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

[Madpacket]

It's about focusing and timing. AMD did this successfully in the past even with an anti-competitive Intel and can do it again with their APU's if they can avoid the foundry issues. Intel is becoming ignorant with how they're positioning the Haswell CPU's and holding back the good bits (proper GPU's) from consumers which is a mistake and will give AMD the crack to fill in the most popular price categories. I see them having a decent chance of taking a large chunk of the low end market this back to school and holiday season.

 

[Homeles]

That was also true around 2003-2006, but yet AMD could surpass them then.

Intel has already learned from that mistake.

 

[mrmt]

That was also true around 2003-2006, but yet AMD could surpass them then.

AMD was never good with native R&D. First they reversed-engineered Intel chips, then acquired NexGen and the K6-2 core, they when DEC went bankrupt they brought in a lot of IP and even an entire design team, the K8 team, all that to manufacture chips in nodes acquired first from Motorola then from IBM.

As there is no free lunch for AMD this time, the company is derailing.

 

[Piroko]

Yeah yeah, choose between the beggar and the bully. One will take your money, the other one will make you give him your money, eh?

 

[dave_the_nerd]

Their TDP limits aren't arbitrary, they have real cost/design implications. Better to sell unlocked processors and let the users worry about accommodating the increased power consumption than put OEMs on the hook (and provide costlier HSFs). Intel has had enough headaches recalling P3s and Prescotts that pushed thermal requirements to levels probably not even as aggressive as what you're suggesting.

::confused dog head-tilt::

Thermal requirements of an OC'd i5 (mine runs at 4.2 quite happily on a stock cooler and 1.1v, while still consuming less than it's 77w TDP) are well within the limits already established for i7s (stock TDP 95w) and the 6-core Extreme-Editions (Stock TDP 130w.)

Intel could easily bin an i5 "Extreme Edition" for single-threaded performance nerds that had a 95w TDP and ran at 4.0GHz Base / 4.4GHz Turbo.

 

[Exophase]

Extreme edition doesn't count, it's a different platform and comes with a huge markup. Intel obviously has incentive to reduce TDP for their main platforms. Yeah they could do a 4 core EE CPU with a higher base clock but who is going to buy it over an unlocked edition? The necessary markup would be too high to justify it..

As for running all cores at 4.2GHz and staying under the 77W TDP, I find that a little hard to believe, was that actually under full load?

 

[ShintaiDK]

That was also true around 2003-2006, but yet AMD could surpass them then.

Chips was much cheaper to design back then and AMD had its own fabs. Plus it was a time were Intel essentially made every mistake they could and AMD had success with everything. Thats how much it takes for AMD to be in front. Its about equal to winning the lottery.

 

[Piroko]

Thermal requirements of an OC'd i5 (mine runs at 4.2 quite happily on a stock cooler and 1.1v, while still consuming less than it's 77w TDP) are well within the limits already established for i7s (stock TDP 95w) and the 6-core Extreme-Editions (Stock TDP 130w.)

My Ph2 x3 runs at 1.225V (rated 1.425V), but I don't assume that every chip will do that on every board. Intel has to bin for stability while taking board variations and elevated temperatures into account...

Chips was much cheaper to design back then and AMD had its own fabs.

Chips weren't that much cheaper to design back then. Heck, Intel already had higher than 40% margins, a wet dream for most, if other companies had seen seen an opportunity back then they would've taken it.

 

[Fjodor2001]

Intel has already learned from that mistake.

Meaning exactly what?

And it's a guarantee that things won't ever change? This is a fast changing business. If the you relax for a few years you will become run over by someone else. 8% CPU performance increase per year for me is relaxing.

Only the paranoid survive...

 

[parvadomus]

AMD was never good with native R&D. First they reversed-engineered Intel chips, then acquired NexGen and the K6-2 core, they when DEC went bankrupt they brought in a lot of IP and even an entire design team, the K8 team, all that to manufacture chips in nodes acquired first from Motorola then from IBM.

As there is no free lunch for AMD this time, the company is derailing.

LOL. Not true. They need better process tech, thats all.
Current arch is not that bad, you will see with Kaveri.

