I know the trend is more cores now but considering a lot of stuff still doesn't make use of everything and as i've not been keeping aware of all the changes for sometime now, im left wondering where individual cores are at when it comes to yearly improvements, are the jumps in performance like they used to be?
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
Is performance still increasing on individual cores?
- Thread starter Radiation
- Start date
Man of Honour
- Joined
- 13 Nov 2009
- Posts
- 11,663
- Location
- Northampton
Nope if anything theyve gone a bit backwards in temps of individual core speed is concered on the intel side, however i3/5/7 chips overclock very well the majority hitting 4ghz easily
Between 775 and 1366 Intel changed the cache layout between cores. On 775, each core has its own cache and there's a collective pool. If it doesn't find the result in its own cache, or the pool, it then has to query the cache of the other cores and finally run the calculation if it doesn't find the result. On 1366, each core has some cache dedicated to it, and the data in this cache is mirrored to the pool. It leads to less cache available overall, but if the core doesn't find data in its own cache or the pool, it knows it isn't stored anywhere on the chip.
The latter layout is more efficient, as you don't have cores repeatedly trying to access each others cache. This, possibly combined with triple channel ram, accounts for much of the performance per clock improvement of the new intel generation.
The point I'm trying to make is that your question doesn't make much sense. Is a core the processing area with cache attached? Or without it attached? And what do you make of any shared memory? At best all you can do is take two chips with the same number of cores and run tests against each other, which does lead to Q6600 < Q9650 < i7 920 without HT, all tested at the same clock speed.
I think 45nm transistors switch quicker than 65nm, and likewise 32nm quicker than 45nm, so shrinking the process size will also account for some of the improvement.
I've no idea what AMD is doing though, over to BigWayne for that.
The latter layout is more efficient, as you don't have cores repeatedly trying to access each others cache. This, possibly combined with triple channel ram, accounts for much of the performance per clock improvement of the new intel generation.
The point I'm trying to make is that your question doesn't make much sense. Is a core the processing area with cache attached? Or without it attached? And what do you make of any shared memory? At best all you can do is take two chips with the same number of cores and run tests against each other, which does lead to Q6600 < Q9650 < i7 920 without HT, all tested at the same clock speed.
I think 45nm transistors switch quicker than 65nm, and likewise 32nm quicker than 45nm, so shrinking the process size will also account for some of the improvement.
I've no idea what AMD is doing though, over to BigWayne for that.
Soldato
- Joined
- 23 Apr 2009
- Posts
- 11,973
- Location
- Cheshire
Yea, they are still pretty good. For example, the nahalem chips are a good deal faster clock-for-clock, core-for-core compare to core 2 chips.
Also, from the initial tests on Intel Sandy bridge chips (coming out early next year) these are clock-for-clock 15-20% faster than nahalem chips. So their is definite progress in the architecture department (AMD are expected to have simialr gains with their new Bulldozer cores).
Furthermore, as the manufactuering process gets smaller Intel and AMD are able to clock the chips higher without generating as much heat or consume as much power. Therefore new a Sandy bridge desktop CPU may use as much power as a current i5 chip, however the architecture is clock-for-clock faster and it is clocked higher than the equivalent i5, so the computing power for the same electrical power is continuing to increase.
Also, from the initial tests on Intel Sandy bridge chips (coming out early next year) these are clock-for-clock 15-20% faster than nahalem chips. So their is definite progress in the architecture department (AMD are expected to have simialr gains with their new Bulldozer cores).
Furthermore, as the manufactuering process gets smaller Intel and AMD are able to clock the chips higher without generating as much heat or consume as much power. Therefore new a Sandy bridge desktop CPU may use as much power as a current i5 chip, however the architecture is clock-for-clock faster and it is clocked higher than the equivalent i5, so the computing power for the same electrical power is continuing to increase.
Associate
- Joined
- 24 Jan 2010
- Posts
- 2,141
It's slowed down a lot compared to how things were progressing years ago. They're close to reaching several "built-in" limits. Heat and power levels can't get too much higher than they are currently...... unless we all want our PC's to act like heaters, and our power bills to go through the roof.
They're also getting close to the limits of how small they can make the transistors using silicon.
The closer they get to these limits the harder its getting for them to push more performance out of a single core, so now they've switched to parallelism in order to keep up with moores law.
They're also getting close to the limits of how small they can make the transistors using silicon.
The closer they get to these limits the harder its getting for them to push more performance out of a single core, so now they've switched to parallelism in order to keep up with moores law.
