AMD CPU speculation... and expert conjecture

[Cazalan]

And the biggest gain in BF4 was an i7-3970X Extreme with two R9 290X cards. 58%

They're actually making it worth buying these $1000 CPUs. What is AMD thinking?!?!?

 

[-Fran-]

That 58% should catapult the 290X's in P/P ratio over the Titans and 780TI in SLI, right? I wonder how MANTLE will work out the framing issue. Since it's a whole new driver part, it should be included, right? (Frame Pacing thingy).

Cheers!

 

[Cazalan]

8x Steamroller cores @2GHz would hit ~60 SPEC_int
8x A57 cores @2GHz hits ~80 SPEC_int

All those who said that ARM was low performance and that couldn't match x86 in decades were wrong. Another of my predictions in this thread has been verified.

If you want to see a modern x86 server CPU then check here. http://www.spec.org/cpu2006/results/rint2006.html

E5-2470 v2 @2.40 GHz has a SPECint_rate of 364.

How long until someone slaps 36 ARMv8 cores together to match that?

 

[Cazalan]

That 58% should catapult the 290X's in P/P ratio over the Titans and 780TI in SLI, right? I wonder how MANTLE will work out the framing issue. Since it's a whole new driver part, it should be included, right? (Frame Pacing thingy).

Cheers!

By enabling Mantle for Intel what did they just do to the i3?

 

[de5_Roy]

By enabling Mantle for Intel what did they just do to the i3?

oh !@#$ ?

 

[blackkstar]

That 58% should catapult the 290X's in P/P ratio over the Titans and 780TI in SLI, right? I wonder how MANTLE will work out the framing issue. Since it's a whole new driver part, it should be included, right? (Frame Pacing thingy).

Cheers!

By enabling Mantle for Intel what did they just do to the i3?

Mantle numbers are going to create absolute chaos with the whole "yo bro I got dis sweet gaymen PC it's Intel + Nvidia + Corsair with water cooling bro!" crowd. It's going to make a lot of people switch from Nvidia to AMD. People that would never consider AMD CPUs in the first place. Because lets be honest, Core i3 is not cost effective at any price range in comparison to AMD's offerings for gaming and if you go i3 or (god forbid) Pentium/Celeron, you're drinking Intel kool aide. And it's been my experience in lurking in forums that newbs to the enthusiast community absolutely love the Nvidia/Intel/Corsair combination. Seeing Radeon get a 50% boost in a high end review using Intel hex is going to be a massive blow to Nvidia.

Here's a review with 290x running on 3960x.
http://www.guru3d.com/articles_pages/gigabyte_radeon_r9_290x_review,20.html
Multiply those scores by 1.58 (or if you're conservative even 1.40) and see what happens.

If you're too lazy to do the math, that puts 290x at 107.44fps (optimistically).

If someone is looking to play Mantle enabled games, AMD is going to be a clear winner. And if the game isn't Mantle enabled, you're not going to end up with a horribly slow card either.

@gamerK, I know that you need to emulate and lot of chips, but I'm referring to what I thought were the largest bottlenecks. Generally, from what I understand, the major bottleneck is usually the CPU emulation itself. The other things usually aren't that bad, but I do know audio can be a problem as well.

Mantle's threading of rendering is all abstracted for the coder. At the very least I was thinking it would alleviate a bottleneck. Dolphin has tried by doing things like GPU accelerated texture decoder. Emulation has a lot of things that depend on syncing and single threaded performance, but IMO any chance you get on spreading the load around should be used. Switching from DX11 to Mantle should at least help move towards figuring things out and getting things to scale better.

Historically, the developers who get applications to scale to multiple cores have been the most successful ones.

 

[gamerk316]

@gamerK, I know that you need to emulate and lot of chips, but I'm referring to what I thought were the largest bottlenecks. Generally, from what I understand, the major bottleneck is usually the CPU emulation itself. The other things usually aren't that bad, but I do know audio can be a problem as well.

CPU emulation for most of the simpler emulators is actually quite simple; you take a group of instructions that does some purpose, and wrap it around an equivalent X86 instruction. Given the somewhat limited instruction sets these CPU's have (most being RISC CPU's, after all), this is more or less a cheap operation to do in the grant scheme of things.

