Am3+ is dead after...

[Dan_D]

They need to ditch the AM3+ socket and all the bullshit that goes with it.

 

[Zarathustra[H]]

"Every indication we have thus far is that Piledriver is faster than Bulldozer at the same clock, yet still slower per core than Thuban..."

....in Cinebench.

Oh i hear you. It's only one benchmark and its not real world use, but it's what we have, and in the grand scheme of things it's a pretty decent benchmark, especially for rendering/encoding type workloads.

 

[Zarathustra[H]]

They need to ditch the AM3+ socket and all the bullshit that goes with it.

Not a big fan of HT, eh?

 

[octoberasian]

Not a big fan of HT, eh?

I think it was Dan or someone else here that gave a good argument against HyperTransport. After looking it through on a few other sites, it's great for CPU-to-CPU communications, but CPU-to-memory or CPU-to-external I/O, not so much.

Another person said it's probably one, well one of many, reasons why there isn't PCI-E 3.0 native on the processor or the chipset-- lack of bandwidth from the HyperTransport bus. Trinity has an on-die northbridge with a single x16 PCI-E interface for the PCI-E slots. A lot of the changes done to Trinity also necessitated a new socket. I think doing away with HT on Trinity is the start of one of many changes in AMD's future processor line, but we may not see those changes immediately.

So, my thinking is that if AM3+ is to be done away with and Jim Keller can work some magic into Excavator, we might see native PCI-E 3.0, DDR4 and a better bus than HT, or no HT, on an entirely new socket.

But, that's going to be years away from now. One person in the Trinity reviews noted we won't see AMD catching up in single threaded performance until Steamroller or after that, so within two or three years.

 

[Dan_D]

Not a big fan of HT, eh?

I have no issues with Hypertransport beyond the constraints it puts on designers to place the DIMM slots so close to the CPU. Intel has this issue as well. My issue is with AM3's dated thermal design which dates back even further than it does. I don't like the CPU socket thermal solution mounting holes or clip system. I much prefer the symmetrical bolt-through setup on the Intel's. The problem with their setup is that it typically mandates modification to solutions designed for Intel hardware first and it creates a less than ideal installation on the AMD side.

And past that there is a more serious issue with the fact that maintaining all the legacy microcode in the BIOS / UEFI for a wider range of CPUs creates issues with the BIOS / Boards. This also increases the difficulty during the design phase at it forces AMD to make future CPUs compatible with a dated VRD implementation, or at the very least modify any new solution to include support for ancient CPUs. Working within the confines of existing design parameters is often harder than just doing something entirely new which only draws on lessons learned from the past. Not the designs themselves. I have to wonder how efficient Bulldozer might have been power wise had they thrown out the old VRD specs and started from scratch. It may also prevent them from matching Intel with iGPU support and PCIe controllers being added into the CPU. Again there are some pros and cons to that as well, but it's limiting at the very least.

I've also come to prefer LGA style sockets. Though my chief argument for them really only applies to CPUs which far eclipse the expense of motherboards, and obviously that's not really the case with AMD as they do not have such expensive, high end CPUs on the desktop side and do support the use of LGA sockets on the Opteron / Server side of things.

People may like the so called upgrade path offered by legacy sockets on newer CPUs, but they don't usually understand all the baggage that goes with them. Also, in many cases you still need a newer motherboard to access all the new features of the CPUs, and lose out on efficiency, functionality and features when coupling newer CPUs with older motherboard platforms. This is OK for bottom dollar upgraders but it can hit performance and in the long run I don't think it does very many people any good.

 

[Zarathustra[H]]

Another person said it's probably one, well one of many, reasons why there isn't PCI-E 3.0 native on the processor or the chipset-- lack of bandwidth from the HyperTransport bus. Trinity has an on-die northbridge with a single x16 PCI-E interface for the PCI-E slots. A lot of the changes done to Trinity also necessitated a new socket. I think doing away with HT on Trinity is the start of one of many changes in AMD's future processor line, but we may not see those changes immediately.

