AnandTech Ivy Bridge Performance Preview

[dma0991]

I'm pretty sure there is enough OC headroom to take it to 5GHz. The HD4000 is closing the gap with the HD6550D in some games, impressive. :thumbsup:

 

[paperwastage]

Why is Intel focusing on the freaking IGP? More CPU performance/cores, please!

if you want better CPU/performance, go SB-E Xeon (that was just released?)

think a 95W TDP is pretty good (or 77W in Ivy Bridge)

 

[bgcpen]

Can't wait. Thx for the preview!

 

[Arachnotronic]

Tocks are never as exciting as ticks I think most would agree. A die shrink, some new features, better thermals and overclocking. Pretty cool but not that much of a thrill. But a tick brings pages and pages of new architectural improvements with lots of cool diagrams and hours of reading to try and figure it all out (for me anyway) before you see the actual IPC improvements in various applications in the benchmarks. We're always looking for the next release to be as dramatic a performance improvement as we witnessed in going from P4 to Conroe. That was nearly a doubling of performance per clock. It spoiled us.

All that said I'm grabbing one of these as soon as I can.

Netburst -> Conroe seemed so awesome because people forget that the Pentium M existed in between them

 

[GammaLaser]

Other than the IGP improvements, the drop in system power consumption under load compared to the 2600K was probably the most impressive result. IB will be great for notebooks.

 

[Khato]

It looks to me that performance of HD Graphics 4000 is geometry limited. GPUs with low geometry performance perform better in higher resolutions/settings since the bottleneck is elsewhere, but in lower resolutions and settings, geometry performance starts to become important.

Quite possible, but the only game demonstrating such behavior is Crysis. For all the other games run the HD 4000 shows a similar percentage drop in performance compared to the rest... and even on Crysis it's not that far off from the HD 6550D of Llano (1680x1050 at 82.2% of the 1366x768 fps for HD 4000 compared to 76.7% for HD 6550D.) Granted I only looked at IVB, Llano, the 5570, and the GT 440, but over all the resolution performance scaling seemed quite reasonable. Crysis is the only one where it's actually above the average - 82.2% vs average of 74% for Crysis, 69.5% vs average of 69.3% for Metro 2033, 64.7% vs average of 69% for Dirt 3, 85.7% vs average of 86.9% for Starcraft 2, and 72.3% vs average of 74.6% for Skyrim.

Anyway, encouraging results over all considering that it's not using the IVB driver. Hopefully we won't have to wait until release to see some power consumption figures while running those games? Though then again maybe it'd be best to wait on that 'til using the proper driver - I've always suspected that the high power draw of SNB integrated graphics compared to Llano was due to remnant issues with the driver.

 

[IntelUser2000]

Tocks are never as exciting as ticks I think most would agree. A die shrink, some new features, better thermals and overclocking.

All that said I'm grabbing one of these as soon as I can.

Hulk,

Tock=New architecture
Tick=Die shrink of an existing architecture

 

[Dadofamunky]

?!?What!?!? No overclocking? Garrrr... :twisted:

 

[cytg111]

Pretty much as expected, allthough not the 3x IGP increase.
While I got the upgrade itch, Haswell is sounding more and more likely.

 

[IntelUser2000]

Though then again maybe it'd be best to wait on that 'til using the proper driver - I've always suspected that the high power draw of SNB integrated graphics compared to Llano was due to remnant issues with the driver.

I meant more in a relative sense to the HD 3000. Given that each EUs are supposed to be 2x faster, and there's 33% more of them along with other changes like 2x texture unit, maybe geometry unit is not the only thing they need to ramp up. On Sandy Bridge they only needed 2x more EUs(same number of texture units) on the HD 3000 to outperform HD 2000 by 40-50%.

Then again maybe its not bad, as the 3770K's GPU is clocked lower than 2600K.

I don't believe drivers will do any meaningful amount, but if it does whatever I said could be thrown out the window. Sandy Bridge already brought nice increases via drivers, can they do it again?

I forgot to mention the CPU. Even with Turbo accounted for its 4-7%, and the increase is general, meaning it gains with everything. I don't think the desktop chips go off its max Turbo too often. That's a pretty nice gain. Only thing I'm disappointed about is the lack of stock clock increases.

 

[Piano Man]

Hi guys;

We've been working on a little something over the weekend that you might be interested in. I assume you all know what Ivy Bridge is?

http://www.anandtech.com/show/5626/ivy-bridge-preview-core-i7-3770k

Can you work a little bit more and get us some OC results with temps/noise as well

 

[meloz]

Wow, the i7 3770K is an amazing processor. In many benches (which do not make best use of more threads / cores) it bests the much more expensive SNB-E parts.

I was expecting a very conseravtive improvement on CPU side over i7 2600K. I am pleasantly surprised by the numbers.

For people using older than Sandy Bridge processors, this will be a most tempting upgrade.

 

[amenx]

I knew SB-e was a bad buy from the start, esp with the near arrival of IB. Its value is only as a niche product where specific mutli-threading scenarios are a priority.

