Intel 4-5 year tick/tock cycles when AMD is done?

[jpiniero]

They could release a SSE2/x64 CPU today. But it sounds more like a bad excuse. Because nVidia is simply too small to be competitive in that segment. They simply dont have the R&D budget or design team for it.

I seriously doubt nVidia could release a x86 product without the license. Turns out ARM was the way to go regardless.

 

[NTMBK]

They could release a SSE2/x64 CPU today. But it sounds more like a bad excuse. Because nVidia is simply too small to be competitive in that segment. They simply dont have the R&D budget or design team for it.

I suspect that the Tegra K1 would be a lot more successful as an x86 part than as an ARM part- the market for fully fledged OpenGL or DirectX 11 GPUs just isn't there on Android, or any other ARM OS. x86 would give NVidia access to the entire gaming back catalogue associated with it, and they could potentially have made efficient integrated laptop SoCs/APUs. Intel are trying to shut them out of laptops with integrated graphics, and this would have been their countermove. But it didn't happen.

 

[ShintaiDK]

I seriously doubt nVidia could release a x86 product without the license. Turns out ARM was the way to go regardless.

There are already more companies than Intel, AMD and VIA that sells x86 CPUs.

 

[jpiniero]

There are already more companies than Intel, AMD and VIA that sells x86 CPUs.

Yeah, but none right now.

 

[ShintaiDK]

Yeah, but none right now.

So these for example are not x86 compatible?

http://www.vortex86.com/

Looks pretty x86 compatible to me.

 

[jpiniero]

So these for example are not x86 compatible?

http://www.vortex86.com/

Looks pretty x86 compatible to me.

Website's not loading for me.

 

[el etro]

Competition?

FarCry 2 destroy both AMD processors and video cards and its made with a so outdated tech... It should not count as a bench for a test...

http://www.lostcircuits.com/mambo//...sk=view&id=64&Itemid=42&limit=1&limitstart=17

 

[podspi]

That's what intel has been doing. SB is basically nehalem.

You're forgetting power use. Yes, the increases in performance have not been monumental, but power usage has decreased significantly. Since portables are where the market is heading (since most people do not do much more than e-mail and content consumption) -- it makes sense to put their focus in this direction.

 

[Keysplayr]

I actually liked AMD's Phenom II CPU's. I had two of them here. The original Phenom wasn't that good at all. It's all been kind of downhill since Phenom II IMHO.

 

[el etro]

And to stay on topic: I think it's clear Intel has already been doing this since Sandy was released. They "parked the bus" since then.

But clock-per-clock grow(like manufacturing technology improvement) never stopped. Pentiums 4 to Core 2(Or better, Core2 to Nehalem) was a unique IPC grow that not always happened. From Nehalem to Haswell we have 20-30% more clock-per-clock performance.

 

[lamedude]

So these for example are not x86 compatible?

http://www.vortex86.com/

Looks pretty x86 compatible to me.

Only i586/Pentium compatible. My google-fu tells me there are quite a few MMX/SSE patents.

I wonder if Intel will go back to 90s style cycles of 4-5 years between new cpus. Milk the most out of a process before jumping to the next one.

Pentium almost saw a yearly shrink from .8 to .25µ.

 

[Torn Mind]

Or maybe they're slowing down because the $$$$ is not flowing in like it used to. AMD lost all of the price-competitiveness it had when bulldozer flopped and the FX-8150 had to have its price cut to i5 levels instead of its intended purpose as an i7 undercutter. As a selfish organization, Intel still would desire the process node shrink because the lower variable costs of the new node grants them more long-term profitability.

And if Intel chooses to let TSMC and GloFlo catchup, then the term Qualcomm Snapdragon will become much of a household name among the clueless desktop-only guys.

 

[Torn Mind]

In addition, having the same design yet doing a die shrink has resulted consistently in a performance boost and power consumption reduction. Core 2 being shrunk to 45 nm netted an IPC boost for Penryn. Likewise, all of those Pentiums in the 90s were NOT manufactured on the same node even though they were part of the same "family". In fact, the gap between node shrinks shortened starting with 600 nm. The shortest timeframe for node shrinks was 1994-1995.

Since Intel is leading in nm size, accusing Intel of slowing down nodes meant that competitors were even slower than they were, which would not make economic sense since catching up or getting ahead in the node cadence is a long-term advantage.

 

[biostud]

I would also like faster processors, but do we actually have any games that are limited by the CPU speed ATM?

So as long as games are not pushing CPU's to their limits there's really no "need" fa faster CPU's. (Unless you of course do some special kind of work that could benefit from more power)

I'm still planning on getting a Haswell-E..... well, just because.

 

[Fjodor2001]

Secondly, ask yourself. Would you upgrade your CPU if there wasnt something new? The answer is obviously no.

Ask yourself this question too: Would you upgrade if there was something new, but it wasn't much better than what you already had? The obvious answer to that question is no too.