Yeah, Personally I think I'd rather have more PCIe lanes over HT than only 16 directly on the CPU, but I see what you mean.

But, that's going to be years away from now. One person in the Trinity reviews noted we won't see AMD catching up in single threaded performance until Steamroller or after that, so within two or three years.

If steamroller or even the gen after that catches up with Intel, I will eat my hat.

Keep in mind, Intel isn't sitting around doing nothing while AMD is rolling out construction equipment after construction equipment.

Even if Steamroller were here today, I'm not convinced it would catch them up to Ivy Bridge, and steamroller won't be here for another year, at which point it will be competing with Haswell.

Heck, I'm not even convinced that if we had an Excavator today, it would be competitive with Intel's current IB parts, and that's at least 2 years off...

We still don't have any real Piledriver FX benchmarks, but when Ivy Bridge has an almost 80% lead - overclocked to overclocked over current bulldozers in single threaded performance, that lead isn't going to vanish overnight, even if AMD were trying, and from what I'm seeing, it doesn't look like they are.

They have given up in competing against Intel in x86 performance, trying to carve themselves out a lead in the APU market instead.

 

[Dan_D]

They have given up in competing against Intel in x86 performance, trying to carve themselves out a lead in the APU market instead.

It's the only thing they have a realistic shot at doing unless another Next Gen Systems is found and AMD can buy them up fast enough to implement the technology before it becomes obsolete anyway. People often forget that AMD has virtually always been behind Intel in terms of performance and the one time they really weren't, they were using technology bought from someone else and engineers who worked for a more competitive company. (Digital Equipment Corporation or DEC.)

 

[Cecil]

To continue on with Dan_D's post about power requirements.
When BD first came out, I had a 990FXA-UD3 with a Thuban. Worked great with Thuban, because the UD3 was still using the same power layout as the 8 series boards. Dropped an FX-8120 in it and disabled power throttling (no BIOS option to do so, involved some work arounds), and even at stock settings, that was enough to kill the board and CPU.

The UD5 and higher had a newer/better power delivery setup that could handle BD fine, but the one based on the previous chipset couldnt handle it properly at stock.

Someone trying to keep the same board for four different CPU architectures is just asking for problems.

 

[Dan_D]

The 990FX chipset based boards are a great example. They were designed more with Bulldozer in mind but when we reviewed a few using older Phenom II X6's we ran into nothing but trouble with them. Many people reported the opposite problem. Especially with older 890FX chipset based boards supposedly sporting newer VRD implementations for Bulldozer. AM3+ is a mess. It took far too long to sort out and was handled poorly. As a reviewer I dread the thought of another generation of boards using the same dated hardware for newer CPUs. At least with Intel we don't have all these processor board / compatibility issues. And surprise surprise when they stuck with LGA775 for so long it was a mess often times too.

 

[Zarathustra[H]]

It's the only thing they have a realistic shot at doing unless another Next Gen Systems is found and AMD can buy them up fast enough to implement the technology before it becomes obsolete anyway. People often forget that AMD has virtually always been behind Intel in terms of performance and the one time they really weren't, they were using technology bought from someone else and engineers who worked for a more competitive company. (Digital Equipment Corporation or DEC.)

Agreed, heck, they started out more or less as a license manufacturer for Intel, for the 8086 and 286 parts. It wasn't until Intel canceled the deal with the 386, and threw them under the bus in 1986 that they designed their own 386 part not based on Intels hardware.

Once the legal battles were over later 386 and 486 parts were reverse engineered Intel chips, and the 586 parts were just marginally improved 486 chips.

It really wasn't until the K5 in 1996 they designed their own CPU (which was much slower than Intels offerings) as were the K6, K6-2 and K6-III's that followed.

The only time they were ever really competitive with Intel was from the launch of the K7 Athlon in 1999 until Intel came out with the C2D in 2006. Those were 7 great years to be a computer enthusiast because of that competition, but its the only time in history the company has made performance competitive CPU's.

