Intel adds Iris Pro Skylake-H - 128Mb of eDRAM, 30% graphical improvement

[patrickjp93]

Just now, Sakkura said:

Your source does not say anything about the transistor count of Skylake GT4e.

And nor does it need to. Compare die sizes, multiply the ratio out, add 10% b/c GPUs can be transistor-heavy and denser, and you wind up with 2.23.

[Sakkura]

1 minute ago, patrickjp93 said:

And nor does it need to. Compare die sizes, multiply the ratio out, add 10% b/c GPUs can be transistor-heavy and denser, and you wind up with 2.23.

No. That's exactly the kind of unreliable estimate that makes your numbers useless.

[patrickjp93]

4 minutes ago, Sakkura said:

Your source does not say anything about the transistor count of Skylake GT4e. There's not even any actual source for the Skylake GT2 number.

It's perfectly reliable. It's exactly how you arrive at the die sizes for Haswell GT3e. Know your enemy, and you will win many battles. How many times do you and Prysin have to go home bloody before you learn to stop fighting battles you don't have any hope of winning?

[Sakkura]

1 minute ago, patrickjp93 said:

It's perfectly reliable. It's exactly how you arrive at the die sizes for Haswell GT3e. Know your enemy, and you will win many battles.

No it's not. There is no real source for the number, nor any details on whether it's schematic or layout transistors, or how it's distributed between CPU and GPU etc.

[patrickjp93]

2 minutes ago, Sakkura said:

No it's not. There is no real source for the number, nor any details on whether it's schematic or layout transistors, or how it's distributed between CPU and GPU etc.

It's the total transistor count for the entire die.

[Sakkura]

1 minute ago, patrickjp93 said:

It's the total transistor count for the entire die.

Without any information about how much of that is taken up by the GPU, whether it's layout or schematic transistors, or whether the number is even real at all - there's no source for it. Wikipedia is not a source.

[Prysin]

14 minutes ago, patrickjp93 said:

Sorry, was thinking Carrizo (stupid half-generation BS AMD). Regardless, the premise stands.

 

Also, no, the 6600K is 122.4

http://www.anandtech.com/show/9505/skylake-cpu-package-analysis

interesting. because...

 

https://www.techpowerup.com/cpudb/1829/core-i5-6500

177mm2

 

https://www.techpowerup.com/cpudb/1836/core-i3-6100

150mm2

 

https://www.techpowerup.com/cpudb/1839/pentium-g4400

150mm2

 

makes me wonder who is wrong. But hey, we can go on twitter and ask Ian Cutress.

[Prysin]

17 minutes ago, patrickjp93 said:

AMD disclosed the exact number as 3.1 billion at ISSCC 2015. Where the Hell have you been?

where the hell is their official slides then? Google arent givin it to me.

 

And did they disclose APU or CPU?

[NumLock21]

I wish that new HP Secptre would use a cpu that runs on Iris Pro instead of the standard 520.

[patrickjp93]

8 minutes ago, Prysin said:

where the hell is their official slides then? Google arent givin it to me.

 

And did they disclose APU or CPU?

Their CPUs are APUs, just with the GPU disabled.

[Dackzy]

Could I get a business laptop with the new CPUs please? Something like a T460s  

[Ansau]

3.1B transistors

 

End of the discussion.

[MageTank]

2 hours ago, patrickjp93 said:

It's perfectly reliable. It's exactly how you arrive at the die sizes for Haswell GT3e. Know your enemy, and you will win many battles. How many times do you and Prysin have to go home bloody before you learn to stop fighting battles you don't have any hope of winning?

Must....Resist...Urge...

 

Any who, back to the topic at hand. Pretty sure I called this like, almost a year ago (August 2015), but I will say what I have always said on this subject. Even if this iGPU is not matching a 750 Ti, look at what these things will do for NUC's and netbooks. GTX 750 level performance in a netbook? Extremely low power consumption and ability to be passively cooled? This is by no means an unimpressive feat. If I see a Skylake i5 with GT4e in a netbook, i'd pay a very hefty sum of cash for that device. Preferably with a 720p screen (GTX 750 is not that great at 1080p, can't quite expect this iGPU to be either unless light gaming or browsing). 