 

[Idontcare]

::confused dog head-tilt::

Thermal requirements of an OC'd i5 (mine runs at 4.2 quite happily on a stock cooler and 1.1v, while still consuming less than it's 77w TDP) are well within the limits already established for i7s (stock TDP 95w) and the 6-core Extreme-Editions (Stock TDP 130w.)

Intel could easily bin an i5 "Extreme Edition" for single-threaded performance nerds that had a 95w TDP and ran at 4.0GHz Base / 4.4GHz Turbo.

Unfortunately that isn't how it works. Take your CPU and put the stock HSF back onto it.

Now OC however much you like but you must ensure it is stable all the way up to, and including, running at TJmax. That takes more volts than getting it to run stable at 60C because of your 3rd party cooler.

Now you've added more volts, and because of the elevated temperature your leakage current is much larger. Suddenly you'll realize you are no longer fitting inside the TDP spec.

That is how it works. You don't just get to set arbitary spec values.

If you set a temperature limit then that bakes in hard minimums on operating voltage to maintain stable clockspeed, all of which further bakes in your maximum expected power consumption which must fit within your spec'ed TDP parameter.

 

[Ajay]

AMD was never good with native R&D. First they reversed-engineered Intel chips, then acquired NexGen and the K6-2 core, they when DEC went bankrupt they brought in a lot of IP and even an entire design team, the K8 team, all that to manufacture chips in nodes acquired first from Motorola then from IBM.

As there is no free lunch for AMD this time, the company is derailing.

But by then, they had an excellent CPU R&D, just unbelievably poor management. Look who they turn to for help, Jim Keller - from Digital and the original K8 team. I think AMD would have been better off buying DEC's entire CPU division and Fab. The might have gotten some good product managers out of the deal and Alpha could have been the last high end RISC CPU standing.

 

[Ajay]

They need better process tech, thats all.
Current arch is not that bad, you will see with Kaveri.

That part is surely true. @28nm, Kaveri will probably be a niche product. @20nm (if GloFo could come anywhere near to their roadmap timetables), it would have been a contender. I can't for the life of me figure out why Malta wasn't built from the ground up to start on 20nm - except that GloFo management didn't have faith in their own process tech team - which is scary.

So for both external and internal reason, AMD is derailing like mrmt said. If there was ever a company in need of angel investor...

 

[mrmt]

But by then, they had an excellent CPU R&D, just unbelievably poor management. Look who they turn to for help, Jim Keller - from Digital and the original K8 team. I think AMD would have been better off buying DEC's entire CPU division and Fab. The might have gotten some good product managers out of the deal and Alpha could have been the last high end RISC CPU standing.

AMD management destroyed the K8 team, they can do it again. Plus Keller is just one man, not an entire team.

 

[Ajay]

AMD management destroyed the K8 team, they can do it again. Plus Keller is just one man, not an entire team.

He is probably doing best and is likely a very good leader, but with engineers being fired or laid off, morale must be in a gutter. Having someone like Keller will help somewhat, just because of his prestige in the industry.

And yes, they can bork it all again. Someone desperately needs to take AMD private and put in a first class management team. Papermaster's ARM strategy didn't take into account the fact that every top ARM company would be coming at them (plus lots of VC $$s going into small companies aiming to become Google & Facebook's next server provider). I doubt Intel would cancel the x86 IP agreement with the new company (though they would be wise to see if the DOJ or some mutual partners. between Intel & AMD, would be ready to back them up). And they need to get out of the WSA anyway they can.

If something like this doesn't happen b/4 the end of the year, I fear ShintaiDK's prediction (AMD will devolve into Via) is going to come true; that or they will have to declare bankruptcy sometime in 2014, maybe 2015 w/some luck. Broadwell is just going to crush AMD (if Intel's new iGPU architecture and drivers are successful).

 

[Homeles]

Meaning exactly what?

Do I really have to spell it out for you?

And it's a guarantee that things won't ever change?

No. But it's improbable. Why do you keep pushing this?

This is a fast changing business. If the you relax for a few years you will become run over by someone else.

And you think AMD of all companies is going to be the one to run them over? How laughable.

8% CPU performance increase per year for me is relaxing.