The only reason they were able to pull this off was:

1.) As you already mentioned, they hired DEC's CPU architecture team as DEC was winding down; and

2.) Intel screwed the pooch big time with Netburst.

If both of these hadn't happened, AMD would never have been competitive with Intel.


For a while they maintained this competitiveness, beating Intel to such innovations as on-die memory controllers and the x64 instruction set. They had an opportunity to reinvest in their CPU designs and position themselves as a real long term CPU competitor to Intel, but instead they decided to spend their earnings from their competitive designs on buying ATI.

The ATI purchase led to not enough enough R&D spending for the Phenom development and a lackluster Phenom II, and this was repeated with Bulldozer, where they saved money by focusing entirely on the server platform, and releasing a Bulldozer/FX Desktop CPU based on the server model as an afterthought.

While this seems like a huge lost opportunity when it comes to the CPU space, maye it was the correct business decision to spend that money on ATi and get into APU's, I'm not sure.

 

[Dan_D]

While this seems like a huge lost opportunity when it comes to the CPU space, maye it was the correct business decision to spend that money on ATi and get into APU's, I'm not sure.

It's hard to tell. With the desktop segment dying, mobile CPUs with strong APU's might really be the future.

 

[reb00tin]

Is AM3+ technologically behind in some way? My brand new ASUS SABERTOOTH 990FX R2.0 with a 965BE idles at 68 Watt at the wall despite all its new digital power stuff, whereas my GIGABYTE Z77X-UD5H idles at 33 watt. The VRMs on the z77 are barely warm, but the Sabertooth draws 200+ watt full throttle.

 

[Zarathustra[H]]

Is AM3+ technologically behind in some way? My brand new ASUS SABERTOOTH 990FX R2.0 with a 965BE idles at 68 Watt at the wall despite all its new digital power stuff, whereas my GIGABYTE Z77X-UD5H idles at 33 watt. The VRMs on the z77 are barely warm, but the Sabertooth draws 200+ watt full throttle.

I noted this as well.

It is particularly interesting when comparing the new Trinity FM2 CPU's to FX Bulldozer AM3+ CPU's.

They are a similar enough design that you'd expect similar idle power draw, yet for some reason FM2 idles at ~30 Watt and bulldozer idles at ~70 watt.

Maybe HT uses a lot of power?

 

[Dan_D]

Is AM3+ technologically behind in some way? My brand new ASUS SABERTOOTH 990FX R2.0 with a 965BE idles at 68 Watt at the wall despite all its new digital power stuff, whereas my GIGABYTE Z77X-UD5H idles at 33 watt. The VRMs on the z77 are barely warm, but the Sabertooth draws 200+ watt full throttle.

The AMD CPUs aren't as efficient power wise as the Intel's are. The dated AM3+ design with all it's legacy CPU support is no doubt part of that problem.

 

[Tsumi]

I noted this as well.

It is particularly interesting when comparing the new Trinity FM2 CPU's to FX Bulldozer AM3+ CPU's.

They are a similar enough design that you'd expect similar idle power draw, yet for some reason FM2 idles at ~30 Watt and bulldozer idles at ~70 watt.

Maybe HT uses a lot of power?

Trinity FM2 uses an integrated northbridge like current Intel CPUs, so that is probably a major factor in the lower idle power consumption. AM3+ still has the northbridge on the motherboard.

 

[reb00tin]

Trinity FM2 uses an integrated northbridge like current Intel CPUs, so that is probably a major factor in the lower idle power consumption. AM3+ still has the northbridge on the motherboard.

I didn't know that.

It seems an unfair comparison, my i5-3570k is 22nm, the Phenom II 45nm, yet CPU power consumption alone at idle does not account for the diff at the wall.

 

[Dan_D]

Trinity FM2 uses an integrated northbridge like current Intel CPUs, so that is probably a major factor in the lower idle power consumption. AM3+ still has the northbridge on the motherboard.

