3770K vs 4770K test

[Hulk]

Whoa... If this is correct that 10% improvement in CPUmark99 is significant. That is a huge jump in legacy integer IPC.

That being said I'm very suspect of this score.

While it is very hard to get accurate generation-to-generation comparisons in this benchmark when you are collecting scores from the web because you don't know how the test was run (background processes, etc..) but I've been tracking this bench since it came out and based on my observations here are my generalizations. I find it hard to believe Haswell could eke out another 10% legacy integer IPC. I guess it is possible with the wider back end execution engine, better branch predictor, and wider cache. And I hope it's true but I'm very skeptical...

Improvements have been almost nil since Nehalem.

The number next to the core is the efficiency in running CPUmark99. It is the MHz/CPUmark99. Or put another way the number of MHz required for a particular core to "earn" one CPUmark99 point during the test.

As you can see any Core processor is over three times more efficient than Prescott.

486 25
Pentium P54C 16.4
Pentium P55C 14.9
Celeron Mendocino 12.5
Pentium II (Dechutes) 13.2
Pentium III (Katmai) 13
Pentium III (Coppermine) 11.2
Pentium III (Tulatin) 11
Pentium 4 Willamette 17
Pentium 4 Northwood 15.8
Pentium 4 Prescott 20.7
Conroe 7.1
Penryn 6.9
Nehalem 6.5
Westmere 6.6
Sandy Bridge 6.5
Ivy Bridge 6.4
Haswell 5.8

 

[guskline]

Funny thing about this discussion is the comparison of cpus to gpus when it comes to price.

The cpu prices have stayed pretty close with little upward creep in price while the top gpus have gone from what @$500 to $1,000! That's an increase I can understand!

Perhaps that explains the comment that GPUs have bigger jumps in performance.

 

[Enigmoid]

Funny thing about this discussion is the comparison of cpus to gpus when it comes to price.

The cpu prices have stayed pretty close with little upward creep in price while the top gpus have gone from what @$500 to $1,000! That's an increase I can understand!

Perhaps that explains the comment that GPUs have bigger jumps in performance.

Also look at power. Intel has dropped power tremendously compared to the p4 era. Modern gpus are using more power than those 5-10 years ago. At the same power envelope intel has seen substantial gains.

 

[guskline]

Also look at power. Intel has dropped power tremendously compared to the p4 era. Modern gpus are using more power than those 5-10 years ago. At the same power envelope intel has seen substantial gains.

Excellent point. It was a factor for me going with the GTX670/680s versus The Radeon 7900s. In either case the GPUs can use a ton of power.

 

[Borealis7]

10%? works for me! (C2D user)

 

[frozentundra123456]

Whoa... If this is correct that 10% improvement in CPUmark99 is significant. That is a huge jump in legacy integer IPC.

That being said I'm very suspect of this score.

While it is very hard to get accurate generation-to-generation comparisons in this benchmark when you are collecting scores from the web because you don't know how the test was run (background processes, etc..) but I've been tracking this bench since it came out and based on my observations here are my generalizations. I find it hard to believe Haswell could eke out another 10% legacy integer IPC. I guess it is possible with the wider back end execution engine, better branch predictor, and wider cache. And I hope it's true but I'm very skeptical...

Improvements have been almost nil since Nehalem.

The number next to the core is the efficiency in running CPUmark99. It is the MHz/CPUmark99. Or put another way the number of MHz required for a particular core to "earn" one CPUmark99 point during the test.

As you can see any Core processor is over three times more efficient than Prescott.

486 25
Pentium P54C 16.4
Pentium P55C 14.9
Celeron Mendocino 12.5
Pentium II (Dechutes) 13.2
Pentium III (Katmai) 13
Pentium III (Coppermine) 11.2
Pentium III (Tulatin) 11
Pentium 4 Willamette 17
Pentium 4 Northwood 15.8
Pentium 4 Prescott 20.7
Conroe 7.1
Penryn 6.9
Nehalem 6.5
Westmere 6.6
Sandy Bridge 6.5
Ivy Bridge 6.4
Haswell 5.8

Very interesting data. There is one point that is missing however. Until ivb we were seeing steady increases in clockspeed as well, so overall performance was increasing more than the ipc alone. Now we are seeing no increase in clock speed, so the only improvement comes from the incremental increase in ipc.

 

[Idontcare]

I think the point was more about 22nm should be clocking past 5ghz but it doesnt.

Who's to say it doesn't?

IB may not overclock any higher than SB but that doesn't mean 22nm is what is holding IB back.

No more than you'd conclude GloFo's 32nm is what is holding Llano clockspeeds back compared to 45nm Thuban.

For all we know IB was intentionally optimized to enable a lower TDP (95W -> 77W) knowing full-well in advance that in doing so the layout of the chip itself was going to impart a lower clockspeed envelope in comparison to SB's design.

Isn't that what happened with Llano, as deftly demonstrated by bulldozer and piledriver (trinity) clockspeeds?