Granted, running pure interpreter mode for some of these CPU's [PS2 in particular; try running PCSX2 in interpreter mode and enjoy 2 FPS] is killer, but HLE is often good enough. Synchronization kills you, and there's nothing you can do about it. That's why a Pentium 3 can handle N64 emulation, but you need a top tier Pentium 4 to run BSNES at full speed; that extra precision kills performance.

 

[griptwister]

Holy cwap. Those mantle numbers are insane.

 

[esrever]

The frame pacing is much better under mantle. Also there seems to be decent improvement with a midrange CPU and GPU.

 

[jimmysmitty]

But they are under strict competition. The era of apus is dawning and, like x64, AMD is leading the charge. Like it or not (I was disappointed at first) the time of dCPU is coming to a close and like single core cpus it's going the way of the dodo. There will be no steamroller fx due in the next two years not because amd had given up on HEDT but there simply is no demand for 8 core cpus anymore and besides who needs 8 cores when you have 12 core apus right? Lol

If you base the performance of the IGP on the APU to be the most important part of it, then yes AMD is ahead of Intel. But what Intel has is process lead and uses less power. A i7-4770K beats out FX-8350 in the majority of CPU applications, has decent integrated graphics for entry level gaming and uses half the power.

Power is a major factor that people leave out. I wouldn't buy a APU for gaming, I have a desktop for that and soon I might be able to just stream games via Steam on my home network so my HTPC wouldn't need anything more than what Intel offers but rather a good wireless or wired connection.

Intel is smarter than people think. They do more than CPUs and have been a driving force in the advancement of many technologies people assume were other companies, such as PCIe, SATA, USB and NICs for example. They have been around long enough to see where markets will go and while they were a bit late to the UMD party, they have the resources to jump in and push hard.

 

[palladin9479]

Granted, running pure interpreter mode for some of these CPU's [PS2 in particular; try running PCSX2 in interpreter mode and enjoy 2 FPS] is killer, but HLE is often good enough. Synchronization kills you, and there's nothing you can do about it. That's why a Pentium 3 can handle N64 emulation, but you need a top tier Pentium 4 to run BSNES at full speed; that extra precision kills performance.

That's because BSNES sucks. The guys at ZSNES did it the best, they had to use ASM though. Was able to run a SNES at full emulated speed on a K6-2 400. People using HLL to write emulators is what kills performance more then anything, to get any sort of real speed you need to code the HW emulation components in ASM.

The sync issue really depends on the HW and system being used, the newer generation systems are actually less sensitive to sync issues due to their slightly abstracted nature. It's still going to be a really large PITA though, your never going to run full parallel emulation, just ain't gonna happen. What can be done is have the primary thread do the HW synch and off lead individual chip emulation to their own threads. You end up with a ton of locks and pure lose efficiency but gain in scalability. The gains are different for each platform and might be more trouble then their worth.

 

[palladin9479]

Power is a major factor that people leave out. I wouldn't buy a APU for gaming, I have a desktop for that and soon I might be able to just stream games via Steam on my home network so my HTPC wouldn't need anything more than what Intel offers but rather a good wireless or wired connection.

That will never work well. Networks' aren't magical data providers, their are very real limits to what you can put on them. 1920X1080x32 comes to 66,355,200 bits per frame. 60 FPS would require 3797 Mbps of bandwidth, before 10/8 encoding happens. Inline dynamic compression forces quality to take a large hit (think youtube video) and essentially downscales the picture and upscales it at the distant end forcing a loss of information. And that's just layer 4+, at layer 3 your data is being packetized with latency being introduced. You want to talk about "input lag".

We have several systems here that digitize video signals and send them over IP to other sources, it works well enough for communication but absolutely would not work for HD gaming.

In the "power" department APU's beat any dCPU + dGPU combo. 45W for what the 7600 puts out is insane. It's why I recommend APU for SFF / power + space limited setups, it only makes sense there. Once your got a large flexible power / space budget then dGPU + dCPU work better as you can afford larger cooling surfaces.

 

[Lessthannil]

I wouldn't take the advertised TDPs as what they pull under load, if that was what you were getting at.

http://www.bit-tech.net/hardware/cpus/2014/01/14/amd-a8-7600-kaveri-review/12

It would of been really awesome if it only did pull 45w on load but it seems way too good to be true. Not even Intel's 22nm + Haswell can do that unless you disable cores, downclock and undervolt. However, the power consumption great considering that this is a revision of the marred Bulldozer, is 28nm, and has a relatively big GPU on it.