Pretty sure AMD's running a similar average right now. And you're absolutely short-sighted if you're looking at performance as your only measuring stick.

For example, Intel could very, very easily release a higher clocked version of Ivy Bridge or Haswell at similar TDP's to AMD's FX. Ta-freaking-da, you now have that dumb performance increase you're looking for. And if you think that Intel doesn't have the fab maturity to do this, you're clueless.

You and everyone else needs to stop blindly looking at performance as the only thing that matters. You're an enthusiast. Learn how to overclock.

Only the paranoid survive...

Intel's already got a fire under their rear: ARM. Make all the vapid comments about Intel "slacking off" that you'd like. They're not.

 

[frozentundra123456]

LOL. Not true. They need better process tech, thats all.
Current arch is not that bad, you will see with Kaveri.

Lack of fabs or poor results from GF is not really an excuse for AMD. It is like saying I have a great car, except the engine is broken. The two are irrevocably tied together. Doesnt matter whose fault it is; it only matters how the final product performs.

 

[Exophase]

Unfortunately that isn't how it works. Take your CPU and put the stock HSF back onto it.

Now OC however much you like but you must ensure it is stable all the way up to, and including, running at TJmax. That takes more volts than getting it to run stable at 60C because of your 3rd party cooler.

Now you've added more volts, and because of the elevated temperature your leakage current is much larger. Suddenly you'll realize you are no longer fitting inside the TDP spec.

That is how it works. You don't just get to set arbitary spec values.

If you set a temperature limit then that bakes in hard minimums on operating voltage to maintain stable clockspeed, all of which further bakes in your maximum expected power consumption which must fit within your spec'ed TDP parameter.

Well he did say he was using the stock cooler, but I still find the whole thing a little unlikely.. and if it does happen to work for him it may not for all chips across the whole allowed ambient temperature range, under all reasonable usage scenarios.

I could buy Intel setting the clock speed lower than what they could under other parameters, what I don't buy is that they left a ridiculous amount of slack in the TDP spec. If they could make that lower they wouldn't hesitate to, and they've been aggressively doing so.

 

[frozentundra123456]

Well he did say he was using the stock cooler, but I still find the whole thing a little unlikely.. and if it does happen to work for him it may not for all chips across the whole allowed ambient temperature range, under all reasonable usage scenarios.

I could buy Intel setting the clock speed lower than what they could under other parameters, what I don't buy is that they left a ridiculous amount of slack in the TDP spec. If they could make that lower they wouldn't hesitate to, and they've been aggressively doing so.

I agree that 4.0 base clock is pretty high, but I would think they could at least go to something like 3.7, 3.8 with turbo around 4.0, 4.1. Even if the TDP went to 90 or 100 watts, so what. Using more power on the desktop doesnt bother me as long as you get better performance. That is what 7% more performance basically free?? That is as much as they have been getting with an entire new generation with the last two new ones.

 

[Homeles]

LOL. Not true. They need better process tech, thats all.

In case you weren't aware, Sandy Bridge was also on 32nm.

Current arch is not that bad, you will see with Kaveri.

But that's not the same arch, is it? Good one.

 

[Idontcare]

I agree that 4.0 base clock is pretty high, but I would think they could at least go to something like 3.7, 3.8 with turbo around 4.0, 4.1. Even if the TDP went to 90 or 100 watts, so what. Using more power on the desktop doesnt bother me as long as you get better performance. That is what 7% more performance basically free?? That is as much as they have been getting with an entire new generation with the last two new ones.

The problem is definitely the TJmax being set so high.

Take the 3770k that I characterized for power vs. temperature vs. voltage vs. clockspeed.

With a TDP of 77W and a TJmax of 105C the maximum clockspeed Intel could set that particular 3770k to is 3.8GHz (all cores), and that is with the operating voltage optimized for LinX stability (meaning no engineering room for margin against parametric degredation over the operating lifetime of the CPU).

There's not much headroom left on the table IMO.

Now bump up that TDP from 77W to 95W and all you gain is just enough head-room to bump up the clocks to 4GHz from 3.8GHz.