Associate
- Joined
- 21 Aug 2010
- Posts
- 754
It's not just the stated speed that matters. It's how the chip works. A 3gig core2 v a 3gig i3 look the same but the i3 will be faster
I dont' think that was the OP point, but feel free to correct me if I have assumed wrong but I think the OP was talking just about the core speeds of the chips.
If I am correct then i see where you are coming from, for example - About 6 or 7 years ago I bought a 3Ghz AMD Athlon, last month I bought spec below but speed is now 3.2Ghz (though overclockable to higher) which isn't a big increase. Yes I know layout has changed we now have quad/hex core but clock speeds are still not much higher. The reason for this I think Jimbob explained well enough.
The first 3Ghz chip AMD released was the FX-74 in november 2006
Associate
- Joined
- 27 Feb 2009
- Posts
- 565
- Location
- Hinckley
If i think right, the only thing that changed in the early pentium 1 - 4 days was the increase in clock speed?
Pentium 90 was awesome btw!
Pentium 90 was awesome btw!
Haha yup, bad memory on my part I apologise. Just looked up my old spec, was an Athlon 3000 - I don't know why I thought it was 3Ghz. Still rest of my post stands to a degree.
No, there was masses of architectural changes between the P1 and P4, firstly the L1 and L2 cache on P1's was minuscule compared to later Pentium II and III chips, which used cartridge designs so they had enough space for the extra cache. The later Pentium III's becoming similar to todays chips and the earlier P1's in that they were a single package.
P4's improved on the power consumption and they were pushed from then on till they hit the 4GHz barrier, past that heat becomes a major issue as the quartz oscillates so fast it hits microwave frequencies, so intel and amd switched to multicore chips to deal with the heat.
As programs become more and more thread orientated and we started to do more tasks simultaneously (I was just gaming, encoding a H.264 video and ripping a dvd at the same time) the extra cores make more sense than just simply boosting the speed of a single core anyway.
P4's improved on the power consumption and they were pushed from then on till they hit the 4GHz barrier, past that heat becomes a major issue as the quartz oscillates so fast it hits microwave frequencies, so intel and amd switched to multicore chips to deal with the heat.
As programs become more and more thread orientated and we started to do more tasks simultaneously (I was just gaming, encoding a H.264 video and ripping a dvd at the same time) the extra cores make more sense than just simply boosting the speed of a single core anyway.
More applications are written to take advantages of multiple cores - so the individual core speed matters less and less.
'Every day' applications such as google chrome and office 2007+ are written to take advantage of more cores.
If AMD's bulldozer can unify threads then a single thread app will be able to run on multiple cores - this will be a milestone!
'Every day' applications such as google chrome and office 2007+ are written to take advantage of more cores.
If AMD's bulldozer can unify threads then a single thread app will be able to run on multiple cores - this will be a milestone!
not much has really changed since i7/phenom2 came out there basicly the same chips with higher frequency.
individual core speed still matters loads actually rickh, you can only program multicore efficiency so much , games are rubbish at multicore you get 1-2 tasks that need a lot of cpu and you cant split it across cores.
even stuff like the @home clients only use 1 core per job.
multicore is great if what your doing is predictable like encoding and rendering where you can easily split it into chunks and assign each core different chunks but when its something with many variables you cant really split the job across cores or everything gets messy
individual core speed still matters loads actually rickh, you can only program multicore efficiency so much , games are rubbish at multicore you get 1-2 tasks that need a lot of cpu and you cant split it across cores.
even stuff like the @home clients only use 1 core per job.
multicore is great if what your doing is predictable like encoding and rendering where you can easily split it into chunks and assign each core different chunks but when its something with many variables you cant really split the job across cores or everything gets messy
Associate
- Joined
- 24 Jan 2010
- Posts
- 2,141
Individual cores are still getting faster, but just not with the speed they used to progress years ago. (due to the reasons I stated above)
Just compare the benchmarks of the first quad cores, to those the i7 975 produce. Even if you were to match the clockspeeds exactly, you'd still see the 975 performing a lot better.
Just compare the benchmarks of the first quad cores, to those the i7 975 produce. Even if you were to match the clockspeeds exactly, you'd still see the 975 performing a lot better.
Watch the first part of this. It's talking about how we have really reached single core speed performance (compared to all the other x86's) - and must program in parallel
http://www.infoq.com/presentations/Modeling-Concurrency-with-Actors-in-Java
http://www.infoq.com/presentations/Modeling-Concurrency-with-Actors-in-Java