 

[Pilum]

For all we know IB was intentionally optimized to enable a lower TDP (95W -> 77W) knowing full-well in advance that in doing so the layout of the chip itself was going to impart a lower clockspeed envelope in comparison to SB's design.

According to David Kanter after he talked with Intel people, IVB was mostly a quick shrink of SNB. HSW is supposed to be really optimized for 22nm. So we might see different OC results for HSW.

 

[Hulk]

Very interesting data. There is one point that is missing however. Until ivb we were seeing steady increases in clockspeed as well, so overall performance was increasing more than the ipc alone. Now we are seeing no increase in clock speed, so the only improvement comes from the incremental increase in ipc.

My metric is MHz/CPUmark99 which effectively takes clockspeed out of the equation. The table I posted simply shows the efficiency of each core at processing CPUmark99.

It shows how many MHz each generation of Intel microprocessor architecture needs to produce 1 CPUmark99 point.

 

[Idontcare]

Whoa... If this is correct that 10% improvement in CPUmark99 is significant. That is a huge jump in legacy integer IPC.

That being said I'm very suspect of this score.

While it is very hard to get accurate generation-to-generation comparisons in this benchmark when you are collecting scores from the web because you don't know how the test was run (background processes, etc..) but I've been tracking this bench since it came out and based on my observations here are my generalizations. I find it hard to believe Haswell could eke out another 10% legacy integer IPC. I guess it is possible with the wider back end execution engine, better branch predictor, and wider cache. And I hope it's true but I'm very skeptical...

Improvements have been almost nil since Nehalem.

The number next to the core is the efficiency in running CPUmark99. It is the MHz/CPUmark99. Or put another way the number of MHz required for a particular core to "earn" one CPUmark99 point during the test.

As you can see any Core processor is over three times more efficient than Prescott.

486 25
Pentium P54C 16.4
Pentium P55C 14.9
Celeron Mendocino 12.5
Pentium II (Dechutes) 13.2
Pentium III (Katmai) 13
Pentium III (Coppermine) 11.2
Pentium III (Tulatin) 11
Pentium 4 Willamette 17
Pentium 4 Northwood 15.8
Pentium 4 Prescott 20.7
Conroe 7.1
Penryn 6.9
Nehalem 6.5
Westmere 6.6
Sandy Bridge 6.5
Ivy Bridge 6.4
Haswell 5.8

I didn't realize before what you had pulled together in this post I am quoting now.

Very nice, very nice indeed :thumbsup:

 

[mikk]

Nehalem 6.5
Sandy Bridge 6.5


Nice list but I don't think this is correct. See here: http://www.coolaler.com/showthread....kd-1156-比較安全?p=2716654&viewfull=1#post2716654

 

[Hulk]

The Ivy Bridge score posted in the Haswell review gets a metric of 6.4, which is a little faster than the 6.5 most people have reported. Still that score is definitely in the margin of error to be legit. That score does lend credence to the Haswell score posted, which shows a metric of 5.8. Which, if true, is a really nice integer IPC increase and could well lead to a nice performance boost on legacy code.

Won't be long now before we know the whole story but this gets me more excited for the official Anandtech review.

It is interesting how the efficiency gets better and better until the P4, then the trend reverses until Core. The one exception is Mendocino (Celeron 300A) but that core had the on die 128KB full speed cache. Since CPUmark99 fits entirely in this L2 that explains why Mendocino is better at this bench than Dechutes and Katmai. Once the L2 made it to the die with Coppermine things got back on track.

 

[Hulk]

Nehalem 6.5
Sandy Bridge 6.5


Nice list but I don't think this is correct. See here: http://www.coolaler.com/showthread....kd-1156-比較安全?p=2716654&viewfull=1#post2716654

No I'm pretty sure those scores you linked to are wrong. I've got literally dozens of SB scores that verify 6.5. Many from previous threads here. And my own system corroborates those scores. Also lots of Nehalems that verify the 6.5. They could be off but I don't think by more than + or - 0.1. There will always be wild scores outside the bounds of reality but I disregard them.

The SB score you linked would be 5.45 which is way off. They got the Nehalem score right but are reporting either the wrong clock or score for the SB.

 

[AustinInDallas]

I went from the i5 750 to the i5 3570k. i think It will last me a few years.

I actually sold my i5 750, motherboard, 4 gb ram, and coolermaster v8 for a few dollars less than I paid for my i5 3570k, ASrock z77 extreme 4 and 8 gigs of ram (thank you microcenter!) . using stock cooler atm, but will get a cooler when I need it

 

[mikk]

No I'm pretty sure those scores you linked to are wrong. I've got literally dozens of SB scores that verify 6.5. Many from previous threads here. And my own system corroborates those scores. Also lots of Nehalems that verify the 6.5. They could be off but I don't think by more than + or - 0.1. There will always be wild scores outside the bounds of reality but I disregard them.

The SB score you linked would be 5.45 which is way off. They got the Nehalem score right but are reporting either the wrong clock or score for the SB.

Do you have a link to other SB results?