Anyway, the GHz race was a race to nowhere and it was going against the grain of the market that is looking for smaller, more power efficent chips without sacrificing power.

 

[palladin9479]

I wouldn't take the advertised TDPs as what they pull under load, if that was what you were getting at.

http://www.bit-tech.net/hardware/cpus/2014/01/14/amd-a8-7600-kaveri-review/12

It would of been really awesome if it only did pull 45w on load but it seems way too good to be true. Not even Intel's 22nm + Haswell can do that unless you disable cores, downclock and undervolt. Its great considering that this is based off of Bulldozer and that its on 28nm. The GHz race was a race to nowhere and it was going against the grain of the market that is looking for smaller, more power efficent chips without sacrificing power.

That's because people don't understand what TDP means. TDP is the expected amount of heat that will need to be removed in order for the CPU to continue operating. Heat is nothing but electric power that has been converted into heat via ohmic heating, so quite literally waste heat = power consumption. A CPU that generated 45W of waste heat has used exactly 45W of electricity.

However, for the power consumption tests we re-enable everything in order to gauge real-world power draw. The power draw is measured via a power meter at the wall, so the numbers below represent the total system power draw from the mains, not the power consumption of a CPU itself. Measuring the power draw of any individual component in a PC is tricky to impossible to achieve.

CPU's require electricity to work, they require that electricity at a certain voltage and will pull an equivalent amount of amperage as needed. It's the MB and PSU's job to convert AC power into useful DC power to feed the system components. No conversion is 100% efficient unless your in a superconducting environment. So you get inefficiencies along the way, from the initial AC-DC conversion to the various DC-DC conversions along the way. Also PSU's are often sold based on maximum draw but that's not their more efficient draw. A PSU needs to operate at between 25 to 75% of it's draw to be most efficient, some are even tighter at 35~60% of draw. Then you have the MB circuitry, chips, bus's, timings and so on / so forth. All of that contributes to power draw, especially when load goes up.

If you look at that chart with this in mind you can actually see it in action. The 45W 7600 in that system has a total draw of 46W. At full load it becomes 90W making a 44W difference which match's to it's stated 45W TDP. Most of that draw would be attributed to the iGPU unless you were OCing the CPU.

Another thing to remember about TDP is that it's not an absolute limit, it's just a guideline on how much waste heat to expect to exhaust in most scenarios. It's also a useful guide on the amount of power a particular part will draw from the main PSU. People like to throw 450W PSU's onto everything but they are horribly inefficient for powering mini-ITX low power systems. You can use a 90W pico-PSU to power a system with a 45W 7600 and still have headway for memory / MB / HDD and not exceed 80% of the PSU's capacity.

 

[juanrga]

AMD is providing also some MANTLE performance gains for Kaveri+290

BF4: 40.9% (1080p) and 40.1% (1600p) performance improvement under Ultra settings and 4xAA on the AMD A10-7700K with an AMD Radeon™ R9 290X.

StarSwarm: 319% (1080p) and 281% (1600p) performance improvement in the “RTS” test on Extreme settings with the AMD A10-7700K and an AMD Radeon™ R9 290X.

http://www.brightsideofnews.com/news/2014/1/29/amd-rele...

Don't forget the not CPU bottlenecked outcomes:

GPU-limited scenario: 2.7% (1080p) and 1.4% (1600p) performance improvement under Ultra settings and FXAA on the Core i7-4960X with an AMD Radeon™ R7 260X

GPU-limited scenario: 5.1% (1080p) and 16.7% (1600p) performance improvement in the “RTS” test on Extreme settings with the Core i7-4960X and an AMD Radeon™ R7 260X

Who was the one here who predicted all the gains would be seen on lower tier systems? Almost nothing on the high end, as expected.

Also worth noting: Who would buy a Kaveri, then pair it with a high-end dGPU? So I'd wager even the CPU bound numbers are inflated; use a PD CPU and a slightly lower GPU (7750?), and I'd expect maybe 10-15% gains, at best. Looks like AMD is testing in an artificially bad and unrealistic CPU bottlenecked situation.

MANTLE's aim is to eliminate the CPU bottleneck generated by the bloatware layer of DX. If you are testing a top eXtreme intel chip with a low end graphics card, you are in a GPU bottleneck situation and MANTLE provides only single digit gains.

However, if you test an eXtreme intel chip with a pair of 290X then you are again in a CPU bottleneck situation and MANTLE provides double digit gains: BF4 MANTLE is 58% faster for an i7-3970X Extreme with two R9 290X cards.