 

Add in the ability to use DDR4 SO-DIMM's (Seriously, why are these not popular yet? Every Skylake laptop uses DDR3) and maybe add in EVGA's laptop memory overclocking design, and we might see some decent bandwidth to throw in with these iGPU's. Look at the 940m. That thing is almost as strong as a desktop GTX 750, but it's severely starved on memory bandwidth (32GB/s on some variants) while even a tame 3200mhz DDR4 kit can get 51.2GB/s peak (around 48-49GB/s realistically). If those 4266mhz kits ever hit SO-DIMM market, then that would be 68.2GB/s peak. To put that into perspective, the GTX 960m only has 80GB/s memory bandwidth, and seems to do fine for the most part at 1080p in terms of bandwidth. 

 

I think i rambled on for too long about my dreams for an OP netbook. Carry on. 

[Dabombinable]

8 minutes ago, MageTank said:

Must....Resist...Urge...

 

Any who, back to the topic at hand. Pretty sure I called this like, almost a year ago (August 2015), but I will say what I have always said on this subject. Even if this iGPU is not matching a 750 Ti, look at what these things will do for NUC's and netbooks. GTX 750 level performance in a netbook? Extremely low power consumption and ability to be passively cooled? This is by no means an unimpressive feat. If I see a Skylake i5 with GT4e in a netbook, i'd pay a very hefty sum of cash for that device. Preferably with a 720p screen (GTX 750 is not that great at 1080p, can't quite expect this iGPU to be either unless light gaming or browsing). 

 

Add in the ability to use DDR4 SO-DIMM's (Seriously, why are these not popular yet? Every Skylake laptop uses DDR3) and maybe add in EVGA's laptop memory overclocking design, and we might see some decent bandwidth to throw in with these iGPU's. Look at the 940m. That thing is almost as strong as a desktop GTX 750, but it's severely starved on memory bandwidth (32GB/s on some variants) while even a tame 3200mhz DDR4 kit can get 51.2GB/s peak (around 48-49GB/s realistically). If those 4266mhz kits ever hit SO-DIMM market, then that would be 68.2GB/s peak. To put that into perspective, the GTX 960m only has 80GB/s memory bandwidth, and seems to do fine for the most part at 1080p in terms of bandwidth. 

 

I think i rambled on for too long about my dreams for an OP netbook. Carry on. 

I still don't get why low bandwidth DDR3 is used with 940M. Its the kind of thing you'd expect 6 years ago with DDR3 1600 being the main choice (though the laptops that have 64MB dedicated DDR2 800, and 256MB shared DDR3 1333 are really odd)

[Daegun]

3 hours ago, patrickjp93 said:

It's perfectly reliable. It's exactly how you arrive at the die sizes for Haswell GT3e. Know your enemy, and you will win many battles. How many times do you and Prysin have to go home bloody before you learn to stop fighting battles you don't have any hope of winning?

They have no hope of winning, yet all you ever do is provide guesswork and if you do miraculously provide a source for anything, it doesn't actually support any of your claims. You're like wccf in terms of reliability. Remember how Jim Keller was already working at Intel and all he had to do was announce it and then he ended up working for Tesla?

 

2 hours ago, Ansau said:

3.1B transistors

 

End of the discussion.

Thank you for posting the slide for us.

[AyeJayKay]

This is good news as I'm probably going to have to upgrade to a MacBook Pro with integrated graphics since Apple refuses to upgrade the Air to a Retina display.

[Big-G]

7 hours ago, zMeul said:

despite providing better iGPs in their APUs, AMD had a hard time selling APUs and their market strategy failed hard

Intel could use some random iGPs in their CPUs and still sell a ton of them - what AMD didn't get: people don't buy integrated graphics for graphical performance

AMD APU's are only relevant to the console market and "console level gaming" PC market... those commonly referred to by the PC master race as peasants...

