[Tom's Hardware] Sandy Bridge-E and X79 Platform Preview

[blackened23]

I already answered that above. Read.



I already answered this as well. No, it's not the fastest; it's just AS fast, but not faster. And yes, it is a status symbol. That couldn't be clearer. It being a "hobby" isn't the best reason for giving $400 free to Intel.

Just as fast in what. Games? EE processors are definitely faster in rendering / encoding. Not everyone games. Whether thats worth the price premium, is up to the user.

 

[LOL_Wut_Axel]

Just as fast in what. Games? EE processors are definitely faster in rendering / encoding. Not everyone games. Whether thats worth the price premium, is up to the user.

I think less than 50% of the people here read the thread before commenting. You probably didn't, so again: both stock, the performance difference is 2-3% because of the 100MHz higher clock speed on the EE. Both at average overclocks, say at 4.5GHz, will be the same speed.

At the same clock speeds, they'll be the same speed, including in rendering/encoding/content creation.

If someone can't OC because it's for enterprise or the like, then it's up to them to see if the 2-3% higher stock performance is worth the extra 67% in cost.

There's a reason why Tom's themselves said the 3930K was the most interesting CPU of the bunch.

 

[Edrick]

At the same clock speeds, they'll be the same speed, including in rendering/encoding/content creation.

So what are we betting that you are wrong?

 

[Shmee]

I think less than 50% of the people here read the thread before commenting. You probably didn't, so again: both stock, the performance difference is 2-3% because of the 100MHz higher clock speed on the EE. Both at average overclocks, say at 4.5GHz, will be the same speed.

At the same clock speeds, they'll be the same speed, including in rendering/encoding/content creation.

If someone can't OC because it's for enterprise or the like, then it's up to them to see if the 2-3% higher stock performance is worth the extra 67% in cost.

There's a reason why Tom's themselves said the 3930K was the most interesting CPU of the bunch.

Actually, this does make sense. 100MHz isn't a very big clock difference, so the only question than is how much difference does the cache make. We shall have to see. However, where the extra cache does not make much difference, his argument will still hold up; either they will achieve similar overclocks, or they will be kept at stock, where there is only 100 MHz difference.

Seems to me for professional computing purposes, the money saved by going with the 3930 is a no brainer, if they need more power, just add additional machines with the 3930, saving money. With a large farm, save thousands, that will pay for the extra computers to make up the 2-3% loss of performance. Now, it could be more lost, depending on apps, but unlikely anywhere near 67%. Still easy to make up in extra machines at lower cost/build.

 

[Edrick]

Actually, this does make sense. 100MHz isn't a very big clock difference, so the only question than is how much difference does the cache make. We shall have to see. However, where the extra cache does not make much difference, his argument will still hold up; either they will achieve similar overclocks, or they will be kept at stock, where there is only 100 MHz difference.

His arguement of saving money and going with the 3930 over the 3960 for most users is valid. No one is disputing that. But he also goes on to say that the 3930 and 3960 will have the exact same performance for every single application out there (clock or clock). Which is not true. Many applications love more cache. Just not games.

 

[StrangerGuy]

Actually, this does make sense. 100MHz isn't a very big clock difference, so the only question than is how much difference does the cache make. We shall have to see. However, where the extra cache does not make much difference, his argument will still hold up; either they will achieve similar overclocks, or they will be kept at stock, where there is only 100 MHz difference.

Seems to me for professional computing purposes, the money saved by going with the 3930 is a no brainer, if they need more power, just add additional machines with the 3930, saving money. With a large farm, save thousands, that will pay for the extra computers to make up the 2-3% loss of performance. Now, it could be more lost, depending on apps, but unlikely anywhere near 67%. Still easy to make up in extra machines at lower cost/build.

Well, 2600K has 2MB more L3 than 2500K yet it performs exactly at the same clocks without HT. Diminishing returns kick in even more at larger caches and more cores, so I'm pretty much expect the same pattern in S2011. Similarly, I won't expect any miracles from quad-channel DDR3 over dual when the SB's memory controller and caches are head and shoulders above Nehalem.

 

[Atreidin]

Assuming that there is not a single instance where more cache will ever make a difference is stupid. There must be at least one case where the processor with more cache will be faster. Whether or not the added cost is worth it is up to the user.

Assuming that you understand the hearts and minds of everyone who would ever buy that chip, by saying that they are ALL buying it as a status symbol, is being incredibly presumptuous and pompous.

 

[LOL_Wut_Axel]

So what are we betting that you are wrong?

I don't have to bet anything. I know I am right. You know it, too, but you love to argue.

If you're gonna continue arguing that it's still not here, it'll be here in two months. By then you'll see I'm right, and for obvious reasons because of how Sandy Bridge works.

 

[LOL_Wut_Axel]

His arguement of saving money and going with the 3930 over the 3960 for most users is valid. No one is disputing that. But he also goes on to say that the 3930 and 3960 will have the exact same performance for every single application out there (clock or clock). Which is not true. Many applications love more cache. Just not games.