So it could make sense for Intel to only release new CPUs when there actually is a large enough improvement giving people a reason to upgrade, i.e. perhaps every ~5 years on the desktop with the current rate of improvements. That would save them a lot of R&D money.

 

[witeken]

I think that would make them lose more money than it would save them from R&D. The biggest cost from R&D is for new process nodes. You can see that both ARM and AMD can do pretty decent things with their ~40 times smaller R&D budget.

 

[scannall]

The next 10 years should be interesting. 7nm seems to be a pretty big wall, and Intel will hit it first. Where do you go from there? Stacking?

The point being that at some point others will hit that same wall, and there will no longer be a process node advantage. When you can't shrink anymore, which direction do you go?

 

[ShintaiDK]

I think that would make them lose more money than it would save them from R&D. The biggest cost from R&D is for new process nodes. You can see that both ARM and AMD can do pretty decent things with their ~40 times smaller R&D budget.

The processnode is the cheapest part R&D wise. You can simply compare TSMC to Intels R&D budget for example. IC design is where the R&D major expense is. A new uarch like Haswell most likely cost 5-6B$.

 

[witeken]

The processnode is the cheapest part R&D wise. You can simply compare TSMC to Intels R&D budget for example. IC design is where the R&D major expense is. A new uarch like Haswell most likely cost 5-6B$.

Where does that 5B$ go to, exactly?

The next 10 years should be interesting. 7nm seems to be a pretty big wall, and Intel will hit it first.

Here is Brian Krzanich's view on that topic:

in my 30 years i think i have seen the forecasted end of Moore's law at least 5 or 6 times... so i tend to be a skeptic when people say it will end.. At any one point we can typically see about 10 years out.. with pretty good clarity in the 3 to 5 years and much less clarity 5 to 10 years.. but so far in that 10 year horizon.. we don't see anything that says it will end in that time frame..

Where do you go from there? Stacking?

There are quite a lot of possibilities. Here are some:

The point being that at some point others will hit that same wall, and there will no longer be a process node advantage. When you can't shrink anymore, which direction do you go?

I don't want to write a whole text wall about it since someone else has already done very well, so I'll link you to the article: Intel: The End Of Moore's Law.

Basically, your assumption "others will hit the same wall" isn't very obvious, because it will cost a lot of time and money to come at the same level as Intel. By the time physical shrinking ends, Intel might be 4-6 years ahead of the first competitor, and there might only be 1 other competitor in the race (total: 2 or 3). And then you're assuming that there won't be any other things that could be done to improve process. Personally, I think this isn't a topic that really needs to be discussed for another 15 years, since it very likely won't happen in that timeframe (but, of course it can be and I like those informative discussions a lot).

 

[Homeles]

With AMD dying. I see already the smoke, delayed Broadwell, 2014 gets only a Haswell refresh.

I wonder if Intel will go back to 90s style cycles of 4-5 years between new cpus. Milk the most out of a process before jumping to the next one.

It would make sense since it seems each new cpu cycle lately seems to provide small gains of 10-15%

also a price increase of 10-15% would probably happen as well to help increase margins.

I see this happening once AMD is gone.

The only people that see these things happening are the ones that have no clue that ARM exists.

 

[teejee]

IC design is where the R&D major expense is. A new uarch like Haswell most likely cost 5-6B$.

I would be very surprised if this is true.

 

[Lepton87]

In addition, having the same design yet doing a die shrink has resulted consistently in a performance boost and power consumption reduction. Core 2 being shrunk to 45 nm netted an IPC boost for Penryn.

You are kind of wrong, simply shrinking the process node won't affect IPC, it's minor architectural changes that increase IPC. They usually go hand in hand with a process shrink. For example Penryn got Fast Radix-16 Divider that was absent in Conroe. Sometimes Intel makes a straight die-shrink like was the case with Nehalem>Westmere.

 

[witeken]

You are kind of wrong, simply shrinking the process node won't affect IPC, it's minor architectural changes that increase IPC.

Can't there be some IPC gains, according to this Intel blog?

 

[jpiniero]

That article seems badly worded. I think it's trying to say that you might be able to increase the frequency easily with a shrink.

 

[nenforcer]

I actually liked AMD's Phenom II CPU's. I had two of them here. The original Phenom wasn't that good at all. It's all been kind of downhill since Phenom II IMHO.

Clock for Clock and Power / Performance AMD hasn't been very competitive since the first generation i7 Nehalem.

However, because both consoles this generation (PS4/XBONE) have gone to AMD 8 Core Jaguar CPU / GCN 7750GPU I believe the AMD FX 8 core cpus will have a long lifetime in terms of PC ports of those console games.

For gaming purposes I don't think an AMD FX 8 core is a bad investment at all, despite the platform limitations (USB 2 / PCI-E 2.0).