Intel Core i5

[dmens]

Originally posted by: taltamir
4 seperate distributed computing apps running concurrently... 100% CPU utilization.

even then is the core functional units in use all the time? or are there little holes that a fine-grained SMT implementation can take advantage of? even the simplest add operation has to go through numerous buffers/queues that can get clogged, SMT partitions or duplicates those resources.

 

[taltamir]

true, it is unlike that MOST code uses the entirety of the CPU effectively. The architecture is just horribly inefficient, this is part of why we need to do away with x86 already.

 

[dmens]

Originally posted by: taltamir
true, it is unlike that MOST code uses the entirety of the CPU effectively. The architecture is just horribly inefficient, this is part of why we need to do away with x86 already.

what does this have to do with x86? latencies exist in all machines. if you cache (better do), you have misses. if you rename, the table can only hold so many aliases. out-of-order dispatch will put a bottleneck on a scheduler, as will out-of-order load/stores. these problems are not unique to x86.

modern x86 cores use proprietary microcode which bears little resemblance to x86. if x86 as an ISA is inefficient, you'd see the translation front-end as the major throughput bottleneck. but that is obviously not the case.

sure it may a little more energy per operation compared to some dedicated ISA's, but how many of those can be as flexible? for example, there were no from-scratch core redesigns to accommodate SIMD extensions, x86-64, or the upcoming AVX.

lastly, x86 is evolutionary. legacy features can and will be disabled with time to enhance future performance.

 

[Idontcare]

Originally posted by: dmens

Originally posted by: taltamir
true, it is unlike that MOST code uses the entirety of the CPU effectively. The architecture is just horribly inefficient, this is part of why we need to do away with x86 already.

what does this have to do with x86? latencies exist in all machines. if you cache (better do), you have misses. if you rename, the table can only hold so many aliases. out-of-order dispatch will put a bottleneck on a scheduler, as will out-of-order load/stores. these problems are not unique to x86.

modern x86 cores use proprietary microcode which bears little resemblance to x86. if x86 as an ISA is inefficient, you'd see the translation front-end as the major throughput bottleneck. but that is obviously not the case.

sure it may a little more energy per operation compared to some dedicated ISA's, but how many of those can be as flexible? for example, there were no from-scratch core redesigns to accommodate SIMD extensions, x86-64, or the upcoming AVX.

lastly, x86 is evolutionary. legacy features can and will be disabled with time to enhance future performance.

dmens I know you tire of it, at least you give me the impression of it thru the exasperation expressed in some of your posts, and I know for darn sure I tire of it, and it is about all we can really expect to encounter in a forum like this but beyond this facet of your life have you ever in your professional experience encountered an honest to goodness microprocessor design engineer who had more than 5yrs experience on the job and had anything truly bad to say about x86?

I understand the argument that other ISA's (epic) offer something different and better for their target crowds but I have not met anyone who (a) knew what they were talking about from firsthand experience when it came x86 ISA and programming, and (b) had a problem with it for the markets it serves.

Somehow, for some reason, there is this lore out there in the interwebs that x86 is a dreadful ISA and creating supporting architectures is just something that requires an act of god that nearly destroys the world. It is one of these urban legends that gets repeated and repeated by people who read an online article about "x86 overhead" somewhere and they've decided it must be true for reasons they have no ISA experience to validate.

But maybe I'm just off my rocker here, limited by my real-life exposure to actual programmers of desktop and server apps which rely on x86 compilers and hardware...so I ask you, ever met anyone with any credibility (to you) have a problem with "x86 overhead"?

 

[taltamir]

actually i was referring to the completely and utter failure at scalability that requires us to have more "cores" and write unique code to take advantage of multiple "cores".

As for scheduling that YOU but not ME were talking about... IA-64 actually did a LOT to improve the system in regards to scheduler issues, if only intel would develop it further... Just because its intel does not mean it is bad (in fact it is usually quite good)

In video cards execution units are split into SPs now that can be utilized massively in parallel more effectively.

There have been articles and dissertation written about "coreless" processors that bounce threads between two buffers using whatever available execution units are there. everything can use all the computational power without overhead, "cores", or cases where a large portion of the power is not used (for example, a recent CPU test shows dual core getting 85% gaming performance increase over single gaming, and both tri and quad core tie out at 99.8% increase over single core)

 

[dmens]

Originally posted by: Idontcare
dmens I know you tire of it, at least you give me the impression of it thru the exasperation expressed in some of your posts, and I know for darn sure I tire of it, and it is about all we can really expect to encounter in a forum like this but beyond this facet of your life have you ever in your professional experience encountered an honest to goodness microprocessor design engineer who had more than 5yrs experience on the job and had anything truly bad to say about x86?

yep we bitch all the time about x86 and its quirks because it is a pain. however, i have never heard any architect say the machine will be totally gated by "limitations" of the x86 ISA. i have never heard any designer say an x86 necessity will cripple the performance of the machine.

