Intel I7 processors
 

Hey Guys,

Been doing a bit of reading up on the new intel i7 cores and the impact they have for overall processing power.

Although the technlogy isnt strictly new, personally I think that they will have a pretty big impact on the overall rendering times and performance of PC's.

Looking at the review in 3d world of one of the speced up machines, it was only 25% slower than a dual Xenon system, and it trounced any other sinlge processor.

When they get more mainstream, which Dell are on their way to making by giving a PC with i7, I feel that they will be the standard in a few months, and become cheaper, and more affordable.

What do you guys think?

The new standard

Or

Nah it wont take off

OK, lets see what is so great about the i7...

Intel replaced the dated FSB with a few unidirectional point to point connections (QuickPath)... Good for bandwidth, except that it is nothing new... A good example is HyperTransport used by AMD. The basic concept was around 10-15 years ago, perhaps even before that (though I can't think of examples).

Intel reintroduced HyperThreading, which can make the CPU more efficient and execute more instructions per a clock cycle... But HT is meh, Intel tried it in their Pentium 4s and then chose to do without it in their Core CPUs for some reason (probably it was not worth the effort considering the performance gain).

HT (generically SMT) is nothing new, again, even in the days of the P4. IBM got to SMT in 2001, I think, and the concept has been around even before that (DEC/Compaq did work on it back in 1998).

HT itself is a bit overhyped anyway. Does anyone remember the hype with HT when it was in the P4? All those Intel ads about how great Intel was, that they, through sorcery that us mere mortals couldn't comprehend managed to get a whopping two "virtual cores" on one die! Except performance wasn't that good... 10-30% improvement is not the same as what two cores will get...

The i7 has an on-die memory controller which supports six DIMM slots. OK, so the CPU can access the memory faster for better performance, but the idea is old. Cheapish CPUs for industrial use have had on-die MCs for around 20+ years, quite a few CPUs had on-die MCs before AMD got around to using them, which is one again, before Intel.

One particularly bad thing about the memory controller in the i7 is that it doesn't have ECC support, which is important in computers when RAM is plentiful as it helps stop the RAM from becoming corrupted. Likely hood of corruption increases when you have more RAM.

The i7's L2 cache is of concern, it is not even an improvement over a previous Intel design. 8 MB of cache shared between four cores and eight threads ("virtual cores" by HT) is a bit scary. Hyperthreading doubles the memory bandwidth and cache capacity requirements. Sharing one L2 cache between four cores, eight threads is not good for both.

Knowing Intel, they probably went with "lower cost, easy design" when they chose to use a smaller shared cache, which is good for the customer, but then, it restricts HT's potential, which means you are paying more for something which doesn't work as well as it could, so you could pay less for something else (if that makes sense).

So, yes, the i7 is fast, and I would like to have one, but the hype isn't justified.

Of course, I haven't really been paying too much attention to the i7 (Intel will probably replace it next year with another model, and the hype will start all over again), so if I got something wrong in my half-rant half-amateur analysis, do tell me! :p

Agreed, the idea of point to point connections is well established. Although dated now, SGI used a similar method of interface with their Xtalk architecture...Hence the reason most SGI systems had such large bandwidth paths for moving SD/HD video between the IO-CPU-VIDEO components in the 90's. Hell, what x86 system now can handle uncompressed HD with a dual 600MHZ processor and 8 year old video card?! But sadly, things move on...! (or do they?) ;)

Ah, yes, the old SGIs... how could I have forgotten? :blush:

lol!


To make the thread a bit more interesting, name a "new" feature of the i7 and we'll try to think of previous or better example... :)

I see a lot of "ohh its been done before", hence its a bit naf.

Most of everything has been before, in some way or another, so I think its a bit unfair to bundle crit in that way, yes its not "totaly new" but geting the performace out of it is an acheivement.

If it was so easy to do and obvious to everyone (which it seems to be), then why didnt say... AMD make the jump? Theres always inherent problems trying to intergrate an new, or old, idea into an exisitng design.

As I said its not new, but the performance out of a single CPU compared to current technology is a pretty big jump, over going up the route of more cores, which have there downfall.

Edit, as for being replaced next year, then well, yeah of course, thats progress and Moore's law. It would be a tad stupid not to...

From what I have seen, a top of the line i7 is only 37% faster than a top of the line Core 2 Quad. That isn't really better than what Intel has been doing (Pentium 4 execpted of course ;)).

Could you clarify, "it isn't really better than what intel has been doing?"

Only 37% faster, well I hope that any improvement in say maya is "only" 37%

Well, hasn't every new Intel processor in recent memory only improved in performance around 20 to 30% in general? The i7 doesn't really break this trend. It might in some benchmarks and tasks, but its not the "amazing wonder chip" some magazines are saying it is.

I for one await Intel's chips due in 2010 with AVX, which will bring 256-bit SIMD. That should give a much larger performance boost and be something of an "amazing wonder chip", although previous designs years ago have either matched or exceeded 256 bits by a large amount.

In your mind wouldent the AVX not fall into the catagory of

I also like the fact that you've again said that "its been done before", which kinda makes me think, if its so obvious to do then intel or AMD are a bit slow on the take up to incorporate it, then get you CV into them, you would be worth a fortune, maybe you could design the "Amazing wonder chips" with completly new technology thats never been done before???

Would AVX fall into the category of the "cycle of hype"?

Not really, it is worth some of the hype. 256-bit SIMD, expandable to at least 512-bit is a big step forward... not like their bolted on 64/128-bit SSE in the Pentium 3.

Design amazing wonder chips? Me? Nah, I only comment on amazing wonder chips...

Intel has always been a bit slow in adopting technology, and I think quite a few times they got grilled in the press for that...

Does anyone remember their, "We need no 64-bit in our Pentium 4s!" stance they took in 2003, only for them to do a backflip and add sorta-AMD64-compatible instructions to their P4s a while later? They even waved around slides of their roadmap showing their vision of then-future 32-bit P4s. :rolleyes:

As we now know, there wasn't really a future in the P4, or in 32-bit chips...

Amazing wonder chips to look out for in the next few years?

Sun's UltraSPARC RK - it has out of order completetion, never done before commercially, AFAIK. Though, Sun might sink before it gets released...

IBM's POWER7 - Projected to have 16 cores per module, 8 cores per chip, 32 GFLOPS per core, 256 GFLOPS in total along with the typical huge amounts of bandwidth.