Intel Ice Lake CPU in 2019?

[tgk300]

Apparently, earlier today Intel confirmed that it is already working on a successor to their 8th Generation CPU's (Coffee Lake) called "Ice Lake". It is unclear whether they will be 4, 6, 8 or 10 core CPU's as Intel is yet to release any details in terms of CPU specifications apart from the fact that it will be based on a 10nm+ dye. 

 

Read the Article Below from Tech Radar:

http://www.techradar.com/news/intel-reveals-first-scant-details-on-future-ice-lake-processors

[Shimejii]

2nd post seems a bit weird  This wont be known for another year and a bit more then likely

[tgk300]

Just now, Shimejii said:

2nd post seems a bit weird  This wont be known for another year and a bit more then likely

Explain what you mean't by second post a bit weird?

[Shimejii]

1 minute ago, tgk300 said:

Explain what you mean't by second post a bit weird?

Making an account just to post this kind of stuff always seems to be weird and fishy. Generally ends up being just spam or some other stuff.

[tgk300]

Just now, Shimejii said:

Making an account just to post this kind of stuff always seems to be weird and fishy. Generally ends up being just spam or some other stuff.

I ain't spamming, i created this account yesterday, came across this article today so decided to share it.

[ScratchCat]

Interesting, can't wait for the ext 2% IPC boost

[FloRolf]

8 minutes ago, ScratchCat said:

Interesting, can't wait for the ext 2% IPC boost

Just use a C2D 2 weeks before icelake launch and you'll feel the power of 2% boost

[PianoPlayer88Key]

12 hours ago, FloRolf said:

Just use a C2D 2 weeks before icelake launch and you'll feel the power of 2% boost

What happened to the days of bigger IPC boosts?

 

Sandy Bridge was a fairly good jump over Westmere, wasn't it?

I think the 286 was 100% better (or more) than the 8086.  Wikipedia also says the original Pentium was about 100% faster than the 486, which in turn had a similar gain over the 386.

 

I understand IPC gain to basically mean improved performance at a given clock speed and thread count.  For example, making up theoretical Cinebench R15 benchmarks:

• CPU A: 160 @ 4 GHz, 1 thread.  CPU B: 320 @ 4 GHz, 1 thread.  IPC boost = 100%.
• CPU C: 800 @ 3.6 GHz, 8 threads.  CPU D: 1200 @ 3.6 GHz, 12 threads.  IPC boost = 0%.

And there's also the big improvements in price to performance (over a given time period) seen 20-30 years ago.

[porina]

9 minutes ago, PianoPlayer88Key said:

What happened to the days of bigger IPC boosts?

They way I look at it is that a given instruction takes a certain amount of time to execute. If that instruction has been around from the start of x86, it's going to be pretty well optimised by now. There might be minor benefits from removing bottlenecks allowing it to reach full potential, but I consider it "done".

 

Where we do gain big IPC jumps is when new instructions are introduced to do specific tasks more efficiently. The drawback here is you have to wait for software to catch up. For Prime95-like compute tasks Sandy Bridge gave about 100% peak IPC improvement over what came before due to AVX. Haswell added another 50% through AVX2. Skylake got around 14% improvement, I think this is due at least in part to a reduction in execution cycle time from 5 to 4 cycles. Skylake-X now has AVX-512 and in theory support for it would give another 100% improvement, although the code isn't there yet. Note I called these peak IPC improvements, since depending on the rest of the system configuration (in particular ram speed) performance may be limited below that.

[Taf the Ghost]

23 minutes ago, PianoPlayer88Key said:

What happened to the days of bigger IPC boosts?

 

Sandy Bridge was a fairly good jump over Westmere, wasn't it?

I think the 286 was 100% better (or more) than the 8086.  Wikipedia also says the original Pentium was about 100% faster than the 486, which in turn had a similar gain over the 386.

 

I understand IPC gain to basically mean improved performance at a given clock speed and thread count.  For example, making up theoretical Cinebench R15 benchmarks:

• CPU A: 160 @ 4 GHz, 1 thread.  CPU B: 320 @ 4 GHz, 1 thread.  IPC boost = 100%.
• CPU C: 800 @ 3.6 GHz, 8 threads.  CPU D: 1200 @ 3.6 GHz, 12 threads.  IPC boost = 0%.