Somehow, for some reason, there is this lore out there in the interwebs that x86 is a dreadful ISA and creating supporting architectures is just something that requires an act of god that nearly destroys the world. It is one of these urban legends that gets repeated and repeated by people who read an online article about "x86 overhead" somewhere and they've decided it must be true for reasons they have no ISA experience to validate.

But maybe I'm just off my rocker here, limited by my real-life exposure to actual programmers of desktop and server apps which rely on x86 compilers and hardware...so I ask you, ever met anyone with any credibility (to you) have a problem with "x86 overhead"?

from my personal experience, i got a lot of that in the academic world. x86 was ignored entirely in my classes and both professors would snicker at its variable instruction length (ewww), lack of registers, whatever. x86 overheard exists. but AMD/Intel engineers have gotten really good at making it go fast regardless of the extra work.

one reason the tubes gives x86 such a hard time might be because they take the evolutionary ease of the system for granted and never have to worry about how hard it might be to evolve a more rigid definition.

 

[taltamir]

Also...
x86 is a closed and patented format that has a total monopolly of the market, all software is made just for it because it is everywhere. Noone can compete because all the software is for x86. It is a catch 22...
Although, I figured a fairly simple and not too expensive way to break that cycle, companies like to ignore the gaming crowd, like gaming is something dirty. They forget that the internet was built on pornography, and that PC hardware built on gaming. Games drive hardware development, not MS office (my einstein DOS word processor was good enough thank you). just avoid the mistake of making a "framework" or hardware... instead BUY a variety of studios and release their games for whatever you want to standardize... that is what microsoft did with the Xbox.

yep we bitch all the time about x86 and its quirks because it is a pain. however, i have never heard any architect say the machine will be totally gated by "limitations" of the x86 ISA. i have never heard any designer say an x86 necessity will cripple the performance of the machine.

You never heard of a multi core machine before? A huge contributer to VM popularity is running one OS copy per core to deal with the shortcoming of x86.

 

[taltamir]

would you stop with the stawman argument dmens? never once did i mention x86.. in fact I specifically named a variety of issues, none of which is "overhaed"

 

[dmens]

dude i wasn't talking to you, take it easy

 

[IntelUser2000]

Lynnfield isn't JUST Bloomfield with 1 channel of memory cut and PCI Express controller added in.

Remember power gating? Well in Bloomfield(the 9xx Core i7), the power gate transistor allows the PCU to turn off the cores when idle.

Lynnfield allows power gating in not only the cores, but the L3 cache and the I/O. Another reason that it can be much more power efficient than Bloomfield can.

 

[taltamir]

mmm... more advantages... nice.

 

[IntelUser2000]

Head to PCOnline for Core i7 870 "Lynnfield" review.

 

[taltamir]

pconline... the chinese website?

 

[Cookie Monster]

So when is the exact launch date?

I can go down to my local store and buy myself a nice i7 870/P55 motherboard like tomorrow if I wanted.

Another question is just how long would the P55 is going live, since intel is preparing the P57 for 2010.

 

[Axon]

Last I recall hearing, the launch is September 6.

 

[eternalone]

Yeah I would like to know an exact date also, I was hoping these would be out by now, so we could compare specs and prices could come down all around.

 

[Idontcare]

Originally posted by: IntelUser2000
Lynnfield isn't JUST Bloomfield with 1 channel of memory cut and PCI Express controller added in.

Remember power gating? Well in Bloomfield(the 9xx Core i7), the power gate transistor allows the PCU to turn off the cores when idle.

Lynnfield allows power gating in not only the cores, but the L3 cache and the I/O. Another reason that it can be much more power efficient than Bloomfield can.

AMD better get with the program and release their own version of (1) power-gating, and (2) dynamic clocking (turbo).

I don't see them having an entry point until bulldozer but 2010 is gonna be a tough year for them if their X4's are competing price/performance and performance/watt with nehalem based powergated and turbo'd lynnfields/clarkdales.

