Haswell to Broadwell IPC

[TomFoster]

Perhaps it's time to upgrade from my Q9550 haha!

 

[PPB]

I'd like to see more data before I am willing to believe that kraken performance really did increase by 23%.

It can, certain software can really be benefited from the uarch changes. Ivy to Haswell wasn't a big IPC increace (most people agree on the 8% average number). Well, in certain renderers the benefit was from 20-25%.

Likewise, the total opposite can happen: Piledriver /L3less to Kaveri was about a 5-8% IPC uplift, well, in a renderer I use the benchmarks shows ZERO improvement, I have to check them again, because I dont remember if they even showed a performance regression (that would be dismissed because of the clock delta between 7850K and 6800K at that time).

 

[crashtech]

@tenks, Actually the OP's post DID say he normalized for frequency. I missed that. It's probably not too far off then, but you can't know for sure since scaling isn't always perfectly linear.

 

[Exophase]

Sorry for my ignorance, but when I first read the OP I understood normalize as he made the clocks the same. And then a few replys later I learned that's not what he did and others are asking for a clock for clock comparison. Thats not what he did? If not, then what did he do and why, I'm really confused?

I took normalize to mean that the Haswell scores were multiplied by (3.7/4.4). But that would assume that the Broadwell and Haswell CPUs always run at 3.7GHz and 4.4GHz respectively and that performance is linear to clock speed. Neither of which are necessarily true.

 

[Hulk]

Sorry for my ignorance, but when I first read the OP I understood normalize as he made the clocks the same. And then a few replys later I learned that's not what he did and others are asking for a clock for clock comparison. Thats not what he did? If not, then what did he do and why, I'm really confused?

Yes I normalized scores based on clock and then calculated percentages.

I assumed 4790K at 4400MHz and 5775C at 3700MHz. I assumed 4400MHz for 4790K because my 4770K will do max frequency on all cores stock. If 4400/3700 aren't the right numbers let me know what they are and I will update the Excel spreadsheet and the results in the initial post.

See edit in initial post.

 

[Yuriman]

Yes I normalized scores based on clock and then calculated percentages.

I assumed 4790K at 4400MHz and 5775C at 3700MHz. I assumed 4400MHz for 4790K because my 4770K will do max frequency on all cores stock. If 4400/3700 aren't the right numbers let me know what they are and I will update the Excel spreadsheet and the results in the initial post.

It depends on the load. In 3- and 4-core loads, a 4790K should only clock up to 4200mhz unless your motherboard is changing turbo frequencies for you, while 2-core loads should see 4300mhz and 1-core, 4400. I dunno what Broadwell's turbo bins are.

 

[Hulk]

It depends on the load. In 3- and 4-core loads, a 4790K should only clock up to 4200mhz unless your motherboard is changing turbo frequencies for you, while 2-core loads should see 4300mhz and 1-core, 4400. I dunno what Broadwell's turbo bins are.

This is why it would be nice if Anandtech noted the CPU frequency while running each test.

 

[know of fence]

Here is my back of the envelope math.

Seeing Hulk's numbers we see clock for clock score improvements of + 5 %, while frequency drops 0.7 GHz or 16 %, which in total is like going back one generation.

So we are back to Ivy bridge = Sandy bridge level of performance, which are conspicuously missing in all those comparisons.
[sarcasm] Possibly because we all are dying to know how 6 different SKUs of the blasted 3rd refresh 28 nm APUs compare. [/sarcasm]

Here's a random benchmark I picked from Bench, showing Broadwell-C ahead of Ivy-bridge. Also what's up with just 6 MB L3 cache? We've only seen that in i5 SKUs.

HandBrake v0.9.9 LQFilm
Intel Core i7 4790K (88W, $339) 4C/8T, 4.0 GHz, 1MB L2, 8MB L3 582.79
Intel Core i7 4770K (84W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 519.23
Intel Core i7 3770K (77W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 460.09
Intel Core i7 5775K (65W, $366)_4C/8T, 3.4 GHz, 1MB L2, 6MB L3 497.54
Intel Core i5 2500K (95W)______4C/4T, 3.3 GHz, 1MB L2, 6MB L3 416.6

 

[frozentundra123456]

Broadwell C is really targeted at igpu performance, not cpu performance. Supposedly Skylake brings back the frequency to close to haswell levels. We still dont have overclocking data on broadwell, although i dont expect much.

