Intel Skylake / Kaby Lake

[DooKey]

With AMD very strongly in the game, Intel may not have a choice but to follow their lead on the structure.

Seeing other comments, I can't understand why anyone would defend Intel these past years if Intel is supposed to be the tech and caring leader.

What the heck is caring leader? None of these tech companies care one whit about any of us. I'm always confused by this kind of meme. Are people like you actually saying a company cares about you?

 

[formulav8]

gasme very strongly, Intel

What the heck is caring leader? None of these tech companies care one whit about any of us.

You get it then. Some person's ACT like a company care's about them. I've personally, never felt that way with any company caring for me. Some would probably defend Intel or AMD till their death it seems sometimes. Just IMO.

 

[MarkizSchnitzel]

Well, depends on what you mean by "company".
Accounting and investors sure as hell do not. They do want you to care for them, so they make the appearance of caring.

On the other hand, to my eye, there are genuine enthusiasts like Miyamoto who do indeed care.

i doubt though that can ever be the case in this sort of business

 

[SpoCk0nd0pe]

Since they've increased the L2 cache to 1MB they had to lower L3 cache to keep the die size in check since its on the same 14nm process from their previous HEDT.

Yes... but cutting the L3 cache size in half?

 

[LTC8K6]

Yes... but cutting the L3 cache size in half?

But with 4X the L2 cache.

Take your pick...

 

[tamz_msc]

But with 4X the L2 cache.

Take your pick...

Everything has tradeoffs.

 

[dullard]

Core i9-7820X
Is it just me or is that a disappointingly small cache size?

Yes... but cutting the L3 cache size in half?

• Intel 68xxK models were 1.5 MB of L2 cache and 15 MB of L3 cache. 16.5 MB total.
• Intel 78xxX models are 6 MB (7800X) or 8 MB (7820X) of L2 cache and 11 MB of L3 cache. 17 MB or 19 MB total, depending on the chip. Assuming the rumored specifications are correct.

The next generation has more total cache and a greater portion of that cache is faster L2. It is an improvement in the statistics that matter. I'm not sure that I would call that a problem. Nor would I say it is cutting L3 in half, since 11 is not half of 15. You might have wanted more and are thus disappointed. But the cache technically is much better.

Lets look at the next step up:

• Intel 6900K had 2 MB of L2 cache and 20 MB of L3 cache. 22 MB total.
• vs.
• Intel 7900X is rumored to have 10 MB of L2 and 13.75 MB of L3. 23.75 MB total.
• Again, more total cache and the portion of faster L2 cache is much higher.

Lets look at the next step up:

• Intel 6950K had 2.5 MB of L2 cache and 25 MB of L3 cache. 27.5 MB total.
• vs.
• Intel 7920X is rumored to have 12 MB of L2 and 16.5 MB of L3. 28.5 MB total.
• Again, more total cache and the portion of faster L2 cache is much higher

 

[tamz_msc]

• Intel 68xxK models were 1.5 MB of L2 cache and 15 MB of L3 cache. 16.5 MB total.
• Intel 78xxX models are 6 MB (7800X) or 8 MB (7820X) of L2 cache and 11 MB of L3 cache. 17 MB or 19 MB total, depending on the chip. Assuming the rumored specifications are correct.

The next generation has more total cache and a greater portion of that cache is faster L2. It is an improvement in the statistics that matter. I'm not sure that I would call that a problem. Nor would I say it is cutting L3 in half, since 11 is not half of 15. You might have wanted more and are thus disappointed. But the cache technically is much better.

Lets look at the next step up:

• Intel 6900K had 2 MB of L2 cache and 20 MB of L3 cache. 22 MB total.
• vs.
• Intel 7900X is rumored to have 10 MB of L2 and 13.75 MB of L3. 23.75 MB total.
• Again, more total cache and the portion of faster L2 cache is much higher.

Lets look at the next step up:

• Intel 6950K had 2.5 MB of L2 cache and 25 MB of L3 cache. 27.5 MB total.
• vs.
• Intel 7920X is rumored to have 12 MB of L2 and 16.5 MB of L3. 28.5 MB total.
• Again, more total cache and the portion of faster L2 cache is much higher

All the corresponding parts have increased their core count by two. So total cache at a given number of cores has gone down.

