Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Daydream]

In Sandy Bridge land the score scaled very much linearly, 4.4GHz CPU - ~4400 points, 5GHz CPU - ~5000 points.

Does Ivy Bridge break that rule, or the aim remains to exceed 5.2-5.3GHz to top Jim? Knowing that might help to not attempt to fry other SB CPUs if Ivy Bridge has other tricks in store.

[JimH]

I've read that Ivy Bridge is 10% faster at the same clock speed.

Also, the program is getting faster.  5000 today is a little easier to achieve than it was a year ago.

Another advantage of Ivy Bridge is lower power consumption.

Where's babgvant when we need him?

[glynor]

I've read that Ivy Bridge is 10% faster at the same clock speed.

It depends on the specific mix of instructions that an application uses, but the Instructions Per Clock improvements in Ivy are more on the order of 4-6%.  Now, if you compare the "replacement" CPUs (the Sandy 2600K vs the Ivy 3770k, for example) at stock speeds, the improvements are greater, but that's also because of clockspeed differences.  The 3770k has a 100MHz advantage over the 2600k at stock.

The two biggest advantages of Ivy over Sandy are the integrated GPU (which is dramatically better, and is now within striking distance of AMD's latest platform IGP and low-end discreet cards), and the power consumption at load (the 3770k runs around 17-18W lower than the 2600k at peak).

If you already have a Sandy Bridge CPU, and you aren't using the integrated graphics (or it is meeting your needs as-is in GPU performance), then there is not a compelling reason to upgrade.  In fact, it seems that Sandy overclocks a bit more easily, so if you plan to overclock on air, you might do better with a Sandy.  You'd be much better served by waiting for the next "tock" (Haswell).  If you really, really need more multithreaded performance now than you can get on Sandy, you'd be well-served to check out a Sandy Bridge-E platform system instead (which has a whole host of other advantages, even though the design is "last gen" from the desktop perspective).

However, if you are still on an older-generation part (maybe a Core 2 or Nehalem based system), because you essentially get the benefit of both the IPC improvements in Sandy, and the power-usage improvements of Ivy, then it is a much more compelling upgrade.  That's why I was looking to upgrade my Core 2 Quad based HTPC system until my Lynnfield server died and I needed to act immediately.

[Sandy B Ridge]

For those with a Sandy/IvyBridge and a weak GPU, would it make sense to try and set up LAV decoding via QuickSync and Upscaling/Deinterlacing by MadVR via GPU. Discussion on Doom9 seem to point that this is possible to do by fooling it into a false VGA on the iGPU.

Anyone tried it in MC?

SBR

[InflatableMouse]

http://nl.hardware.info/nieuws/27706/warmteprobleem-ivy-bridge-door-koelpasta-bewezen#comments

Ivy Bridge has repeatedly turned out a worse overclocker than Sandy Bridge. People theorized this was due to the smaller dies and the heat concentrating on a smaller area.

Turns out Intel stopped using some special kind of soldering material between the die and the heatspreader and used some cheap ass thermal paste.

Removing the cover voids warranty obviously, but as you can see from the graphs (I know, text is dutch, sorry) replacing that cheap paste for something better can drop temperatures by as much as 15C.

From what I gather its still not as good as that special solder though .

[hulkss]

First off, Intel is not stupid.

As far as thermal compound goes, the issue is not the compound itself. There must be no gaps in the compound layer and the compound must be thin.
As the thickness approaches zero so does the thermal resistance.

Here is a reputable test showing that mayonnaise works as well as most advanced thermal compounds. Why? The layer of compound (mayo) must be very thin.
http://www.hardwaresecrets.com/article/Thermal-Compound-Roundup-February-2012/1490/5

To get a thin compound layer, proper clamp uniformity, stiffness, and force is required.

