When will we see a CPU thats double the performance of a 5 year old cpu?

[jonneymendoza]

Bought my 920 within weeks of release (stupidly didn't wait for D0 stepping) but its still going strong!!

i got the D0 stepping

mines running like a champ.poor that 5 years later and nothing beats it over 50% clock for clock

 

[JonSmith]

Single thread in Cinebench R15 for a Q6600 @ 3.6GHz gives a result of 93. A 4770k at stock gives 170, so very close to twice as fast especially after a small OC.

 

[dvs8]

Just to add some real figures to this discussion, at the weekend I upgraded from my 5 year old i5 750 (Lynnfield) to a Core i5 4670K

Ran the SuperPI 1M test (which I've used for years as a general CPU speed measure) - almost exactly the same time for my turbo-clocked Haswell as my manually overclocked Lynnfield - both running at 3.8GHz.

So in 5 years, the clock-for-clock performance in that particular test has moved precisely nowhere.

 

[jonneymendoza]

Just to add some real figures to this discussion, at the weekend I upgraded from my 5 year old i5 750 (Lynnfield) to a Core i5 4670K

Ran the SuperPI 1M test (which I've used for years as a general CPU speed measure) - almost exactly the same time for my turbo-clocked Haswell as my manually overclocked Lynnfield - both running at 3.8GHz.

So in 5 years, the clock-for-clock performance in that particular test has moved precisely nowhere.

Just as i thought. no advancement in CPU tech

 

[JonSmith]

SuperPI is not a great benchmark to use since it uses legacy, hand optimised assembly instructions. The limitation then really becomes FPU pipeline length which I believe is the same for Lynnfield and Sandy/Ivy/Haswell.

I would've expected it to be faster due to the cache improvements though, although it's quite possible that the cache bandwidth is sufficient to feed the FPU given that it's barely being utilised. A modern FPU is very wide which is where the performance comes from when using 128-256bit instructions.

 

[Adrianr]

Single thread in Cinebench R15 for a Q6600 @ 3.6GHz gives a result of 93. A 4770k at stock gives 170, so very close to twice as fast especially after a small OC.

Interesting, thanks for posting. I don't suppose you know the numbers for multi-threaded performance at all?

 

[JonSmith]

~ 360, the scaling is practically 100%

 

[Adrianr]

~ 360, the scaling is practically 100%

Fair enough. I think the last time I ran it at 4.5ghz returned 866, though granted not all 4770k's can hit that on sensible volts. Even 4ghz would easily surpass 720 though.

 

[easyrider]

Still on my trusty 920 here...see no reason to upgrade....the 920 is legendary

 

[Phixsator]

Still on my trusty 920 here...see no reason to upgrade....the 920 is legendary

It truly is. Most of them, even C0 steppers, would do 4,0ghz on a proper board which is a pretty darn impressive 50ish % OC.

 

[Ryan-3]

Still on my trusty 920 here...see no reason to upgrade....the 920 is legendary

Same here!

Legendary chip.

 

[melvis]

The trouble is theres really no mainstream application where a high end CPU is the limiting factor.

Just to add some real figures to this discussion, at the weekend I upgraded from my 5 year old i5 750 (Lynnfield) to a Core i5 4670K

Ran the SuperPI 1M test (which I've used for years as a general CPU speed measure) - almost exactly the same time for my turbo-clocked Haswell as my manually overclocked Lynnfield - both running at 3.8GHz.

So in 5 years, the clock-for-clock performance in that particular test has moved precisely nowhere.

i did the same upgrade a little while ago, i got a small boost ~10% simply because my 750 was a very poor clocker, my main motivation was mitx motherboards for 1156 were damn near impossible to find at any sensible price anymore, even 2nd hand.

 

[Nutella33]

Superpi is a basically obsolete test though? It's an appalling judge of a new cpu's ability/performance even for IPC

 

[Terminal_Boy]

I only upgraded my Q6600 a couple of weeks ago! Funnily enough for the space of time between components, I'm still not running any more ram (16GB in both systems), however a Q6600 at 3.4ghz to a 4770K at 4.5ghz is quite a big jump in performance.

Q6600 @ 3GHz (air) to i5 2500K @ 4.3GHz (water) was a big jump 2 years ago and I haven't seen any need to upgrade since as it's a gaming rig, not a benching or mining rig.

 

[Calabi]

Superpi is a basically obsolete test though? It's an appalling judge of a new cpu's ability/performance even for IPC

So those only optimised for the current cpu's are valid?

Sounds like a clever trick to make us think were getting somewhere.

So from this I'm guessing its not worth upgrading my i5 760.

 

[the_director]

When I moved from a Q6600 @ 3.2 to a 3570K it made a massive difference in games because I had crossfire GPUS - frame rates easily doubled.

 

[Nutella33]

So those only optimised for the current cpu's are valid?

Sounds like a clever trick to make us think were getting somewhere.

So from this I'm guessing its not worth upgrading my i5 760.

Are you regularly going to be running software based on an obsolete instruction set? and why would you believe it to be the most valid measure of performance given how it doesnt scale to any real-world application?

I don't think it's a trick as no sane person would buy something based on superpi scores. Even in the tiny percentage of the market that looks at benchmarks before purchase.

As for the upgrade. Depends if it does everything you require of it or not or if you just fancy an upgrade *shrugs*

 

[Nasher]

We are getting quite close to the limit now, at least with silicone

The next big technology jump will be to carbon based chips.

 

[badcompany]

AFAIK
Q9550 @3.4 cinebench was 4.0 ish
Fx8350 @ 4.7 cinebench was 8 ish

 

[uksoldierboy]

Just as i thought. no advancement in CPU tech

It's depends totally on your needs, i especially wouldn't base my conclusion on a Super PI run.

It's purely subjective and based on what it is your using it for.

No way would i go back to my 920.