 

7 hours ago, MrDynamicMan said:

This is still dumb. Their best iGPUs should be on CPUs that would not typically be paired with a dgpu. When they put anything other than the most basic iGPU on a top tier chip theyre just wasting die space.

Not everybody games... thus requiring dedicated GPU's. The iGPU's do enough to not need to get a separate low spec dGPU.

 

 

[79wjd]

41 minutes ago, AyeJayKay said:

This is good news as I'm probably going to have to upgrade to a MacBook Pro with integrated graphics since Apple refuses to upgrade the Air to a Retina display.

A Retina Air (assuming you're referring to the 13" variant) would basically be the Retina Pro (or have abysmal battery life).

[AyeJayKay]

1 hour ago, djdwosk97 said:

A Retina Air (assuming you're referring to the 13" variant) would basically be the Retina Pro (or have abysmal battery life).

Yes, I want a Retina Air 13" but if Apple doesn't update the Air then I will be forced to get a MacBook Pro. Not ideal but what can you do.

[MrDynamicMan]

Just now, GidonsClaw said:

AMD APU's are only relevant to the console market and "console level gaming" PC market... those commonly referred to by the PC master race as peasants...

 

Not everybody games... thus requiring dedicated GPU's. The iGPU's do enough to not need to get a separate low spec dGPU.

 

 

If someone build a system with a 6700k, does intensive work eith it, and still doesnt have a basic dGPU thats their problem. A more basic iGPU with more cores/cache/etc. would be far smarter. For those people who dont use at least a 950/270x/W.E. With a 300+ cpu, thats on them.

[patrickjp93]

4 hours ago, MrDynamicMan said:

If someone build a system with a 6700k, does intensive work eith it, and still doesnt have a basic dGPU thats their problem. A more basic iGPU with more cores/cache/etc. would be far smarter. For those people who dont use at least a 950/270x/W.E. With a 300+ cpu, thats on them.

Okay, let's look at this purely from a compute power standpoint.

 

GFlops on 8 Skylake cores at 3.3GHz:

8 * 256/32 (AVX2) * 2 (FMA) * 3.3*10^9 = 422.4 GFlops

 

GFlops on 6700K w/GT2.

4*8*2*4.2*10^9

+

24(EU)*8(SP)*2(FMA)*1.15*10^9

= 710.4GFlops

 

Integrated graphics is the answer to getting performance density in a central processor. Heterogeneous acceleration will get here, eventually...

[MrDynamicMan]

Just now, patrickjp93 said:

Okay, let's look at this purely from a compute power standpoint.

 

GFlops on 8 Skylake cores at 3.3GHz:

8 * 256/32 (AVX2) * 2 (FMA) * 3.3*10^9 = 422.4 GFlops

 

GFlops on 6700K w/GT2.

4*8*2*4.2*10^9

+

24(EU)*8(SP)*2(FMA)*1.15*10^9

= 710.4GFlops

 

Integrated graphics is the answer to getting performance density in a central processor. Heterogeneous acceleration will get here, eventually...

Eventually.

But until then, why waste die space on a iGPU that almost no-one will use?

[patrickjp93]

2 minutes ago, MrDynamicMan said:

Eventually.

But until then, why waste die space on a iGPU that almost no-one will use?

The same reason AMD does: to have the best product available when that day comes, as future-proofing sells.

[MrDynamicMan]

Just now, patrickjp93 said:

The same reason AMD does: to have the best product available when that day comes, as future-proofing sells.

I guess so.

but (as a genuine question) why is it important for the gpu to be on-die for HSA?

[patrickjp93]

1 minute ago, MrDynamicMan said:

I guess so.

but (as a genuine question) why is it important for the gpu to be on-die for HSA?

Because otherwise you have to copy memory to a dGPU's local memory. That takes time and bandwidth. For latency-sensitive applications, integrated accelerators are king.