And the chipset of any AM3+ motherboard is little more than a stepping change of the 8x0FX / SB850 series chipsets. Quite dated.

 

[xorbe]

yet CPU power consumption alone at idle does not account for the diff at the wall.

Some PSUs are really inefficient at super low wattages. I built a custom EPS splice-in cable (simultaneous V*A) just so I could actually measure the CPU power alone. But if the motherboard is hungry, not much one can do (like my 920's asus P6T deluxe).

 

[Dan_D]

Some PSUs are really inefficient at super low wattages. I built a custom EPS splice-in cable (simultaneous V*A) just so I could actually measure the CPU power alone. But if the motherboard is hungry, not much one can do (like my 920's asus P6T deluxe).

Motherboard power consumption can vary a lot even on boards with the same or similar chipsets. The amount of integrated hardware can severely increase power consumption. Also, disabling these devices in BIOS may not power them off. It will just prevent their OROMs from loading and eating up memory / system resources.

 

[pelo]

They need to ditch the AM3+ socket and all the bullshit that goes with it.

I hope so...

Actually, I'll take it further. They need to ditch FM2 after Kaveri as well. Unified North Bridge is the first step of a long series of steps they need to make in order to lower power consumption, increase efficiency and decrease overall platform cost. Much like Intel is doing with Haswell, AMD needs to integrate VRMs and more, like the South Bridge onto the die. Though this won't happen until later for Intel (Sky Lake? Broadwell?), if we're to judge by AMD's FM2 life expectancy, AMD won't be making any of these improvements for a period of 3 years...

I'd be perfectly content with dumping a Llano FM1 system or Trinity FM2 if it means they change their approach with respect to the GPUs dependency on DDR3 bus width and caching, because that would signify a significant increase in GPU performance and a decrease in power consumption and increased efficiency. Push the size of that GPU all you want, if it's still relying on DDR3 it's never going to get anywhere. It would be like having a 7970 with a dual 64-bit bus width relying on GDDR3. Great graphics being bottlenecked by a crappy IMC, slow RAM and a cramped bus width.

The GPU has a separate interface for sending memory requests that target the coherent system memory. The Fusion Control Link (or Onion) is a 128-bit (16B) bi-directional bus that feeds into a memory ordering queue shared with the coherent requests from each of the 4 cores. Onion runs at up to 650MHz for notebook variants of Llano (10.4GB/s read + 10.4GB/s write) and 492MHz for Zacate. An arbiter in the IFQ is responsible for selecting coherent requests (based on memory ordering) to send to the memory controller. Desktop versions of Llano will probably run Garlic and Onion faster still, given the extra power budget.
The memory controller arbitrates between coherent (i.e. ordered) and non-coherent accesses to memory. Llano has two 64-bit channels of DDR3 memory that must operate independently, while the smaller Fusion cousin only has a single channel. The GPU memory is interleaved across both channels for maximum streaming bandwidth and requests will close DRAM pages after an access. In contrast, system memory is optimized for latency and locality; contiguous requests will tend to stay to one memory channel and keep DRAM pages open. The memory can run up to 1.86GT/s for a total of 29.8GB/s memory bandwidth on Llano. It also contains an improved hardware prefetcher that tracks 8 different strides or sequence of strides and speculatively fills into the memory controller (rather than the caches).
In contrast, Sandy Bridge has tigher integration – using the on-die ring interconnect and L3 cache. Data is passed through the ring interconnect, but can be shared either through the cache or memory. The ring interconnect is 32B wide with 6 agents and operates at the core frequency (>3GHz). Data usually coming from either the 4 slices of the L3 cache or the memory controller, which resides in the system agent. The peak bandwidth is over 400GB/s, but the practical bandwidth since many accesses have to go through multiple stops on the ring interconnect. The Sandy Bridge power management is also fully unified for both CPU and GPU, so that when one is idle, the other may ‘borrow’ the thermal headroom.