 

[dullard]

All the corresponding parts have increased their core count by two. So total cache at a given number of cores has gone down.

That is like going from an i5 to an i3 and saying, hey there is less cache. While true, you are now comparing apples to oranges. You should compare one chip to the chip that is replacing it.

 

[DrMrLordX]

That is like going from an i5 to an i3 and saying, hey there is less cache. While true, you are now comparing apples to oranges. You should compare one chip to the chip that is replacing it.

You have that a bit backwards. It's like we're going from an i3 to an i7.

 

[dullard]

You have that a bit backwards. It's like we're going from an i3 to an i7.

Going back to the start of this topic, it was about the 7820X. The 7820X has 19 MB of cache. It is replacing the 6850K with 16.5 MB of cache. The amount of cache increased.

tamz_msc essentially wanted to compare the 78xx line to the 6900K chip just because the core count was the same. Call it i3 to i7 or i5 to i3 comparison, it doesn't matter. What matters is that he is comparing one category to a different category at a whole different price point.

 

[tamz_msc]

Going back to the start of this topic, it was about the 7820X. The 7820X has 19 MB of cache. It is replacing the 6850K with 16.5 MB of cache. The amount of cache increased.

tamz_msc essentially wanted to compare the 78xx line to the 6900K chip just because the core count was the same. Call it i3 to i7 or i5 to i3 comparison, it doesn't matter. What matters is that he is comparing one category to a different category at a whole different price point.

You're essentially saying that the second digit of the name differentiates between some feature, while in reality it doesn't.
Core for core, total cache has reduced, and the 8 core part has been relegated to 28 PCIe lanes.

If you're directly comparing 6900K vs 7900X, then the increased cache is obviously due to the two extra cores. What's so special about that?

 

[dullard]

You're essentially saying that the second digit of the name differentiates between some feature, while in reality it doesn't.
Core for core, total cache has reduced, and the 8 core part has been relegated to 28 PCIe lanes.

If you're directly comparing 6900K vs 7900X, then the increased cache is obviously due to the two extra cores. What's so special about that?

Um, the second digit is THE important feature in Intel's Core chips right now.

Try going from a 7700 to a 7600. You lose hyperthreading and 2 MB of cache.
Try going from a 7700 to a 7300. You lose two cores and 4 MB of cache.

The second digit describes the features and price point. So you need to compare the second digit when comparing processors.

 

[frozentundra123456]

That is like going from an i5 to an i3 and saying, hey there is less cache. While true, you are now comparing apples to oranges. You should compare one chip to the chip that is replacing it.

No, I what one needs to look at is the cache per core. The cpu obviously doesn't know what chip it is replacing, the workload is split up among the cores and cache that are physically on the chip. That said, since the structure is different, I think trying to compare to older chips is pointless at this time. We will just have to wait for actual performance figures.

 

[tamz_msc]

Um, the second digit is THE important feature in Intel's Core chips right now.

Try going from a 7700 to a 7600. You lose hyperthreading and 2 MB of cache.
Try going from a 7700 to a 7300. You lose two cores and 4 MB of cache.

The second digit describes the features and price point. So you need to compare the second digit when comparing processors.

No, those are more arbitrary when it comes to the HEDT.

5930K vs 5820K - second digit differentiates between no. of PCIe lanes.
6850K vs 6800K - now it's the third digit that does the same.

Intel never laid out a naming scheme that allowed for consistently meaningful comparison.

 

[DrMrLordX]

tamz_msc essentially wanted to compare the 78xx line to the 6900K chip just because the core count was the same.

Why wouldn't he? Forget model numbers, just look at core count and cache sizes.

Intel has reduced L3 to compensate for the increase L2 size and its effect on die size/transistor count. We don't know what will be the impact on performance.

 

[dullard]

Why wouldn't he? Forget model numbers, just look at core count and cache sizes.

Intel has reduced L3 to compensate for the increase L2 size and its effect on die size/transistor count. We don't know what will be the impact on performance.

Because they aren't in the same target market. The price difference is almost double. Forget core count and look at prices.