 

[LtGoonRush]

I knew SB-e was a bad buy from the start, esp with the near arrival of IB. Its value is only as a niche product where specific mutli-threading scenarios are a priority.

If it had launched with 8-core CPUs as planned with the prices of the current 6-cores, and 6-cores in between that and the LGA-1155 Core i7s, the value proposition would be a lot better. Calling it now, Ivy Bridge-E will be groundbreaking, especially when we overclock it to 5Ghz on an 8-core chip.

 

[Diogenes2]

Calling it now, Ivy Bridge-E will be groundbreaking, especially when we overclock it to 5Ghz on an 8-core chip.

Which 8 core chip are you referring to ?

 

[Ferzerp]

Wow, the i7 3770K is an amazing processor. In many benches (which do not make best use of more threads / cores) it bests the much more expensive SNB-E parts.

You have highlighted the major flaw in Intel's current release cycle. Why buy an "-E" proc when you can wait 4 months and save a ton of money on something that will beat it everywhere except extremely threaded tasks.

 

[IntelUser2000]

If it had launched with 8-core CPUs as planned with the prices of the current 6-cores, and 6-cores in between that and the LGA-1155 Core i7s, the value proposition would be a lot better. Calling it now, Ivy Bridge-E will be groundbreaking, especially when we overclock it to 5Ghz on an 8-core chip.

Actually I don't believe Sandy Bridge E(not EP) was ever planned for 8 cores. With 8 cores, they would have needed to sacrifice frequency by 10-15% resulting in lower performance in most enthusiast-oriented applications compared to the Core i7 990X. That would have been bad. You see already in except the most threaded applications, the advantage is only 10-15%. Law of diminishing returns for consumer applications are hitting in full force here.

What could have been different is if it hadn't been delayed, and 3960X was released early Q4 last year. Then it would have had 6+ month lead time over Ivy Bridge LGA 1155 chips.

Its still probably worth it for Intel though, because they sell the Sandy Bridge E rebranded as Xeon E5 16xx. It would be a good single socket workstation, probably rivaling dual Westmere EP systems in HPC like scenarios.

 

[pauldun170]

What I got out of the article was a nagging sense I should wait for Haswell.
damn
At this rate I'm never going to upgrade my Q9550.

 

[iCyborg]

Actually I don't believe Sandy Bridge E(not EP) was ever planned for 8 cores. With 8 cores, they would have needed to sacrifice frequency by 10-15% resulting in lower performance in most enthusiast-oriented applications compared to the Core i7 990X. That would have been bad.

This kind of sacrifice is exactly what turbo was designed to help with though.

 

[philipma1957]

well I for one am glad I purchased the 2500k and the intel extreme mobo from mc at a steep discount.

Next stop haswell 2013 2 second boot times sound good to me.

 

[exar333]

Review was very impressive for mostly a process change. Lot's of promise for Haswell.

 

[Adam8281]

Tocks are never as exciting as ticks I think most would agree. A die shrink, some new features, better thermals and overclocking. Pretty cool but not that much of a thrill. But a tick brings pages and pages of new architectural improvements with lots of cool diagrams and hours of reading to try and figure it all out (for me anyway) before you see the actual IPC improvements in various applications in the benchmarks. We're always looking for the next release to be as dramatic a performance improvement as we witnessed in going from P4 to Conroe. That was nearly a doubling of performance per clock. It spoiled us.

All that said I'm grabbing one of these as soon as I can.

I think you've got your tocks and your ticks confused.

 

[pelov]

I don't believe drivers will do any meaningful amount, but if it does whatever I said could be thrown out the window. Sandy Bridge already brought nice increases via drivers, can they do it again?

This being Intel, I'm not expecting performance gains via drivers but at least the capacity to fix issues would be a good start. I think what makes people cringe when they hear "Intel IGP" is the bad rep they've received over the years for essentially no driver support whatsoever. HD4000 performs very well, but the monkey on their back isn't so much the performance but the consistency and reliability.

These chips look awesome. Thanks a ton for the preview. The performance-per-watt is incredible and the gains, though modest over previous generation, are actually more than I was expecting. I'm currently considering just giving up my desktop for a laptop replacement (a hefty laptop with a good GPU) so these should be right up my alley. I don't find myself gaming as much as I used to, so whoever wins the cost-to-performance race when both Trinity and IB arrives will be earning my money.

 

[Makaveli]

Why is Intel focusing on the freaking IGP? More CPU performance/cores, please!

Because that is where they are behind.

The average user doesn't need more cores!

 

[Makaveli]

I knew SB-e was a bad buy from the start, esp with the near arrival of IB. Its value is only as a niche product where specific mutli-threading scenarios are a priority.

After I read the review I thought the same thing.

Anyone that bought SB-e might be feeling little foolish right now unless of course do they have need for all 6 cores.

Kinda of reminds me of people that built 1366 rigs just before SB's was launched.

Review was very impressive for mostly a process change. Lot's of promise for Haswell.

This is what i'm waiting for I can get another year out of my system easy.