 

[Dufus]

^^^ Sweepr posted a link where overclocking past 4.2GHz was a problem.

http://forums.anandtech.com/showpost.php?p=37450200&postcount=2455

Of course it's only a single result, more will be needed to get a better idea.

 

[NTMBK]

Here's a random benchmark I picked from Bench, showing Broadwell-C ahead of Ivy-bridge. Also what's up with just 6 MB L3 cache? We've only seen that in i5 SKUs.

2MB of it is sacrificed to act as tags for the L4 cache.

 

[yacoub]

Here is my back of the envelope math.

Seeing Hulk's numbers we see clock for clock score improvements of + 5 %, while frequency drops 0.7 GHz or 16 %, which in total is like going back one generation.

So we are back to Ivy bridge = Sandy bridge level of performance, which are conspicuously missing in all those comparisons.
[sarcasm] Possibly because we all are dying to know how 6 different SKUs of the blasted 3rd refresh 28 nm APUs compare. [/sarcasm]

Here's a random benchmark I picked from Bench, showing Broadwell-C ahead of Ivy-bridge. Also what's up with just 6 MB L3 cache? We've only seen that in i5 SKUs.

HandBrake v0.9.9 LQFilm
Intel Core i7 4790K (88W, $339) 4C/8T, 4.0 GHz, 1MB L2, 8MB L3 582.79
Intel Core i7 4770K (84W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 519.23
Intel Core i7 3770K (77W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 460.09
Intel Core i7 5775K (65W, $366)_4C/8T, 3.4 GHz, 1MB L2, 6MB L3 497.54
Intel Core i5 2500K (95W)______4C/4T, 3.3 GHz, 1MB L2, 6MB L3 416.6

Yeah, I've been wondering why Ivy Bridge was left out of the test charts. It's much more relevant than some AMD stuff nobody buys. I have an Ivy Bridge setup and was wondering whether there's any point in upgrading to this new Broadwell series rather than waiting for Skylake but it seems like it'd be paying to basically just get a 3-year newer CPU with a couple new instruction sets and maybe 5-15% performance improvement?

 

[MrTeal]

Here is my back of the envelope math.

Seeing Hulk's numbers we see clock for clock score improvements of + 5 %, while frequency drops 0.7 GHz or 16 %, which in total is like going back one generation.

So we are back to Ivy bridge = Sandy bridge level of performance, which are conspicuously missing in all those comparisons.
[sarcasm] Possibly because we all are dying to know how 6 different SKUs of the blasted 3rd refresh 28 nm APUs compare. [/sarcasm]

Here's a random benchmark I picked from Bench, showing Broadwell-C ahead of Ivy-bridge. Also what's up with just 6 MB L3 cache? We've only seen that in i5 SKUs.

HandBrake v0.9.9 LQFilm
Intel Core i7 4790K (88W, $339) 4C/8T, 4.0 GHz, 1MB L2, 8MB L3 582.79
Intel Core i7 4770K (84W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 519.23
Intel Core i7 3770K (77W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 460.09
Intel Core i7 5775K (65W, $366)_4C/8T, 3.4 GHz, 1MB L2, 6MB L3 497.54
Intel Core i5 2500K (95W)______4C/4T, 3.3 GHz, 1MB L2, 6MB L3 416.6

Interesting that the result seems to show a performance/MHz regression from Haswell even with the eDRAM. Where the turbo clocks disabled in that test?

 

[know of fence]

Interesting that the result seems to show a performance/MHz regression from Haswell even with the eDRAM. Where the turbo clocks disabled in that test?

Bench just shows base frequencies, but since HandBrake is multithreaded, turbo at best adds +100 MHz anyway. Base frequency is probably a better reference than max Turbo, not that it matters much.

More data comparisons from Bench, showing Broadwell-C ahead of Ivy-bridge or behind it. Sorted by score.