 

[taltamir]

AFAIK the first version of bulldozer isn't even gonna have integrated GPU+CPU...
While intel is slating those for this year! Intel is on a roll...

axon... sep 6 eh? A month and a day is not too bad. Although considering just how MANY previews of P55 boards I have SEEN the last few weeks i'd expect it to be sooner.

 

[Denithor]

And then count on at least two months for the price gouging to settle down before you can actually get one of these chips for a price anywhere near those shown in all the previews.

 

[Idontcare]

Originally posted by: taltamir
AFAIK the first version of bulldozer isn't even gonna have integrated GPU+CPU...
While intel is slating those for this year! Intel is on a roll...

axon... sep 6 eh? A month and a day is not too bad. Although considering just how MANY previews of P55 boards I have SEEN the last few weeks i'd expect it to be sooner.

Personally I'm not so much worried about fusion or clarkdale, to me these are cost enablers not performance enablers. It will/should lead to lower-cost mobile compute solutions, but I don't think its going to make anything use less power or be faster versus their chipset-based IGP counterparts would have been with the same process technology.

I'm more worried about AMD's staid approach to power management and single-threaded performance boosts by tapping into that unused TDP when one or two cores are only being used.

If BD even has these "tricks" built-in, by the time it debuts Intel will have had them on the market for over 2yrs by that point.

 

[IntelUser2000]

Originally posted by: Idontcare

Personally I'm not so much worried about fusion or clarkdale, to me these are cost enablers not performance enablers. It will/should lead to lower-cost mobile compute solutions, but I don't think its going to make anything use less power or be faster versus their chipset-based IGP counterparts would have been with the same process technology.

I'm more worried about AMD's staid approach to power management and single-threaded performance boosts by tapping into that unused TDP when one or two cores are only being used.

If BD even has these "tricks" built-in, by the time it debuts Intel will have had them on the market for over 2yrs by that point.

The PCU is a culmination of years of research by Intel that has been trickled down from the original "Foxton" project in Montecito. Though Montecito were severely late and defeatured(2.0GHz, with 2.2GHz Boost down to 1.6GHz, 100W TDP to 108W, Foxton disabled), whatever they learned with the advanced analog based temperature/current/power measurement circuit was implemented in Tukwila/Nehalem.

The digital power management microprocessor that's basically in the Nehalem called the PCU is done with no small feat. So they did suffer few years of failure before turning it radically around with the new versions.

In regards to Clarkdale, I think these will be the Intel IGP that people will actually take notice, believe it or not. You'd also want to take note on the current AMD IGP solutions, the memory bandwidth can be no higher than the bandwidth of the HT bus. Integrated GPU via Fusion won't have those problems.

(IMO, Bloomfields are the worst of the Nehalem generation. That is sort of understandable as features like Turbo Mode is set low since it won't matter to enthusiasts who overclock anyway. Lynnfield will be significantly better than Bloomfield and Clarkdale will be even better. But the architecture itself is the most impressive I've ever seen)

 

[IntelUser2000]

So anyone look at the Core i7 870 benchmark?? BTW, these are the CPUs the reason Intel is phasing out Core i7 940 and 950 and releasing the 960 instead. Performance will exceed the i7 940 and according to Intel are only 2-3% behind the 950.

Of course the reason for beating a higher clocked i7 9xx's is because of Turbo Mode. Thanks to being able to clock +2 speed grades higher even in 4 cores.

 

[taltamir]

Originally posted by: Denithor
And then count on at least two months for the price gouging to settle down before you can actually get one of these chips for a price anywhere near those shown in all the previews.

if you camp it out (or luck out), you can buy one before the price gouging on the first day.

 

[taltamir]

Originally posted by: IntelUser2000
So anyone look at the Core i7 870 benchmark?? BTW, these are the CPUs the reason Intel is phasing out Core i7 940 and 950 and releasing the 960 instead. Performance will exceed the i7 940 and according to Intel are only 2-3% behind the 950.

Of course the reason for beating a higher clocked i7 9xx's is because of Turbo Mode. Thanks to being able to clock +2 speed grades higher even in 4 cores.

the current i7s have a +2 speed increase, the lynfields have a +5 increase.
So I am not sure what you mean by +2

 

[IntelUser2000]

Originally posted by: taltamir

the current i7s have a +2 speed increase, the lynfields have a +5 increase.
So I am not sure what you mean by +2

Careless post I'd say.

Read again please:

"Thanks to being able to clock +2 speed grades higher even in 4 cores."

Single thread increase won't be as relevant as the uptime won't be as high and there aren't that many single thread apps anymore.