, to get into or out of this state.) Now: try web browsing or reading email while it's running like this. Tell the difference? Didn't think so. You might notice a _slight_ slowdown (when rendering a complex page, for example): that's because without the heatsink on, the CPU won't go back into the "normal" S0 power state.

So, just as a Pentium 4 doesn't dissipate 103W without a heatsink installed, so to does it not dissipate 103W if you're not doing anything.

This can all be found in http://developer.intel.com
Handy, that.

Intel 925 chipset feature

[rumpledstiltskin]

One of the most useful things about the 925 chipset, IMHO, is the interesting possibilities it offers for SATA RAID. Say you want the performance capabilities of RAID 0, but at the same time, you need the redundancy for RAID 1. let's also say that you can only afford two SATA drives.
the intel 925 chipset has native support for a mixed raid, where you can create a raid 0 partiiton across two hard drives, using only part of the hdd capacity on each drive for the raid 0 partition. the rest of the unpartitioned space can be set aside as a raid 1 partition. that way you can install the OS and other non-critical files tha can be lost to the raid 0 partition and get the performance, but if one of the drives fail, you can store your important stuff on the raid 1 partition. I'm trying to find a controller card that will do this functionality, but I can't find anyone that claims to explicitly support it. the only reason I know about the 925 features is I got a chance to play with a pre-production board. definitely a cool feature.

Re:Is not

[Alexis de Torquemada]

A higher clockrate is ALWAYS better from a performance standpoint. ALWAYS. ALWAYS. ALWAYS. If you know anything about synchronous logic design you would know there is no debate about this.

True, provided that you are comparing processors with identical design that only differ in clockrate. But of course this is by far not the case, the P4 and Athlon 64 are implemented in fundamentally different ways. For example, in order to achieve the high clockrates with which they want to market their products to the uninformed (obviously), the Intel guys have increased the pipeline length beyond good and bad, with the consequence that mispredictions for out-of-order execution cost some real time. HyperThreading was introduced as sort of a hack for reducing the negative effect of their long pipelines, at least for multi-threaded applications. Running only a single thread, the P4 just has trouble keeping its functional units busy.

The speed of a processor is not measured in GHz. It's measured by the amount of work it gets done in one second. This depends on the application, but it's no secret that AMD CPUs perform substantially more work per processor cycle than Intel CPUs. E.g. my Ahlon XP 2400+ operates at "only" 2GHz. However, I took a the results from comparative Benchmark tests from the German computer magazine c't, and averaged (over all tests) the clockspeed that a Pentium 4 would need in order to be as fast as the Athlon. The result was 2800MHz, so the Athlon XP is on the average 40% faster than a Pentium 4 operating at the same speed. In other words, clockrate isn't everything.

The main problem with your analysis is that there exist algorithms that mathmatically CANNOT be solved in parallel, making SMP, hyperthreading, clusters all useless.

Actually, that's a good argument against Intel's hyperthreading, though there's a problem with it anyway: In practice, the question is not "Is this problem serial or parallelizable?", but how well it can be parallelized. For example, going from 1 to 8 CPUs may allow you to speed up computation of a certain problem by factor 7, however going from 128 to 512 CPUs may give you a speed increase of only 3%, because the communication and syncrhonization overhead becomes the bottleneck.

Oh, and 64-bit only buys you a larger memory space.

First of all, this "only" is misleading since even desktop machines will soon reach the 4GB boundary (actually, the 4GB limit virtual memory, which is often required in substantially larger quantities than physical RAM). You can use PAE for up to 64 gigs, but it's a performance killer.

And second, this is not true. AMD64 allows you to use wider adresses as well as wider integers, and this is a great boon for certain types of application, most notably cryptography. I've seen a benchmark that showed an 2GHz Athlon 64 outperform a P4EE 3.4GHz by factor two in AES encryption. Obviously, 64-bit integer operations benefit AES greatly. On 32-bit machines, they have to be split up into sub-operations - e.g. a 64-bit multiplication (discarding the upper 64 bits of the result) requires 3 32-bit multiplications plus several additions. For comparison, the Athlon 64 requires 3 clock cycles for a 32-bit multiplication, but only 4 for a 64-bit multiplication! Compare this to about 11 or more cycles the CPU would have spent on an equivalent sequence of 32-bit operations, which also would have increased code size (more cache misses) and forced you to use more of the already scarce registers (AMD64 doubles the size and number of the general purpose registers, some of which aren't even that general-purpose...).