With the current rate of Intel CPU performance increases, could AMD be catching up?

[colonelciller]

The lower power consumption means that there is enough headroom for extra frequency if Intel needs, meaning that there are improvements, it's just that those improvements aren't where you want them to be, but they are there.

Offering "improvements" that people don't care about only works in a monopoly market.

As a matter of fact, Intel's 32nm and 22nm chips can reach clocks as high as any AMD chips but with much less power consumption. It's just marketing that prevents a 4 or even a 4.2 IVB-4C SKU to be launched.

Performance "potential" unavailable for consumers to use is... useless.

Useless, unavailable, locked-down performance potential is not feature... well not a feature for consumers at least, although it probably is considered a great feature over at Intel Headquarters.

 

[ShintaiDK]

Offering "improvements" that people don't care about only works in a monopoly market.



Performance "potential" unavailable for consumers to use is... useless.

Useless, unavailable, locked-down performance potential is not feature... well not a feature for consumers at least, although it probably is considered a great feature over at Intel Headquarters.

Looking at sales, its exactly what people are asking for. Seems to be you asking for a minority request.

 

[Pilum]

Intel spec's higher and higher TJmax values because it allows them to then spec lower and lower performing stock HSF's which then saves them a couple bucks per retail CPU.

(I am not joking or making that up, they even had a slide in one of their IDF presentations where they specifically outlined that being the sole motivation for increasing the TJmax for G0 steppings of Kenstfield over the TJmax of B3 steppings)

The downside of allowing your CPU to operate at higher temperatures is the power consumption spirals upwards. And by setting the TJmax value to a higher value that means the TDP "budget" gets used up rather quickly.

But from this follows that the current performance limitations of IVB are an economical decision, not a technological limit. If Intel would switch back to a decent TIM and use more expensive stock coolers, they could push out higher IVB SKUs even with the current silicon.

So Intel still has performance reserves, which are currently transformed into additional profits. If AMD should manage to catch up considerably, Intel could just increase production costs slightly and everything would be back to square one.

 

[mrmt]

Useless, unavailable, locked-down performance potential is not feature... well not a feature for consumers at least, although it probably is considered a great feature over at Intel Headquarters.

I'm glad you acknowledged that there are improvements, it's just that they are not for you. Just ask the guys who need battery life, or the guys managing datacenters to see if they consider efficiency improvements useless. They don't, and this is the reason for the bloodbath AMD is receiving now. The problem is that you are not a priority for Intel, there are more people willing to pay more for more efficiency than you are willing to pay for more performance.

As a matter of fact, desktop isn't a priority even for AMD. They didn't even bother to design a specific SKU for desktops. What you buy are crippled server SKUs. On top of that AMD stated multiple times that their focus is in mobile, not in desktops. In fact, they are so "strong" in desktops because there are not enough demand for them in servers and mobile.

 

[Fjodor2001]

Did you calculate for the baseclock for FX8350 went up 400Mhz compared to the FX8150?

I've never mentioned the FX8150/8350. I'm talking about the AMD chips (Llano/Trinity/...) that correspond to the Intel mainstream desktop CPUs (SB/IB/H/...).

And IPC didnt increase 20-30% either. Not to mention the chip run out of spec with power consumption as well.

I've never said that IPC went up 20-30%. Read my post again. I said that performance went up by 20-30% (or 21% / 29% specifically in the benchmarks I mentioned in that post).

 

[Fjodor2001]

The lower power consumption means that there is enough headroom for extra frequency if Intel needs[...]

Not that much headroom:

The TDP increases quite rapidly beyond 3.5 GHz.

(Image created by Idontcare )

 

[Idontcare]

But from this follows that the current performance limitations of IVB are an economical decision, not a technological limit. If Intel would switch back to a decent TIM and use more expensive stock coolers, they could push out higher IVB SKUs even with the current silicon.

So Intel still has performance reserves, which are currently transformed into additional profits. If AMD should manage to catch up considerably, Intel could just increase production costs slightly and everything would be back to square one.

Of course they are matters of economical decisions. So too is true for AMD.

AMD went to great lengths and expense to prove to the world that their bulldozer chips can clock higher than 8GHz, but they don't want to go to great expense selling them with the required cooling solutions.

Where things work for Intel is they can (1) easily raise the TDP spec and release higher clocked SKUs (AMD's back is already against the wall there), and/or (2) easily lower the TJmax value by tens of degrees and release higher clocked SKUs (AMD's back is already against the wall there too).

This is where you can tell Intel doesn't feel inclined to push themselves or their products. There is only one reason we do not have a 3870k clocked at 4GHz base with a 95 or 105W TDP and that is because AMD isn't forcing their hand to do so.

 

[ShintaiDK]

I've never mentioned the FX8150/8350. I'm talking about the AMD chips (Llano/Trinity/...) that correspond to the Intel mainstream desktop CPUs (SB/IB/H/...).

I've never said that IPC went up 20-30%. Read my post again. I said that performance went up by 20-30% (or 21% / 29% specifically in the benchmarks I mentioned in that post).

You do know you compare a double jump in generations? Its like comparing Lynnfield to Sandy Bridge.

http://images.anandtech.com/graphs/graph4083/35030.png

So... :whiste:

 

[Idontcare]

Not that much headroom:

The TDP increases quite rapidly beyond 3.5 GHz.

(Image created by Idontcare )

What you say is true, but its not a practical limit until you try and get above 4.5GHz.