The reason people think of their OC'ed processors as being proof that Intel is leaving lots of clockspeed on the table is that they are keeping the operating temperatures well below TJmax. That keeps the leakage currents quite low as well as enabling the user to set their operating voltage even lower than would otherwise be required for stable operation up to TJmax.

Intel doesn't have that luxury. Not unless they drastically reduce TJmax like AMD did.

 

[frozentundra123456]

The problem is definitely the TJmax being set so high.

Take the 3770k that I characterized for power vs. temperature vs. voltage vs. clockspeed.

With a TDP of 77W and a TJmax of 105C the maximum clockspeed Intel could set that particular 3770k to is 3.8GHz (all cores), and that is with the operating voltage optimized for LinX stability (meaning no engineering room for margin against parametric degredation over the operating lifetime of the CPU).

There's not much headroom left on the table IMO.

Now bump up that TDP from 77W to 95W and all you gain is just enough head-room to bump up the clocks to 4GHz from 3.8GHz.

The reason people think of their OC'ed processors as being proof that Intel is leaving lots of clockspeed on the table is that they are keeping the operating temperatures well below TJmax. That keeps the leakage currents quite low as well as enabling the user to set their operating voltage even lower than would otherwise be required for stable operation up to TJmax.

Intel doesn't have that luxury. Not unless they drastically reduce TJmax like AMD did.

Sorry, I dont really understand TJmax, or at least how lowering it would allow for higher clocks. Just intuitively I would think raising the TJmax would allow higher clocks, but obviously I am no engineer, so maybe you could explain it to me in layman's terms. Are you saying that the 3770k is running near its thermal limits at stock without better cooling?

 

[Idontcare]

Sorry, I dont really understand TJmax, or at least how lowering it would allow for higher clocks. Just intuitively I would think raising the TJmax would allow higher clocks, but obviously I am no engineer, so maybe you could explain it to me in layman's terms. Are you saying that the 3770k is running near its thermal limits at stock without better cooling?

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.

Think of it like this - if Intel says a 3770k has a TDP of 77W, operates at 3.5GHz, and has a max operating temp of 105C then they have a rather limited range of operating voltages that would be capable of making all that happen.

If the required voltage for stability at 3.5GHz and 105C is too high then they blow their TDP spec.

But what if they lower TJmax? What if they set it to 95C instead of 105C?

Well now they only need the operating voltage to be high enough to enable 3.5GHz stability at 95C, that allows them to set the voltage lower than would have been required if the CPU needed to be stable all the way up to 105C. Lower voltage means lower power usage.

Further, by only allowing the CPU to heat up to a max of 95C the leakage current is also reduced. Less leakage means lower power usage.

So everything revolves around the max allowed operating temp. Your voltage, your clockspeed, your power consumption, and your HSF costs.

Look at the FX8350 for comparison. Considerably more expensive HSF because they need the max temperatures to stay below 65C. Why 65C? Because the power usage would get too high if they allowed the 8350 to get above 65C, and it would require higher voltages too. (or they'd need to drop the clockspeed)

 

[Haserath]

...

 

[frozentundra123456]

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.

Think of it like this - if Intel says a 3770k has a TDP of 77W, operates at 3.5GHz, and has a max operating temp of 105C then they have a rather limited range of operating voltages that would be capable of making all that happen.

If the required voltage for stability at 3.5GHz and 105C is too high then they blow their TDP spec.

But what if they lower TJmax? What if they set it to 95C instead of 105C?

Well now they only need the operating voltage to be high enough to enable 3.5GHz stability at 95C, that allows them to set the voltage lower than would have been required if the CPU needed to be stable all the way up to 105C. Lower voltage means lower power usage.

Further, by only allowing the CPU to heat up to a max of 95C the leakage current is also reduced. Less leakage means lower power usage.

So everything revolves around the max allowed operating temp. Your voltage, your clockspeed, your power consumption, and your HSF costs.

Look at the FX8350 for comparison. Considerably more expensive HSF because they need the max temperatures to stay below 65C. Why 65C? Because the power usage would get too high if they allowed the 8350 to get above 65C, and it would require higher voltages too. (or they'd need to drop the clockspeed)

I think I understand, thanks for the explanation.