Your predictions about MANTLE are destroyed once again. Myself predicted MANTLE would bring 30--50% gains to BF4. Kaveri with an AMD Radeon™ R9 290X falls in the middle with 41%. The 58% for an i7-3970X Extreme with two R9 290X cards is ~8% on top of my maximum predicted value. The 58% destroys your "almost nothing on the high end" prediction.

Finally, I believe that you also missed this pre-build Kaveri plus a 290

http://www.tomshardware.com/news/cyberpowerpc-zeus-mini-kavari-haswell-pc-gaming,25890.html

 

[juanrga]

It looks better than I expected: the A57 core is about 35% faster than the jaguar core with the bonus of consuming much less power! This implies

8x Steamroller cores @2GHz would hit ~60 SPEC_int
8x A57 cores @2GHz hits ~80 SPEC_int

Thus A57 cores are above the Steamroller level of performance (I predicted they would be at the same level, but A57 cores are finally better).

Regarding floating point:

8x Steamroller cores @2GHz offers maximum of 128 GFLOPS
8x A57 cores @2GHz offers maximum of 128 GFLOPS

It is now made evident why AMD abandoned the 4/6/8 core Piledriver based Opterons, why is releasing the new 12/16 core Warsaw Opterons only for legacy customers, and why doesn't release any Steamroller based Opteron. An 8-core Opteron obtained from dual packaging quad-core Kaveri CPUs (plus extra L3 cache) would be slower and inefficient than the new Opteron A1100. And without Steamroller Opteron there is no Steamroller FX.

It is worth to mention that the 16 core Warsaw Opteron offers 2.6 GFLOPS/W whereas the new 8 core Opteron A1100 offers 5.12 GFLOPS/W, which is ~4x the efficiency of Piledriver! Recall Steamroller is only 10-15% more efficient than Piledriver.

All those who said that ARM was low performance and that couldn't match x86 in decades were wrong. Another of my predictions in this thread has been verified.

Those who claimed here that a friend at AMD supposedly said him that ARM is an experiment would read what AMD said during the presentation of the new Opterons:

And custom cores such as Nvidia Denver will be better than standard A57. There are rumors that AMD will release custom cores for Cambridge SoC.

that means wiki is right
i had suspicions too
as arm arch is known for low power consumption, and cortex-a15 and krait are hitting >2GHz clock at 4 cores using power in single digit (watt) with ipc per core of 3.2-3.5mips/mhz compared to 3.78 of bulldozer (wiki is showing 2.9 for piledriver )

http://en.wikipedia.org/wiki/Instructions_per_second
amd's cpus are getting trashed by arm's cpu in ipc per core and in energy efficiency too
performance of mobile cpu's and gpu's are increasing very quickly as compared to desktop

the gpu in tegra k-1 is supposed to have 365gflops which is better than hd4400 desktop gpu (max 260gflops @1350Mhz) and is equal to radeon 7640G (350gflops @685mhz) :lol:

so will we be playing GTA 5 on mobile devices before its desktop launch :lol:

ARM is a high-efficiency design. It provides great performance in a given power constraint.

Performance = efficiency x power

For 1--2W power constraint, ARM provides more performance than x86 consuming the same 1--2W. We see now that 25W ARM provides about 2.8 times more performance than jaguar x86 cores (22W). Future 50--100W ARM chips will provide more performance than similar rated x86 chips.

I would take those Wiki numbers with a grain of salt, specially when they put Piledriver FX-8350 much slower than bulldozer FX

According to my numbers the Piledriver FX-8350 has ~3.6 DMIPS/MHz cores. And I was expecting the A57 core to be a 4.1 DMIPS/MHz design. Thus I was expecting it to be at the Steamroller core level of performance. My original prediction was

8 A57 @2GHz ~ 4 Steamroller @4GHz

However numbers shown by AMD say that it is about 1/3 faster than Steamroller.

Intel has a problem as well. Apple with its Cyclone A7 showed how a dual core offered the same performance than their last Intel quad-core clocked higher.

Some scores leaked in the web show Tegra K1 iGPU beating a higher TDP rated Haswell i5 iGPU. But I am more interested in the Denver cores, with a 7-wide design and Nvidia claiming that it will be faster than any A57, the Denver core could be at the Haswell level of performance in integer performance.

