team blue Ice Lake Details Released

[jdubya421]

6 hours ago, RejZoR said:

Manufacturing process doesn't inherently improve IPC. It gives space to make architectural changes to achieve IPC boost, but without architectural changes you can downsize it to single atom and it wouldn't make it any more efficient per clock.

Not even true. They can get different sputtering and trimming techniques that could decrease loss, timing, or any other factor. As the process matures the performance can increase. As the process matures materials can change as well, they can find defects with a current material or improvements in a new one, different dopants etc. I work in the semiconductor industry, it happens all the time. 

[porina]

8 minutes ago, jdubya421 said:

Not even true. They can get different sputtering and trimming techniques that could decrease loss, timing, or any other factor. As the process matures the performance can increase. As the process matures materials can change as well, they can find defects with a current material or improvements in a new one, different dopants etc. I work in the semiconductor industry, it happens all the time. 

I think there is a misunderstanding here. The original claim that sparked it off was that a note shrink, by itself, improves IPC. Instructions Per Clock. It just can't. It is a logic level design function. What process may change is that you could get more clock, but that isn't improving IPC.

[jdubya421]

58 minutes ago, porina said:

I think there is a misunderstanding here. The original claim that sparked it off was that a note shrink, by itself, improves IPC. Instructions Per Clock. It just can't. It is a logic level design function. What process may change is that you could get more clock, but that isn't improving IPC.

Oh, by "manufacturing process" do you just mean technology size? I.e. the shrink down to 10nm?

[porina]

28 minutes ago, jdubya421 said:

Oh, by "manufacturing process" do you just mean technology size? I.e. the shrink down to 10nm?

At least in the enthusiast community, yes. Process node. 

[dtaflorida]

18 hours ago, floofer said:

Lets go back to 90nm pentium 4 then. 

Ah, good old Netburst, back when Intel chased high clocks instead of improved architecture. 

[Tristerin]

18 hours ago, Genwyn said:

 adequately compete with ryzens midranged options would be great for the market

This is what I would like to see - during the great RAM crisis I had to settle for a 2133mhz kit that met my budget at the time.  Ive since replaced it out and have a case, PSU, GPU, RAM just sitting around but I didn't want to build a budget Ryzen because 2133mhz just doesn't play well with it, compared to intel.  Im hoping something cheap, but okay performance comes out of this so I can apply this RAM and spare parts and make a used/new build to sell that can perform decently and not be bottlenecked by that kit.

[RejZoR]

12 minutes ago, dtaflorida said:

Ah, good old Netburst, back when Intel chased high clocks instead of improved architecture. 

Remember those claims how Netburst architecture will be good up to 20GHz I believe? Lol, yeah, it's been 20 years since and we're just going past 5GHz now (if we exclude Bulldozers).

[Mira Yurizaki]

39 minutes ago, dtaflorida said:

Ah, good old Netburst, back when Intel chased high clocks instead of improved architecture. 

I don't blame them though. Historically high clock speeds do contribute more to performance than architecture improvements alone.

 

For example, to poke at a benchmark result:

 

Between the i7-9700K and i7-2600K (stock speed), there's about a 70% improvement in a single threaded test. However, a singificant portion of this performance increase is simply because the 2600K runs at 3.8 GHz max turbo while the 9700K runs at 4.9GHz. This should account for around a 29% overall performance improvement. And if you normalize the 9700K to the 2600K speeds, you get a value that's about 31% better. So it took Intel seven generations to achieve 31% better IPC. Although I suppose technically there were only three microarchitectures throughout this period, so I guess technically, it only took three generations to achieve 31% better IPC. And looking back at various reviews of older processors, when normalized, jumps between generations tends to be about 10%-15% better IPC.

 

As another point of interest, the Pentium III had a 500 MHz model. Not even a year later, the same implementation had a 1GHz model.

 

Processors that we use in general purpose computers really spend more time trying to rearrange instructions in such a way that makes the most efficient use of a simple execution engine. And there's only so much reshuffling of instructions you can do before it takes more effort to do than it's worth.

[vanished]

22 hours ago, gutz00 said:

https://www.anandtech.com/show/14664/testing-intel-ice-lake-10nm

 

Intels Next Gen Processors code named Ice Lake have been benchmarked. 

Do you think they can compete with third gen Ryzen?

This is the laptop line up, is it not?  So it's not competing with third gen Ryzen, only Zen+, and fo that reason I think this just might be a segment where they still have a chance until zen 2 chips make it into laptops, bit as for everything else they've already lost.  Server cares about security, efficiency, and multicore performance, which they've destroyed intel in (all while being significantly cheaper too).  Desktops care about high single threaded performance with good multithreaded to back it up, and they've accomplished that as well, again, while consuming much less power.  I personally don't see how intel can possibly come back at them.  If they don't node shrink, they won't improve.  If they do, they'll probably lose clock speed, or best case, keep it the same, and that's assuming they keep core count the same, which they really can't afford to do since AMD now has them beat significantly in that department.  If they increase core count, they'll definitely lose clock speed, increase power consumption even more, and also cost, and it'll be a laughing stock.

22 hours ago, gutz00 said:

Personally I don't think they can. It's great to see Intel hit 10nm but we see a dip in Base Frequency. Personally not a trade off I'd like to see. 

Yeah