Although some of that has changed, like the dynamic turbo being shared between CPU and GPU rather than just GPU and an IMC that supports 1866mhz DDR3, it's still pretty much the same in Trinity as it was in Llano. They need to address that and quickly. Unfortunately, because FM2 is supposed to live through 2 generations of APUs, they're not going to make the changes that need be made for another 3 years
http://www.realworldtech.com/fusion-llano/2/

 

[octoberasian]

Haswell, if Fudzilla is correct, will have its own VRMs on-die. I think it was posted in the Intel forums here as well. There was some controversy about this because some thought this would kill off enthusiasts who like to overclock their processors.

Haswell is bringing a lot of things to the processor. Intel is hoping to fully integrate a south bridge along with the NB on the same die. Like you said, that's probably not until Broadwell or Skylake afterwards.

AMD is moving towards combining both the GPU and CPU as one cohesive unit. That may happen by Excavator or later, where the GPU and CPU are one single unit. The idea is that since the current Bulldozer/Piledriver modules and Steamroller modules have a shared FPU, the GPU portion can handle the compute part for the FPU. If you can just imagine a successor to the GCN cores as the future replacement for an FPU in a processor, that's going to boost desktop computing power by a lot.

I'm hoping that by the time Excavator rolls out, there is no Hypertransport and we get a new socket, and an on-die northbridge with PCI-Express support like Intel has had for two generations now. Integrating VRM and southbridge would probably be Excavator or later at the current rate AMD is going at integrating features Intel has already gotten in their current and upcoming processors.

So, if Intel has 3D transistors and microOP cache, AMD is going to have them in their processors at least a year or two away after Intel introduces them. Steamroller will have the microOP cache, and since GF just started on 3D transistors for ARM, AMD might not see it until Excavator or the architecture after that.

AMD may still be behind, but I'm sure they'll probably be feature-parity with Intel within the next five years at the current rate. Performance-wise is an entirely different issue. Hopefully AMD will be on par with Intel's processors the same time both are released to market.

It'd be interesting to see AMD's Excavator perform as good as an Intel architecture released at the same when Excavator processors start hitting the market. That's wishful thinking though, and may or may not happen.

 

[pelo]

They've already unified the North Bridge

It was part of the reason for requiring a new socket, FM2. Had they continued this path, I think they would have been on a much more aggressive timeline in catching up to Intel. Instead, it looks like they're just going to address CPU performance and it might even come at the cost of GPU performance because they're not changing their dependency on DDR3 for at least 2 years.

 

[Zarathustra[H]]

Haswell, if Fudzilla is correct, will have its own VRMs on-die. I think it was posted in the Intel forums here as well. There was some controversy about this because some thought this would kill off enthusiasts who like to overclock their processors.

I can definitely see this.

While it will be a good thing for budget systems. (It will lower cost of motherboards, and bring them more consistent and reliable VRM's) it very well might hurt the more extreme overclockers.

Many good overclocking boards are so, because they - among other things - have better VRM's than other boards.

Intel is likely going to put solid and reliable VRM's in Haswell for running CPU's at stock, but they are unlikely to over spec them just in case someone wants to overclock, which is what the high end overclocking motherboards makers do.

This will likely result in fewer shitty experiences due to shitty motherboards, but also in fewer fantastic overclocking experiences due the VRM's being the same for everyone.

That is - unless - Intel provides a way in their socket for motherboards to bypass the on die VRM's and go directly to the CPU, but I doubt they'd do this. It would cost them more, and give them little to no benefit.

 

[pelo]

They're not integrating all of the VRMs on-die. I believe it's only minimal integration for Haswell, though this will certainly pick up later on.

Frankly, I'd take the decrease in power consumption and higher stock clocks and IPC over extreme overclocks. People overclock for the fun of it, rarely because they need the extra performance (which can be solved by even more aggressive turbo scaling, btw).

I've always thought it was more fun to see how much you can get out of very little power

 

[Dan_D]

They're not integrating all of the VRMs on-die. I believe it's only minimal integration for Haswell, though this will certainly pick up later on.