Some info from Intel http://www.intel.com/content/dam/www/public/us/en/documents/design-guides/3rd-gen-core-lga1155-socket-guide.pdf

Intel recommends 70 to 135 pounds of clamp force to hold the processor in the socket and up to 50 pounds of clamp force between the cpu and heatsink (cooler).

Mechanical variations in the thermal interface material layer is what causes all the variation in these temperature studies.

Here is a report where a guy "de-lidded" the Ivy Bridge CPU and mounted a cooler directly on the CPU chip. Result - no improvement.
http://www.maximum-tech.net/ivy-bridge-heat-problem-persist-even-after-ihs-removal-12404/
http://www.overclock.net/t/1249419/pcevaluation-intel-i7-3770k-temperature-measured-without-ihs

[Hendrik]

The 22nm process is not 100% perfect yet, it'll probably improve over time.
So watch out for new steppings to show up, which indicate a change in the manufacturing process.

[InflatableMouse]

First off, Intel is not stupid.

As far as thermal compound goes, the issue is not the compound itself. There must be no gaps in the compound layer and the compound must be thin.
As the thickness approaches zero so does the thermal resistance.

Here is a reputable test showing that mayonnaise works as well as most advanced thermal compounds. Why? The layer of compound (mayo) must be very thin.
http://www.hardwaresecrets.com/article/Thermal-Compound-Roundup-February-2012/1490/5

To get a thin compound layer, proper clamp uniformity, stiffness, and force is required.

Some info from Intel http://www.intel.com/content/dam/www/public/us/en/documents/design-guides/3rd-gen-core-lga1155-socket-guide.pdf

Intel recommends 70 to 135 pounds of clamp force to hold the processor in the socket and up to 50 pounds of clamp force between the cpu and heatsink (cooler).

Mechanical variations in the thermal interface material layer is what causes all the variation in these temperature studies.

Here is a report where a guy "de-lidded" the Ivy Bridge CPU and mounted a cooler directly on the CPU chip. Result - no improvement.
http://www.maximum-tech.net/ivy-bridge-heat-problem-persist-even-after-ihs-removal-12404/
http://www.overclock.net/t/1249419/pcevaluation-intel-i7-3770k-temperature-measured-without-ihs

Yeh I know all that about the thermal compound and I know Intel isn't stupid too .

That last link is interesting though because it contradicts the Japanese test 100%. Makes me wonder what those Japanese have been doing, and why.

[glynor]

Ivy Bridge has repeatedly turned out a worse overclocker than Sandy Bridge. People theorized this was due to the smaller dies and the heat concentrating on a smaller area.

The thermal changes are certainly all having an effect of some kind, but I'm not convinced at all that the thermal changes Intel made were detrimental compared to the "Sandy Bridge" method.  In fact, I'd guess that it is equally likely that Intel changed the process to improve thermal efficiency as they did to save cost.  They aren't talking, so the only way to find out is to do a scientific study of the new process, which would be very difficult without access to Intel's fab.

Ivy certainly doesn't overclock the same way Sandy Bridge did.  But that doesn't mean it "isn't a good overclocker".  It means it is different (read the Anandtech article I linked above).  Some people are getting dramatic overclocks on them, so it isn't impossible (it might be impossible to match Sandy on air though).

I don't think there is any one answer to why Ivy is so sensitive to voltage:

1. Smaller thermal footprint for cooling.
2. More dense heat production due to the new "3d Transistor" process Intel used at 22nm.
3. Because Ivy is a "tick" in Intel's development cadence, the underlying architecture wasn't designed for this new 22nm process (it is mostly a process-shrunken Sandy) which means it might not be optimized for leakage on 22-nm (or other issues).
4. They might have cheaped out a bit on the thermal design, since at stock voltage Ivy runs so cool (most people don't overclock).
5. Intel's 22-nm process is young and isn't yielding optimal chips yet.

In fact, all or any of those things might have a lot to do with why Ivy was delayed!  Intel said they were having yield issues.  Perhaps now we know one result of those problems.