Intel could sell a 4.5GHz (base) 3770K with a TDP of say 110W and a TJmax of 80C and they'd be fine. But it would really undermine/cannabilize their efforts to build the "extreme" brand based on S2011.

 

[mrmt]

Ihe TDP increases quite rapidly beyond 3.5 GHz.

I never said it doesn't. I just said that there is headroom there. 3.5GHz is by no means the fastest commercially viable IVB SKU.

 

[Haserath]

The fact is AMD has to improve their performance per watt just as much as Intel to keep up. We can't have higher wattage processors without serious cooling changes.

The industry will trend toward smaller and cooler anyway.

 

[Blandge]

Fanboy reading comprehension fail. He said "performance", not performance/watt. Obviously, because Intel has leading process technology, they are going to be ahead in performance/watt. But the very fact that AMD is indeed competitive in performance, speaks volumes about the quality of their engineering.

Except that AMD is not competitive in performance when you compare systems at similar power ratings and die sizes (FX-8350 to i7-3960X). That's the point. AMD's performance numbers only speak volumes about their pricing model.

 

[Fjodor2001]

You do know you compare a double jump in generations?

Really? Then please remind me of the CPU generation between Llano and Trinity.

 

[Blandge]

Really? Then please remind me of the CPU generation between Llano and Trinity.

Stars -> Bulldozer -> Pildriver

Llano is equivalent to Stars
Trinity is equivalent to Piledriver

 

[ShintaiDK]

Really? Then please remind me of the CPU generation between Llano and Trinity.

Bulldozer.

Just like the step from Lynnfield to Sandy Bridge is a one step.

So you see, its no different. Well there is a difference. Intel managed to improve multithreading as well as single. Unlike AMD as you can see here:
http://images.anandtech.com/graphs/graph6347/50398.png

So you actually proved that AMD falls further behind.

 

[Fjodor2001]

What you say is true, but its not a practical limit until you try and get above 4.5GHz.

Intel could sell a 4.5GHz (base) 3770K with a TDP of say 110W and a TJmax of 80C and they'd be fine. But it would really undermine/cannabilize their efforts to build the "extreme" brand based on S2011.

Yes, that's right.

But note that the Intel chips also have a Turbo frequency at about 0.5 GHz higher than the base frequency. So either they'd have to skip that, or they'd be at 5.0 GHz in Turbo. That would mean a TDP @ 200 W...

So in Turbo mode you'd gain 5.0/3.9 GHz = 28% performance at the cost of 200-77 W = 123 W TDP.

 

[Fjodor2001]

Bulldozer.

Please remind me what iGPU the Bulldozer CPUs have, that makes them fit between Llano and Trinity.

Also, there may be 2 uarch steps between Llano and Trinity, but Trinity was released 1.5 years after Llano. I.e. about same time difference as between SB and IB. The time difference should be what matters after all (i.e. performance increase per year).

 

[Blandge]

Yes, that's right.

But note that the Intel chips also have a Turbo frequency at about 0.5 GHz higher than the base frequency. So either they'd have to skip that, or they'd be at 5.0 GHz in Turbo. That would mean a TDP @ 200 W...

So in Turbo mode you'd gain 5.0/3.9 GHz = 28% performance at the cost of 200-77 W = 123 W TDP.

Explain to me again why you need turbo if you're running at 4.5 GHz?

 

[Blandge]

PleaAlso, there may be 2 uarch steps between Llano and Trinity, but Trinity was released 1.5 years after Llano. I.e. about same time difference as between SB and IB. The time difference should be what matters after all.

Umm no. Llano was released late. IIRC it was slated for early (January?) 2011 and was released in June. The difference should be closer to 2 years.

 

[Fjodor2001]

Umm no. Llano was released late. IIRC it was slated for early (January?) 2011 and was released in June. The difference should be closer to 2 years.

Llano was released in July 2011 (A8-3850), Trinity in October 2012 (A10-5800K). So actually only about 1 year between them.

 

[Fjodor2001]

Explain to me again why you need turbo if you're running at 4.5 GHz?

For the same reason you need it when running at 5.0 Ghz. Or at 3.5 GHz. Or at...

 

[Blandge]

Llano was released in July 2011 (A8-3850), Trinity in October 2012 (A10-5800K). So actually only about 1 year between them.

Ok so it was only 22 months from when Llano was supposed to be released to when Trinity was released. The Stars microarchitecure which Llano is based on was released with Phenom II in 2008. So 4 years.

In any case. The time scales don't matter in the slightest. It's the microarchitectural generations that determine how different one CPU is from another. The performance comparison of the Husky (Read: Stars) cores in Llano and Piledriver cores inside Trinity is significantly different from the comparison of Stars to Bulldozer in when performance is more or less a wash.

That's why it's 2 generations.

 

[Homeles]

Please remind me what iGPU the Bulldozer CPUs have, that makes them fit between Llano and Trinity.

Also, there may be 2 uarch steps between Llano and Trinity, but Trinity was released 1.5 years after Llano. I.e. about same time difference as between SB and IB. The time difference should be what matters after all (i.e. performance increase per year).

I'm with you on this one. Llano was an updated version of Phenom/Athlon II. It does feature higher IPC. Therefore it counts as another generation.

 

[AtenRa]

Mobile Llano was released on 14th of June 2011, Mobile Trinity was released on May 15 2012.

 

[Blandge]

I'm with you on this one. Llano was an updated version of Phenom/Athlon II. It does feature higher IPC. Therefore it counts as another generation.

Doesn't look like a higher IPC to me.