Frankly, I'd take the decrease in power consumption and higher stock clocks and IPC over extreme overclocks. People overclock for the fun of it, rarely because they need the extra performance (which can be solved by even more aggressive turbo scaling, btw).

I've always thought it was more fun to see how much you can get out of very little power

I don't care how much electricity the thing uses or even how much heat it generates so long as I can dissipate it sufficiently. I want maximum performance and I don't really care how this is achieved.

 

[xorbe]

I don't see how the VRMs could be on-die ... that would greatly increase heat density. Maybe it works for stock, but it would kill overclocking without water cooling.
edit: Ah, I see now ... nevermind.

 

[pelo]

I don't care how much electricity the thing uses or even how much heat it generates so long as I can dissipate it sufficiently. I want maximum performance and I don't really care how this is achieved.

The days where the majority were in your shoes has greatly diminished. Intel's chasing the money, and that means using a single overarching architecture and core design throughout all of their products, low end to desktop. Considering the rather huge lead they have over AMD in the desktop and server arena, they're not going to really care about the enthusiast much. -- crappy TIM? Who gives a shit Is it cheap? Use it.

Think about the Sandy and Ivy CPU architectures. Great performers, sure, but where were they even better? The graphics capabilities were 2x greater than the prior gen and the efficiency improvements were stellar. Those are mobile goals, not desktop enthusiast ones.

The days where overclocking, high TDP, high clock speed processors were the design goals have been left behind years ago. Haswell is an efficiency architecture first and Intel made a point of stating the architecture is shared throughout all of their products. That trend is going to continue

-- I also think it's a bit silly to think of overclocking as an enthusiast-only shindig. How hard is it to overclock today? Any idiot can do it. It's not like the olden days where we needed a graphite pen. Those were the enthusiast days, today's overclocking to 4.5ghz is akin to tying your shoes.

 

[RamonGTP]

If we go by enthusiast track record of predicting overclockability on a new Intel product, we have nothing to worry about.

 

[Zarathustra[H]]

crappy TIM? Who gives a shit Is it cheap? Use it.

I thought this was pretty much not the case with IB, just suspected early on.

From what I recall, reading the threads on the forums from people who removed the cover, wiped the TIM and replaced it with a direct on chip cooler and good TIM, found that their results didn't improve much or at all...

Think about the Sandy and Ivy CPU architectures. Great performers, sure, but where were they even better? The graphics capabilities were 2x greater than the prior gen and the efficiency improvements were stellar. Those are mobile goals, not desktop enthusiast ones.

The days where overclocking, high TDP, high clock speed processors were the design goals have been left behind years ago. Haswell is an efficiency architecture first and Intel made a point of stating the architecture is shared throughout all of their products. That trend is going to continue

Agreed, but at the same time, IPC and clocks have been improving with every generation since - well - the first Yonah Intel Core CPU's in 2006. If you consider only IPC, that's improved with every generation since Pentium 4.

So, while IPC and clock improvements are not the top goals, they have been going hand in hand with mobile improvements, and I'm OK with that.

-- I also think it's a bit silly to think of overclocking as an enthusiast-only shindig. How hard is it to overclock today? Any idiot can do it. It's not like the olden days where we needed a graphite pen. Those were the enthusiast days, today's overclocking to 4.5ghz is akin to tying your shoes.

Well, it's certainly easier than it once was, but remember, the average user is still afraid of the BIOS and has no idea how to stress test an overclock.

 

[pelo]

Zara, check out IDC's thread on Anandtech. He removes the TIM and takes analytical data regarding the improvements.

http://forums.anandtech.com/showthread.php?t=2261855

 

[RamonGTP]

I thought this was pretty much not the case with IB, just suspected early on.

From what I recall, reading the threads on the forums from people who removed the cover, wiped the TIM and replaced it with a direct on chip cooler and good TIM, found that their results didn't improve much or at all...