And there's also the big improvements in price to performance (over a given time period) seen 20-30 years ago.

We're not going to see gains like the 90s anymore. Most of the improvement came from outright 100% or higher frequency bumps.

 

We went from 33 Mhz to 1 Ghz in less than an decade. We're not going to go from 4 Ghz to 120 Ghz in another decade because of Physics.

[PianoPlayer88Key]

1 hour ago, porina said:

They way I look at it is that a given instruction takes a certain amount of time to execute. If that instruction has been around from the start of x86, it's going to be pretty well optimised by now. There might be minor benefits from removing bottlenecks allowing it to reach full potential, but I consider it "done".

 

Where we do gain big IPC jumps is when new instructions are introduced to do specific tasks more efficiently. The drawback here is you have to wait for software to catch up. For Prime95-like compute tasks Sandy Bridge gave about 100% peak IPC improvement over what came before due to AVX. Haswell added another 50% through AVX2. Skylake got around 14% improvement, I think this is due at least in part to a reduction in execution cycle time from 5 to 4 cycles. Skylake-X now has AVX-512 and in theory support for it would give another 100% improvement, although the code isn't there yet. Note I called these peak IPC improvements, since depending on the rest of the system configuration (in particular ram speed) performance may be limited below that.

The bolded part reminds me of another thing.  When I do see IPC jumps, I want them to benefit already-existing software.  Sometimes there are things that older software does well, or user interfaces I like, or other factors, and sometimes newer software is a bit of a step back for me.  (I understand some businesses, etc, are even more hardcore with the "if it ain't broke, don't fix it" mentality - aren't there some places that were recently still using Windows 3.1?)

 

For example, the same pre-existing build of Blender or Handbrake, when run on a newer CPU with 100% better IPC, should complete the same project (at the same settings) in half the time as it took on the older CPU running at the same clock speed and core/thread count.

 

1 hour ago, Taf the Ghost said:

We're not going to see gains like the 90s anymore. Most of the improvement came from outright 100% or higher frequency bumps.

 

We went from 33 Mhz to 1 Ghz in less than an decade. We're not going to go from 4 Ghz to 120 Ghz in another decade because of Physics.

I didn't mean gains in clock speeds, I meant gains in performance at the SAME clock speed.  I screenshotted parts of a few wikipedia articles, highlighted a few things, and sized the windows as necessary.  Also included is an abridged table (which I didn't save) from the instructions per second page, timeline of instructions per second section.

On the table, focus more on the "instructions per clock cycle per core" column instead of the one to its left, at least at the bottom of the chart where there's a few multi-core CPUs.  I removed some CPUs with SMT/HT before I made the screenshot (as I think the chart wasn't taking that into account properly), as well as everything non-x86.

 

Also, take a look at the 286 in the table on the right, and compare it to the Intel DX4.  Then, look at the 2 photos below.

There's a dramatic increase in price to performance over that time. (Unfortunately I don't have the price for the 286-10 CPU by itself - I'm guessing around $300-400(?), but I DO have the price for the 486 motherboard and RAM & CPU Cooler, and a 1.44MB floppy drive purchased a little earlier than the 486 parts.)  In the time since Sandy Bridge launch, I have seen nothing remotely close to improvements like that.

 

 

In modern terms, if the i7-7700K had a 100% IPC gain over the i7-6700K, it would do about 1,686 or so in Cinebench R15 multi-thread, and about 334 in Single Core mode, at 4 GHz.  (Not at 8 GHz, if so, that would be no IPC improvement.  Not that any 7700Ks hit 8 GHz even under LN2, anyway.)

[Taf the Ghost]

@PianoPlayer88Key

 

Solo CPUs were really, really rare in the 286 period. 386 was really the first to be sort of available. 386 & 486 were still being sold right when the Pentium dropped, but Windows 95 changed everything.  Then the PII was the game-changer for popular adoption.

 

So there's a difference between Intel's side they sold them for and some very wild costs to the end user.