Oh hell yeah. Even at just the PnR stage alone, there's a ton of variability - do you want to optimize your clock and buffer trees a certain way, macro placement, non-default routing rules used, etc - that _will_ impact yield and performance.
Oh hell yeah. Even at just the PnR stage alone, there's a ton of variability - do you want to optimize your clock and buffer trees a certain way, macro placement, non-default routing rules used, etc - that _will_ impact yield and performance.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.
Heh, i have just remembered this,
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.
So would this be a fair estimate? 2600K @ 4.8Ghz = 3770K @ 4.5Ghz = 4770K @ 4.2Ghz
Do you have a link to other SB results?
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%.
http://forums.anandtech.com/showthread.php?t=2057154&page=3
First two line up with his, and I have confirmed Nehalem is 6.5.
All I know is my 3770k at 4.4Ghz is FAST!!!!!!! If a 4.2Ghz Haswell is equal to a 4.62Ghz IB it is smoking fast!
merom @ 2.1 = 7.23
i7 920 @ 3.8 = 6.4716
deneb @ 2.0ghz = 7.54
FX8350 @ 4600 = 8.88 its an esxi guest so their is extra overhead
wow, your easy to please..............
FX8350 @ 4600 = 8.88 its an esxi guest so their is extra overhead
Here's some straight-up FX8350 numbers.
The thing with CPUmark99, being a single-threaded benchmark, is you really have to disable the turbo-core/boost if you want consistent results across platforms and users.
...
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.
10% ontop of Ivy when Ivy is already a little faster than sandy = at least an upgrade if you are a sandy user. We need to wait for OC results in a legit review.
Doesn't Intel Turbo Boost practically stay in turbo when the CPU is loaded the whole time? For example if only 1 thread is being loaded on an i7 3930k for 5 minutes, wouldn't a 3930k run at 3.8Ghz for the whole 5 minutes provided that the CPU temperatures is does not reach it's limits since 3.8GHz is the official turbo clock for single core turbo on the 3930k?
Yup. And better if you are on Nehalem. I'll probaby break and go with a Haswell platform..
Prescott 31 stage pipeline, less IPC than Northwood. Branch mispredictions were painful with Prescott. And HT is of no use with this bench. The scores are correct. The benchmark is old and has limitations of course.
Also keep in mind this tests integer performance only and is single core. Lots of SB improvements came from additional instructions, better FP, larger/faster caches. All that doesn't matter for this little Int bench that fits in L1.
When overclocking a processor, we can expect a kind of linear trend in performance. As Northwood's speed increases, its performance increases. The same is true for Prescott , but what is important to look at is increase in performance compared to increase in clock speed.
Prescott 's enhancements actually give it a steeper increase in performance per increase in clock. Not only can Prescott be clocked higher than Northwood, but as its clock speed is increased, it will start to outperform similarly clocked Northwood CPUs.
We can even see this trend apparent in our limited 3 clock speed tests. Most of the time, the 2.8GHz Northwood outperforms the 2.8GHz Prescott, but the percentage by which Prescott is outperformed decreases as clock speed increases, meaning that the performance delta is significantly less at 3.2GHz.
Here's some straight-up FX8350 numbers.
The thing with CPUmark99, being a single-threaded benchmark, is you really have to disable the turbo-core/boost if you want consistent results across platforms and users.
Point is you can't be sure, it is dependent on uncontrolled factors including the ambient, cooling solution, mobo/bios, and end-user settings.
If you are trying to reduce performance to a "per GHz" metric for IPC comparisons then you better be doing all you can to ensure the "GHz" is actually the correct one.
Whether it is manually fixed at 3.8GHz (turbo-boost), or manually fixed at 3.2GHz (base), it doesn't matter so long as the potential for it being a variable has been eliminated.
Leaving it up to turbo-boost and power-management means you aren't controlling the very parameter you are interested in characterizing (IPC).
Well there would be no reason to disable turbo boost unless you are trying to compare performance per a clock speed in benchmarks. I mean who disables turbo boost in real computing situations. Disabling turbo boost on my i7-3930k would cause a big loss in performance. For example in a single threaded program my CPU could be running at 3.8GHz, or at least much faster than with turbo boost disabled if not at 3.8GHz the whole time, and even in a dual threaded program it still can run at 3.8GHz but not stick at 3.8GHz as often mainly because something running in the background. The base clock is only 3.2GHz on my CPU and I consider turbo boost a very important feature, especially for lightly threaded programs and it could be the difference between my CPU running at 3.2GHz or 3.8GHz.
Not comparable. On GPUs you can add more and more shaders every year and we get better performance in games which doesn't work on CPUs. The current Core architecture already is on a high level IPC and high frequency too. Big gains are unrealistic. For Skylake I expect 6 cores for the mainstream. But you need the software, not everything gets a speedbump from more cores.
10% ontop of Ivy when Ivy is already a little faster than sandy = at least an upgrade if you are a sandy user. We need to wait for OC results in a legit review.
It went the way of the dodo. Optimistically speaking, even if Haswell has 10% better IPC than IVB and hits 5GHz easily :Whatever happened to computer power doubling every 18 months?