HandBrake v0.9.9 LQFilm
Intel Core i7 4790K (88W, $339) 4C/8T, 4.0 GHz, 1MB L2, 8MB L3 582.79
Intel Core i7 4770K (84W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 519.23
Intel Core i7 5775K (65W, $366)_4C/8T, 3.4 GHz, 1MB L2, 6MB L3 497.54
Intel Core i7 3770K (77W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 460.09
Intel Core i5 2500K (95W)______4C/4T, 3.3 GHz, 1MB L2, 6MB L3 416.6

More favorable benchmark.
Agisoft PhotoScan Benchmark - Total Time Time in Minutes (Lower is Better)

Intel Core i7 4790K (88W, $339) 4C/8T, 4.0 GHz, 1MB L2, 8MB L3 18.05
Intel Core i7 5775K (65W, $366)_4C/8T, 3.4 GHz, 1MB L2, 6MB L3 18.24
Intel Core i7 4770K (84W) _____4C/8T, 3.5 GHz, 1MB L2, 8MB L3 19.01
Intel Core i7 3770K (77W)______4C/8T, 3.5 GHz, 1MB L2, 8MB L3 19.79
Intel Core i5 2500K (95W)______4C/4T, 3.3 GHz, 1MB L2, 6MB L3 25.28

Office workloads, 3D Particle Movement Single threaded
Intel Core i7 4790K (88W, $339)4C/8T, 4.0 GHz, 1MB L2, 8MB L3 146.06
Intel Core i7 4770K (84W)_____4C/8T, 3.5 GHz, 1MB L2, 8MB L3 129.37
Intel Core i7 3770K (77W)_____4C/8T, 3.5 GHz, 1MB L2, 8MB L3 126.54
Intel Core i7 5775K (65W, $366)4C/8T, 3.4 GHz, 1MB L2, 6MB L3 125.98
Intel Core i5 2500K (95W)______4C/4T, 3.3GHz, 1MB L2, 6MB L3 117.5

Add 3% score for the 0.1 GHz frequency disparity and we are at 129 in the above benchmark...

 

[Arachnotronic]

^^^ Sweepr posted a link where overclocking past 4.2GHz was a problem.

http://forums.anandtech.com/showpost.php?p=37450200&postcount=2455

Of course it's only a single result, more will be needed to get a better idea.

Probably why Intel wasn't showing off this chip's OC'ing capabilities at Computex.

 

[Hulk]

This is what I was really hoping to see in the review.

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6

 

[Exophase]

This always bugged me. If all four cores run at 4.2....then why is it marketed as just 4.0?

Several reasons. Not all CPUs will have the same power consumption at a given clock speed. Not all workloads will either; AVX heavy code will use more power than code that's predominantly scalar. And there's other stuff on the die that could use a significant amount of power besides the CPU cores, particularly the GPU.

 

[yacoub]

This is what I was really hoping to see in the review.

http://www.anandtech.com/show/7003/the-haswell-review-intel-core-i74770k-i54560k-tested/6

Awesome, that's what I wanted to see too!

 

[know of fence]

So the Xeons have launched and they are substantially more expensive, without being all that much faster, there is a 95 W TDP version that just bumps base clock to 3.5.
When a 0.1 base clock comes with a 30W TDP increase, I'd be looking at temperatures. There aren't any xeons out wit disabled graphics this time, which in the past allowed to see how much of the TDP the iGPU accounted for (a nominal 7 W).

Devil's Canyon 4970k certainly was quite an extraordinarily high clocked chip, even though it came pretty late.
But this is the first time since the reveal of the tick-tock release cadence, that a die shrink (tick) actually went down a notch in frequency.

 

[ClockHound]

Devil's Canyon 4970k certainly was quite an extraordinarily high clocked chip, even though it came pretty late.
But this is the first time since the reveal of the tick-tock release cadence, that a die shrink (tick) actually went down a notch in frequency.

True, however, the Broadwell 5775C was meant to be released last year as the successor to the 4770k - which would have been clock neutral, but then stuff (bad 14nm yield stuff) happened and we got DC instead with it's higher clocks, better TIM and devilish performance boost.

The Tick Tock clock rhythm is irregular - Maybe Intel is getting into its 'Free Jazz' period. Where the cool concept is the important part, not some stupid rigid delivery schedule. Chip geeks - they're artists too ya know. ;-)

 

[know of fence]

True, however, the Broadwell 5775C was meant to be released last year as the successor to the 4770k - which would have been clock neutral, but then stuff (bad 14nm yield stuff) happened and we got DC instead with it's higher clocks, better TIM and devilish performance boost.