 

[JAG87]

Marginal. Especially once you factor that TDP went up 7W.

Haswell is actually more exciting for GT3e than anything else it brings to the table.

 

[Face2Face]

The IPC increase is pretty much expected. I am more interested in what kind of overclocks Haswell is capable of. If most chips can do 5Ghz+ - The 3570K will be for sale

So would this be a fair estimate? 2600K @ 4.8Ghz = 3770K @ 4.5Ghz = 4770K @ 4.2Ghz

 

[Pheesh]

Marginal. Especially once you factor that TDP went up 7W.

Haswell is actually more exciting for GT3e than anything else it brings to the table.

TDP rating going up is kind of misleading with the on die VRM's, we'll need to see the actual power consumption/efficiency of the platform to see what the differences are. Looks like a good upgrade from my core2.

 

[lagokc]

Funny thing about this discussion is the comparison of cpus to gpus when it comes to price.

The cpu prices have stayed pretty close with little upward creep in price while the top gpus have gone from what @$500 to $1,000! That's an increase I can understand!

Perhaps that explains the comment that GPUs have bigger jumps in performance.

The Quantum Obsidian X-24 was $699 and that was in 1998 dollars ($942.14 in 2012 dollars)! Exotic gaming video cards have always been expensive.

Oh god I just realized the Voodoo2 came out 15 years ago. Brb, I have to go tell some kids to get off my lawn.

 

[AtenRa]

According to David Kanter after he talked with Intel people, IVB was mostly a quick shrink of SNB. HSW is supposed to be really optimized for 22nm. So we might see different OC results for HSW.

If Intel will focus on lower power consumption and try keeping the die size as small as possible on HW due to bigger iGPU than IvyBridge, then they will use smaller (denser) transistors. That will make the CPU OC headroom being lower than if they would use bigger transistors.

I believe the same happened with Llano.

 

[AtenRa]

I would not conclude that Intel's 22n is "very poor".

i7-3770K vs. i7-2600K: Temperature, Voltage, GHz and Power-Consumption Analysis

Heh, i have just remembered this,

From the bellow diagram it is clear that the 22nm 3D Tri-Gate advantage is more down to lower Voltage than at Higher V where High End CPUs will be operate.

The question is, does that has anything to do with a Thicker Tox vs the Planar 22nm ?? or is it a FinFET design characteristic ??

What i mean is, did Intel intentionally made the Tox thicker in order to have lower leakage and lower Gate delay (higher performance) at lower Voltages where Mobile and SOC's chips will benefit most ?? and at the same time they managed to keep the same Gate delay as the Planar 22nm at 1V (actually is a little better than the planar).

Im asking to see if they have starting to focus more at the mobile/SOCs and keep a nice ~20% higher performance for the High End CPUs (from the manufacturing point of view) per every shrinking process.

From your slide, it is clear that IvyBridge Higher Delta of Vcc to Frequency occurs at almost 0.7V up to 0.9V. Passing 1.0V and it starts to react almost as the 32nm.

 

[JAG87]

TDP rating going up is kind of misleading with the on die VRM's, we'll need to see the actual power consumption/efficiency of the platform to see what the differences are. Looks like a good upgrade from my core2.

That's true, I forgot about that

 

[Hulk]

Do you have a link to other SB results?

Here's one and there have been others here over the years. I have my own database of scores compiled from this site and from my own systems over the years the other systems (friends) that I have personally tested. So I pretty much know the legit range of scores for a given processor range.

http://forums.anandtech.com/showthread.php?t=2057154&highlight=cpumark&page=9

 

[Lepton87]

486 25
Pentium P54C 16.4
Pentium P55C 14.9
Celeron Mendocino 12.5
Pentium II (Dechutes) 13.2
Pentium III (Katmai) 13
Pentium III (Coppermine) 11.2
Pentium III (Tulatin) 11
Pentium 4 Willamette 17
Pentium 4 Northwood 15.8
Pentium 4 Prescott 20.7
Conroe 7.1
Penryn 6.9
Nehalem 6.5
Westmere 6.6
Sandy Bridge 6.5
Ivy Bridge 6.4
Haswell 5.8

That benchmark has some quirks or one score is just reported wrong. P4 Prescott 20.7, Willamette 17. Prescott was definitely faster clock for clock then Willamette, actually it had IPC very close to Northwood, at low clocks it was slightly lower but it caught up at high clocks. Nehalem to SB also doesn't paint a true picture as SB was clearly faster then Nehalem, IPC went up north of 15%.

 

[Hulk]

Marginal. Especially once you factor that TDP went up 7W.

Haswell is actually more exciting for GT3e than anything else it brings to the table.

TDP has to go up when you add all those xtors at the same process.

Truth be told, legacy integer performance (ie CPUmark99) isn't all that important these days as very few people are constrained by MS Office performance or the like. Now I realize that there are people that my have niche apps that need this, which is why I wrote "very few people." It's video, 3D rendering, and gaming that we are looking to for major performance improvements.

Still it is interesting to note that Intel is still able to increase IPC for this old benchmark.