AMD CPU speculation... and expert conjecture

[-Fran-]

Well, you still have the SoC approach. Or the Package route as well. I know it's cheaper to just slap everything into the same die, but heat production will get too messy IMO. Now, like noob2222 points out, AMD has a very clever design to have better dissipation. Adding low power parts (generating heat parts, that is) won't be such a hard thing to do, but since the die could get too big, there's also the yield issue at hand.

It's not a simple balance to just slap things into the die (now that I think about it), since the yield and ramping are tied too damn close. You could lower the cost of the entire platform for sure by slapping everything to the CPU die (NB, GPU, SB, IMC, PCIe, etc), but then low yields will be expected and prices would still rise unless you do so on a mature node.

Intel has been slowly slapping stuff to the CPU and I think it's the correct "rhythm" to follow. Since GF and TSMC don't have the same yield and ramping (and process), Intel is leading the pack here.

And regarding the HT links. I remember doing OC to the IMC and the HT link in the PhII. It worked fine for the most part, but don't remember the exact OC. I'm sure I have some pictures of it, haha.

Cheers!

 

[iceclock]

quantum computing is interesting but way too costly. gonna be a long time before we see public avaible quantum computers. also what do u mean by optical cpus?

 

[truegenius]

quantum computing is interesting but way too costly. gonna be a long time before we see public avaible quantum computers. also what do u mean by optical cpus?

optical transistor based cpus
http://en.wikipedia.org/wiki/Optical_computing
http://www.gsu.edu/news/63294.html

 

[iceclock]

interesting uses the energy of light to compute stuff.

any known release date for this technology?

 

[mayankleoboy1]

before consumer "optic procs", i think we could be switching to graphene or gallium based processors.
Optics, the way they are currently being researched is just to increase the internal bandwidth of buses. There isnt any clock speed increase per se. And light isnt being used to make gates, or for logic. Its being used like an on chip fiber optic.

 

[Cazalan]

Total system heat production, yes. But you're also putting all the components on a smaller total space, so the thermals will all be contained to a smaller area, making them harder to dissipate.

The transistor layouts will have to continue to improve to accommodate the heat generation. We're only at rev 1 of FinFet technology. The goal is to further reduce leakage current so voltages can get much lower and still be stable. Power goes up as a function of Voltage squared times frequency, so every mV reduction helps.

The thermal modeling software/simulations have to get better too. Knowing where to add additional thermal vias in "hot" areas. It adds some die area but you can pack the "cool" areas more densely to compensate.

http://www.electronics-cooling.com/2004/08/thermal-vias-a-packaging-engineers-best-friend/

That's why they say we need faster computers to make faster computers. Simulating all this physics and verifying it in the lab is a tedious and time consuming process.

 

[Cazalan]

Some details on the differences between Bobcat and Jaguar from ISSC this week.

http://www.techpowerup.com/180394/AMD-quot-Jaguar-quot-Micro-architecture-Takes-the-Fight-to-Atom-with-AVX-SSE4-Quad-Core.html

 

[truegenius]

is it possible to insert some metal wires in die (just like candles in cake) during manufacturing so these wires will help in faster heat transfer but increased die size (atleast it will be better than 95w in under 1mm² area ) ?

or in other words
just like ventilation system in subway/tunnels

 

[Cazalan]

is it possible to insert some metal wires in die (just like candles in cake) during manufacturing so these wires will help in faster heat transfer but increased die size (atleast it will be better than 95w in under 1mm² area ) ?

or in other words
just like ventilation system in subway/tunnels

Yes they're called thermal vias. See link a couple posts back.

 

[gamerk316]

before consumer "optic procs", i think we could be switching to graphene or gallium based processors.

Apparently, Graphine has some issues due to lacking the ability to shut off electron flow, making solid state devices impossible. Its loosing favor as the next-gen material of choice as a result.

http://spectrum.ieee.org/semiconductors/nanotechnology/graphenes-new-rival
http://spectrum.ieee.org/nanoclast/semiconductors/nanotechnology/graphene-is-losing-favor-as-the-twodimensional-material-of-the-future

 

[Cazalan]

Per the previous discussion about core use in games, it looks like Crysis 3 is heavily multi-threaded and the FX-8350 is the best cpu for the game aside from the 6-core intel chips. Impressive for a $200 cpu.

It does appear to utilize more cores but those 8350 to 6300 numbers look odd. It's 14% faster which is the same 14% clock speed gain. That means 2 more cores added nothing.