You stopped reading after the first random person did this. There have been several others that have delid their IB with great results. That first guy is essentially the only person who didn't get results iirc.

 

[Zarathustra[H]]

You stopped reading after the first random person did this. There have been several others that have delid their IB with great results. That first guy is essentially the only person who didn't get results iirc.

Ahh,

Fair enough. I don't have an IB chip, so when the info was first flying around I read it cause I was curious, but I didn't keep up with it, as it really wasn't relevant to me.

 

[Dan_D]

The days where the majority were in your shoes has greatly diminished. Intel's chasing the money, and that means using a single overarching architecture and core design throughout all of their products, low end to desktop. Considering the rather huge lead they have over AMD in the desktop and server arena, they're not going to really care about the enthusiast much. -- crappy TIM? Who gives a shit Is it cheap? Use it.

Think about the Sandy and Ivy CPU architectures. Great performers, sure, but where were they even better? The graphics capabilities were 2x greater than the prior gen and the efficiency improvements were stellar. Those are mobile goals, not desktop enthusiast ones.

The days where overclocking, high TDP, high clock speed processors were the design goals have been left behind years ago. Haswell is an efficiency architecture first and Intel made a point of stating the architecture is shared throughout all of their products. That trend is going to continue

-- I also think it's a bit silly to think of overclocking as an enthusiast-only shindig. How hard is it to overclock today? Any idiot can do it. It's not like the olden days where we needed a graphite pen. Those were the enthusiast days, today's overclocking to 4.5ghz is akin to tying your shoes.

I know my thoughts on the subject are quickly becoming more of a minority than they had been before.

 

[octoberasian]

I don't see how the VRMs could be on-die ... that would greatly increase heat density. Maybe it works for stock, but it would kill overclocking without water cooling.

The idea is better power usage and TDP.

I'm going to guess here since I don't remember the entire reasoning behind this, but it's to do several things:

• Reducing board parts for increased efficiency.
• Better power regulation based on work load and temperature.
• Ensuring that when the load increases, processor can automatically and properly clock up or down as long as it stays within its max TDP by using as little power as possible for each processor state.

If you recall, more and more processors in the last few years have had the ability to shutdown certain parts of the processor to get better power efficiency and performance-per-watt. We've seen this a lot with Intel's Sandy Bridge and then Ivy Bridge. AMD has done similarly to its FX processors beginning with Bulldozer based modules and its Llano APUs. For both companies, though their implementations are different, come down to the ability to shut off cores and graphics cores (if on-die GPU exists); to lower or raise the frequency of the processor depending on work load; and if in the case of Intel, the ability to shut off certain parts of the core itself.

An on-die VRM is meant to give the processor more control over how much voltage it uses when it does any of its power-saving features. And, that's where all the red flags were being thrown up in the air by overclocking enthusiasts. Does that mean that enthusiasts will not be able to properly overclock their processor anymore? Does that mean they can't increase the voltage of the processor as they bump up the frequency? Will Intel's on-die VRM implementation prevent unnecessary voltage increases?

It's a bunch of things that some users on another website were and still discussing about. We won't know for sure until sometime next year when Haswell is released somewhere in the first half of 2013.

But, when you read something like this, where Intel's Haswell graphics part does the same amount of work at 7W as Ivy Bridge's 3000-series HD graphics core does at 17W, is really impressive how far Intel has gotten in increasing performance-per-watt efficiency. That and Haswell's processors will be rated between 35W to 77W TDP like Ivy Bridge, with 100W TDP reserved for higher-end processors. Just going by this alone, a Socket 2011 Haswell may use a lot less power than a Socket 2011 Sandy Bridge or Ivy Bridge processor.

 

[Dan_D]

The idea is better power usage and TDP.

I'm going to guess here since I don't remember the entire reasoning behind this, but it's to do several things:

• Reducing board parts for increased efficiency.
• Better power regulation based on work load and temperature.
• Ensuring that when the load increases, processor can automatically and properly clock up or down as long as it stays within its max TDP by using as little power as possible for each processor state.