Just looking at the flagship desktop CPUs and base clocks, the much maligned and delid-happy 4770K is actually faster by 6%, which is why it beats the new i7 in a slew of benchmarks.

Does that look like progress, or clock neutral?
http://ark.intel.com/compare/35428,88040,75123,65523,61275

We've seen OC headroom shrink, but now it's official, desktops are slowing down. Is it in favor of graphics, yield optimization, "mobile first", lack of competition, reasons unknown or all of the above?

 

[dahorns]

Just looking at the flagship desktop CPUs and base clocks, the much maligned and delid-happy 4770K is actually faster by 6%, which is why it beats the new i7 in a slew of benchmarks.

Does that look like progress, or clock neutral?
http://ark.intel.com/compare/35428,88040,75123,65523,61275

We've seen OC headroom shrink, but now it's official, desktops are slowing down. Is it in favor of graphics, yield optimization, "mobile first", lack of competition, reasons unknown or all of the above?

I thought we already established that the 5775c isn't meant as a replacement to the 4770k? Pretty clear Intel is using Skylake to do that. 5775c is a niche chip designed for those that want the best in integrated graphics to go along with a high end cpu. Also, saying the 4770k beat the 5775 a "slew of times" is misleading. Looking at the reviews, the 5775 seems faster than the 4770k in about everything. The 4790k does have the upper hand, but barely.

 

[Dave2150]

I thought we already established that the 5775c isn't meant as a replacement to the 4770k? Pretty clear Intel is using Skylake to do that. 5775c is a niche chip designed for those that want the best in integrated graphics to go along with a high end cpu. Also, saying the 4770k beat the 5775 a "slew of times" is misleading. Looking at the reviews, the 5775 seems faster than the 4770k in about everything. The 4790k does have the upper hand, but barely.

Yes this is indeed the case - just a few silly people who don't understand that the 5775C is not meant to compete with the 4790k at all.

Wait for the 6700k if you want a 4790k successor.

 

[PPB]

It will be odd to see a comparison between both 3 of them:

4790k being the first push into the sweet spot between frequency and voltage (mine does 4.0 at 1.002 VID, prefixed LLC makes it 1.02 actually) and thus delivering good performance even if its the older and with the worse IPC of the trio.

Then you have the 5775c, having small ipc improvements over haswell but an L4 cache that skews rhe avg IPC increase more in its favour, making it a tick+ in terms of IPC improvement.

Sadly clocks dont follow the trend with DC 4790K, reverting those gains in performance a notch and making it discretely match the 4790K.

Lastly you have 6700K. It regains the clocks and should further improve the ipc from broadwell. But here is the catch: IPC gains statements in rumours always compared haswell to skylake, they accounted the IPC gain already in broadwell. They didnt also account the perf boost the L4 gives too, making it look more of a tock- than anything.

My conclusion is as follows: expect the stagnation to diminiah a little but still tarnish any claim of revolutionary improvenents of skylake. Specially because broadwell now looks better than it supposed to with the L4 cache. If 6700K was indeed a C suffix part and had crystalwell to, it would have lived up to the hype. Sadly only the clocks remain from 4790K.

 

[know of fence]

I thought we already established that the 5775c isn't meant as a replacement to the 4770k? Pretty clear Intel is using Skylake to do that. 5775c is a niche chip designed for those that want the best in integrated graphics to go along with a high end cpu. Also, saying the 4770k beat the 5775 a "slew of times" is misleading. Looking at the reviews, the 5775 seems faster than the 4770k in about everything. The 4790k does have the upper hand, but barely.

Actually the i7-5775C is a replacement for whatever you put into Socket 1150 of the Z97 / H97 motherboard before, except for people who build systems now the i7-4970K actually may be a better choice.

Could you also link some of those reviews you mentioned. I genuinely hope to see some proper reviews and comparisons.
Including the ol' Haswell in benchmark comparisons would make the new CPU look bad, something Intel probably wouldn't sponsor. A website that includes a direct comparison to the 4770K and/or the 3770K would be my personal "most trusted in tech since 2015".