Incorrect. This was true with Core 2, but Sandy Bridge is not cache starved at all and higher cache has very minimum effect, even when we get to low numbers like 1MB/core vs 1.5MB/core.

Celeron G540 (2MB L3 cache, 2.5GHz) vs. Pentium G620 (3MB L3 cache, 2.6GHz). 50% difference in L3 cache, 4% difference in clock speeds.

7% faster.

4% faster

4% faster

5% faster

5% faster

As you'll see, even getting down into only 1MB vs 1.5MB L3 cache-per-core, the difference in performance is easily explained by the 4% higher clock speed of the G620. The only application that presents a benefit from higher cache is WinRAR, but even then it's 3% only and the higher you go, the more diminshing the returns get. At 1.5MB vs 2MB L3 cache-per-core, the difference should already be reduced to 1%. After that, we start to look at numbers that are statistically insignificant and within the margin of error.

So no, cache makes very little difference in performance because Intel made it not depend on it.

 

[HendrixFan]

I don't have to bet anything. I know I am right. You know it, too, but you love to argue.

If you're gonna continue arguing that it's still not here, it'll be here in two months. By then you'll see I'm right, and for obvious reasons because of how Sandy Bridge works.

No, you keep framing your argument in a manner that suits you. Your basic premise of price to performance is one most anyone will agree with, even those who don't care about saving a lot of money for a little performance. As you have been put to task on your insistence that nobody should pay top dollar, that buyers shouldn't choose how to spend their own money, you have slid your argument into unprovables or into opinions that you masquerade as fact.

 

[LOL_Wut_Axel]

No, you keep framing your argument in a manner that suits you. Your basic premise of price to performance is one most anyone will agree with, even those who don't care about saving a lot of money for a little performance. As you have been put to task on your insistence that nobody should pay top dollar, that buyers shouldn't choose how to spend their own money, you have slid your argument into unprovables or into opinions that you masquerade as fact.

No, I frame it in a manner that's accurate. Even if you don't believe it with the proof I presented above, you'll see in two months. There's no "little" performance to be gained, which has been my main argument. It's the same, there's no "little".

 

[HendrixFan]

No, I frame it in a manner that's accurate. Even if you don't believe it with the proof I presented above, you'll see in two months. There's no "little" performance to be gained, which has been my main argument. It's the same, there's no "little".

Of course there will be many programs that have no benefit in additional cache. Showing benchmarks that prove that point doesn't show anything. There is a benefit to having a faster chip. It is faster. It may not be much faster, and it certainly won't scale well in relation to the increased price, but it will be faster.

Even if you only get a 1% gain from additional cache in a select few programs, then you are wrong. Even if you only get a 3% gain from the 3% clockspeed increase, then you are wrong. They have different clocks and different cache, they will have some performance difference.

Overclocking will always be luck of the draw, some chips go faster, some perform better at lower voltages, some do both, some do neither. The way CPUs are binned ensures that you will have a better chance at overclocking with a more expensive CPU. How much better? That certainly changes, even from week to week.

Going back to your earlier post, you claimed:

Why would you buy the 3960X over the 3930K when they're both binned the same? The only difference is that the 3930K had some unneeded L3 cache disabled and a meager 100MHz difference, probably to serve as an excuse to jack up the price 67%.

First off, they run at different clocks including in turbo mode. The fact that they have different clocks should be clear enough that they are binned differently, even if it is only a few chips in the 3930K range that can't clear the bar to the 3960X level. The chips are still being fabbed, yet you can somehow now what the chips look like coming off the wafers? Intel is still gathering data on the binning process as it happens. Where do you get your insider information from?

Secondly, the 3930K is listed at $583 versus $999 for the 3960X. While it is a considerable increase in price it is nowhere near 67%.

 

[LOL_Wut_Axel]

*snip*

No, there won't be. I already gave you examples with different desktop workloads. If you want to believe more than 2MB L3/core will do magical things for Sandy Bridge, you're simply wrong.

The higher cache/core you get, the lower the impact. Like I said earlier, at 1MB/core vs. 1.5MB, we're only seeing a 3% difference. Once you get higher, you rapidly get into diminishing points of return. By 2MB/core vs 2.5MB, it'll be statistically significant: less than 1%.

You could make an argument for servers wanting more cache, but for desktop workloads going into a quantity this high means no difference. These are also desktop CPUs, not server. If it's for your own or an enterprise's work machine and therefore you don't want to overclock, then it's up to you to see if the 2-3% performance difference between the 3930 and 3960 is worth it.

 

[Idontcare]

That water cooling unit is made by Asetek (same manufacturer as Corsair's water cooling solution). :hmm:

I have it on good authority that Intel is planning to release SB-E in two (or more) variants - one being no cooling included whatsoever (the buyer adds their own 3rd party cooler to their retail SB-E), the other being the already reported self-contained water-cooling gear but that it will contain something more than just a mere Intel logo as the value-add feature above and beyond buying the same kit from the OEM in question.

^