 

[-Fran-]

It does appear to utilize more cores but those 8350 to 6300 numbers look odd. It's 14% faster which is the same 14% clock speed gain. That means 2 more cores added nothing.

It could also mean they used the 6 cores from the XBox as a threading target, haha.

Cheers!

 

[gamerk316]

It does appear to utilize more cores but those 8350 to 6300 numbers look odd. It's 14% faster which is the same 14% clock speed gain. That means 2 more cores added nothing.

Which makes sense if none of the cores are near 100% on the 6 core system. Its not a bottleneck, so the higher clock speeds win out.

 

[JAYDEEJOHN]

Theres this one too
http://www.pcgameshardware.de/Crysis-3-PC-235317/Tests/Crysis-3-Test-CPU-Benchmark-1056578/

 

[-Fran-]

And talking about Crysis 3... Where are the Tom's benchmarks for it?

Come on Mr. Angelini! hahaha

Cheers!

 

[JAYDEEJOHN]

Looks like they started from scratch here, and/or heavily integer usage
It can be done, but they have to start early

 

[esrever]

Impossible.

 

[BeastLeeX]

Theres this one too
http://www.pcgameshardware.de/Crysis-3-PC-235317/Tests/Crysis-3-Test-CPU-Benchmark-1056578/

Don't just look at the FX-8350, check out the FX-6300, its trading blows with the i5 2500k, and right behind the 3570k!

 

[JAYDEEJOHN]

Thats the point, it isnt easy, and it isnt the way weve always done it, but it can be done.
Leave it to crytek to innovate

 

[Blandge]

This is only useful for low power systems. Copper can only dissipate soo much heat. Shrinking and adding more power to a single chip creates thermal density, the one thing copper can't change, its thermal dissipation. Look at sandy vs ivy bridge. Roughly 10 less watts, up to 20 degrees hotter with the stock fan, and maximum overclock reduced by 700 mhz under water cooling. This is the effect of thermal density. Packing more crap on the same die will only make it worse.

Everyone called me crazy when I suggested the mighty 14nm broadwell might not see DT speeds without some help with dissipating the hot spot.

This is wrong. The 20 degree increase in IVB temps was due to using a thermal interface material between the heat spreader and the CPU die which causes a gap to form between the cpu die and heat spreader (on sandy bridge the heat spreader was soldered on which replaces this gap with solder). Removing the head spreader and reducing the distance between the head spreader and cpu die and replacing the TIM decreases temperatures by.... guess how much? 20 degrees.

The lower overclockability of IVB has to do with 22nm finFET device characteristics, not heat dissipation.

Check out this thread over at Anandtech for a detailed discussion.

Delidded my i7-3770K, loaded temperatures drop by 20°C at 4.7GHz

 

[JAYDEEJOHN]

Finfet is designed for lower v threshold, allowing for same perf, less wattage, less heat per mm.
As finfet gets better, the better question is, will it be enough to compensate for newer nodes?
I think it will, as like blandge (almost mispelled that) points out above, it could be some sandbagging going on, as better heat dispersion can be had down the road easily, combined with 2nd gen finfet should be no problem, after that tho, who knows?

 

[Blandge]

Finfet is designed for lower v threshold, allowing for same perf, less wattage, less heat per mm.
As finfet gets better, the better question is, will it be enough to compensate for newer nodes?
I think it will, as like blandge (almost mispelled that) points out above, it could be some sandbagging going on, as better heat dispersion can be had down the road easily, combined with 2nd gen finfet should be no problem, after that tho, who knows?

Mark Bohr over at intel gave an interview where he talks about several transistor technologies that have better physical characteristics than even finFETs have today. Just like HKMG and finFET technology, there are more innovations that will be introduced in the 10nm and 7nm timeframe that will push transistors to lower voltage and leakage, and smaller dimensions.

 

[griptwister]

Sooo... how about that PS4? Looks like we'll finally see some better quality games.

 

[BeastLeeX]

Sooo... how about that PS4? Looks like we'll finally see some better quality games.

Will it get the Oculus Rift? As of now, it won't, so I really don't care about it!

Edit: Just saw a picture of the controller, and it seems innovative, but a also seems like a feature the 50% of the crowd will like and 50% will hate, also, it looks like poo.

 

[JAYDEEJOHN]

Mindshare like this, with the 8350 doing so well, should help AMD
http://www.xtremesystems.org/forums/showthread.php?284767-Geforce-Titan-GK110-will-be-a-consumer-part&p=5171823&viewfull=1#post5171823