 

[dullard]

No, those are more arbitrary when it comes to the HEDT.

5930K vs 5820K - second digit differentiates between no. of PCIe lanes.
6850K vs 6800K - now it's the third digit that does the same.

Intel never laid out a naming scheme that allowed for consistently meaningful comparison.

You are correct that the numbers are a bit more arbitrary when it comes to HEDT. But your PCI lane count thought only works on that one generation. For example, the 4930K and the 4820K both have 40 PCI lanes. Neither the second nor the third digit is a useful rule for the number of PCI lanes in HEDT chips.

http://ark.intel.com/products/77780/Intel-Core-i7-4930K-Processor-12M-Cache-up-to-3_90-GHz
https://ark.intel.com/products/77781/Intel-Core-i7-4820K-Processor-10M-Cache-up-to-3_90-GHz.

What Intel has done though is differentiate on price. The higher the number (within a generation), the more features and the more the price. For HEDT, Intel has always had a chip near $1000, one near $600, one near $300 to $400. And recently they have a chip near $1700.

 

[RichUK]

You'd think increasing the L2 cache would be a good move, considering it's faster than L3 cache.

Simplistically speaking.

 

[Sweepr]

I have an eye on that Core i9-7820X. High-clocked 8C/16T goodness with Skylake+ IPC and 4-channel DDR4. Skylake-X is the spiritual successor to Bloomfield IMHO.

Only 15 days till formal launch if the rumors are correct.

 

[SpoCk0nd0pe]

I really don't care how version numbers compare. What I do care about is that they have cut the amount of L3 cache per core in half (you can keep the 0.125 mb).

 

[DrMrLordX]

Forget core count and look at prices.

Um . . . no? I just want to know if more l2 will compensate for the loss of L3, that's all. If it doesn't then the expected increase in IPC may not emerge, giving Intel a small problem (which they will compensate for partially with advanced clockspeed vs Broadwell-E).

 

[pantsaregood]

Um . . . no? I just want to know if more l2 will compensate for the loss of L3, that's all. If it doesn't then the expected increase in IPC may not emerge, giving Intel a small problem (which they will compensate for partially with advanced clockspeed vs Broadwell-E).

Initially, the answer appears to be "the larger L2 should easily compensate for the lack of L3 cache due to the considerably lower latency," but there's a bit more complexity.

Unless Intel has significantly altered their cache model, the L3 cache is still inclusive. This means that everything that exists in L2 cache on a given core also has to exist in L3 cache - this means that for every 1 MB of L2 cache, there is only 1.375 MB of L3 cache. Compare this to Broadwell-E, which had 0.25 MB of L2, there existed 2.5 MB of L3 cache.

This effectively means that 72.7% of Skylake-X's L3 cache could be used for mirroring L2 cache, where Broadwell-E only used 10% of its L3 for mirroring L2 cache.

Does this mean cache will be slower? No. We need benchmarks.

 

[shortylickens]

Stop worrying about level 3 cache.
Level 3 cache is no better than RAM. And you really dont need an extra megabyte of RAM.

 

[Topweasel]

Initially, the answer appears to be "the larger L2 should easily compensate for the lack of L3 cache due to the considerably lower latency," but there's a bit more complexity.

Unless Intel has significantly altered their cache model, the L3 cache is still inclusive. This means that everything that exists in L2 cache on a given core also has to exist in L3 cache - this means that for every 1 MB of L2 cache, there is only 1.375 MB of L3 cache. Compare this to Broadwell-E, which had 0.25 MB of L2, there existed 2.5 MB of L3 cache.

This effectively means that 72.7% of Skylake-X's L3 cache could be used for mirroring L2 cache, where Broadwell-E only used 10% of its L3 for mirroring L2 cache.

Does this mean cache will be slower? No. We need benchmarks.

So if I am reading this correctly the Skylake now only has lets use the 7920x model only has 16.5 MB of unique cache between the L2 and L3 and the 6950x has 25MB of unique cache? So the hope here would be that the extra speed on the L2 make up for losing about 10MB worth of extra data total which in theory would mean more wasted cycles on refreshing the cache from memory?