If you recall, more and more processors in the last few years have had the ability to shutdown certain parts of the processor to get better power efficiency and performance-per-watt. We've seen this a lot with Intel's Sandy Bridge and then Ivy Bridge. AMD has done similarly to its FX processors beginning with Bulldozer based modules and its Llano APUs. For both companies, though their implementations are different, come down to the ability to shut off cores and graphics cores (if on-die GPU exists); to lower or raise the frequency of the processor depending on work load; and if in the case of Intel, the ability to shut off certain parts of the core itself.

An on-die VRM is meant to give the processor more control over how much voltage it uses when it does any of its power-saving features. And, that's where all the red flags were being thrown up in the air by overclocking enthusiasts. Does that mean that enthusiasts will not be able to properly overclock their processor anymore? Does that mean they can't increase the voltage of the processor as they bump up the frequency? Will Intel's on-die VRM implementation prevent unnecessary voltage increases?

It's a bunch of things that some users on another website were and still discussing about. We won't know for sure until sometime next year when Haswell is released somewhere in the first half of 2013.

But, when you read something like this, where Intel's Haswell graphics part does the same amount of work at 7W as Ivy Bridge's 3000-series HD graphics core does at 17W, is really impressive how far Intel has gotten in increasing performance-per-watt efficiency. That and Haswell's processors will be rated between 35W to 77W TDP like Ivy Bridge, with 100W TDP reserved for higher-end processors. Just going by this alone, a Socket 2011 Haswell may use a lot less power than a Socket 2011 Sandy Bridge or Ivy Bridge processor.

What we may see out of LGA2011 Haswell CPUs is a power envelope that's not too different from today's current hex core parts, but with 8 cores or more. Though I'd expect idle efficiency to be a bit better than current generation parts are.

 

[pelo]

I'm going to guess here since I don't remember the entire reasoning behind this, but it's to do several things:

SoC improvements. On Intel's recent IDF they made a point of stating they're using a single architecture throughout their product lines. The efficiency improvements needed for the low end are going into their high end processors. Remember that they're still a ways away from getting x86 down to a level where it's both efficient on power and cheap.

That 10W Haswell chip sports a GT4 iGPU. They also said the improvements on the 17W Haswell GT3 are about 30% over the same ULV Ivy processor.

 

[octoberasian]

What we may see out of LGA2011 Haswell CPUs is a power envelope that's not too different from today's current hex core parts, but with 8 cores or more. Though I'd expect idle efficiency to be a bit better than current generation parts are.

That would be impressive right there.

SoC improvements. On Intel's recent IDF they made a point of stating they're using a single architecture throughout their product lines. The efficiency improvements needed for the low end are going into their high end processors. Remember that they're still a ways away from getting x86 down to a level where it's both efficient on power and cheap.

That 10W Haswell chip sports a GT4 iGPU. They also said the improvements on the 17W Haswell GT3 are about 30% over the same ULV Ivy processor.

I remember reading something about that.

 

[pelo]

They announced that they originally had 12-core designs. I'm not sure whether that will come to fruition on 2011, as having that many cores and cache is going to drop clock speeds by a good amount. Maybe 10-core? 8-core? I'm thinking the planned 12-core parts would be strictly server.

 

[Cecil]

2011 is dead to me. They arent releasing IB-E until end of next year. Long after Haswell will be out. So their high end 2011 platform wont have the single threaded performance their mainstream platform (1150) does. They should have brought IB-E this year, then Haswell mid way next year, and Haswell-E end of next year.

 

[Dan_D]

That would be impressive right there.



I remember reading something about that.

It wouldn't shock me. With die shrinks and increased core counts in the past this has occured a few times. Hell my QX9775 processors had a TDP of 150watts. Today's quad cores have a TDP of less than half that. The Core i7 980X had a TDP which was in line with previous generation quad cores like the Core i7 975EE but with two more cores. (As I recall.)