Spoiler

8x Steamroller cores @2GHz would hit ~60 SPEC_int
8x A57 cores @2GHz hits ~80 SPEC_int

All those who said that ARM was low performance and that couldn't match x86 in decades were wrong. Another of my predictions in this thread has been verified.

If you want to see a modern x86 server CPU then check here. http://www.spec.org/cpu2006/results/rint2006.html

E5-2470 v2 @2.40 GHz has a SPECint_rate of 364.

How long until someone slaps 36 ARMv8 cores together to match that?

Do you really believe that AMD needs 49 Steamroller cores to match an Ivy Bridge Xeon?

Besides that, I note how you pretend to compare the raw performance of a 25W SoC to a 95W CPU.

We could invert your 'logic' and compare a 95W ARM SoC to a 25W Intel CPU... but if you want know what will be beat the high-end Xeons, google Nvidia Boulder...

 

[colinp]

Time to open the thread "AMD Excavator speculation... and conjecture... and totally accurate retrospective predictions"

 

[gamerk316]

Granted, running pure interpreter mode for some of these CPU's [PS2 in particular; try running PCSX2 in interpreter mode and enjoy 2 FPS] is killer, but HLE is often good enough. Synchronization kills you, and there's nothing you can do about it. That's why a Pentium 3 can handle N64 emulation, but you need a top tier Pentium 4 to run BSNES at full speed; that extra precision kills performance.

That's because BSNES sucks. The guys at ZSNES did it the best, they had to use ASM though. Was able to run a SNES at full emulated speed on a K6-2 400. People using HLL to write emulators is what kills performance more then anything, to get any sort of real speed you need to code the HW emulation components in ASM.

The sync issue really depends on the HW and system being used, the newer generation systems are actually less sensitive to sync issues due to their slightly abstracted nature. It's still going to be a really large PITA though, your never going to run full parallel emulation, just ain't gonna happen. What can be done is have the primary thread do the HW synch and off lead individual chip emulation to their own threads. You end up with a ton of locks and pure lose efficiency but gain in scalability. The gains are different for each platform and might be more trouble then their worth.

No, BSNES is accurate, ZSNES isn't, never has been, and never will be. There's a long laundry list of games with significant issues, workarounds, and hardcoded hacks to get them to work. And some stuff never got fixed; major issues with SMRPG in some areas being one of the more famous ones, or the necessity to toggle between the different rendering engines to get certain games to work properly.

BSNES is much more accurate and doesn't need a single hack for any game to get it to work, but is much, much, much slower as a result. BSNES is as close to emulating the real life SNES HW as you can likely get.

 

[gamerk316]

AMD is providing also some MANTLE performance gains for Kaveri+290

BF4: 40.9% (1080p) and 40.1% (1600p) performance improvement under Ultra settings and 4xAA on the AMD A10-7700K with an AMD Radeon™ R9 290X.

StarSwarm: 319% (1080p) and 281% (1600p) performance improvement in the “RTS” test on Extreme settings with the AMD A10-7700K and an AMD Radeon™ R9 290X.

http://www.brightsideofnews.com/news/2014/1/29/amd-rele...

Don't forget the not CPU bottlenecked outcomes:

GPU-limited scenario: 2.7% (1080p) and 1.4% (1600p) performance improvement under Ultra settings and FXAA on the Core i7-4960X with an AMD Radeon™ R7 260X

GPU-limited scenario: 5.1% (1080p) and 16.7% (1600p) performance improvement in the “RTS” test on Extreme settings with the Core i7-4960X and an AMD Radeon™ R7 260X

Who was the one here who predicted all the gains would be seen on lower tier systems? Almost nothing on the high end, as expected.

Also worth noting: Who would buy a Kaveri, then pair it with a high-end dGPU? So I'd wager even the CPU bound numbers are inflated; use a PD CPU and a slightly lower GPU (7750?), and I'd expect maybe 10-15% gains, at best. Looks like AMD is testing in an artificially bad and unrealistic CPU bottlenecked situation.

MANTLE's aim is to eliminate the CPU bottleneck generated by the bloatware layer of DX. If you are testing a top eXtreme intel chip with a low end graphics card, you are in a GPU bottleneck situation and MANTLE provides only single digit gains.

However, if you test an eXtreme intel chip with a pair of 290X then you are again in a CPU bottleneck situation and MANTLE provides double digit gains: BF4 MANTLE is 58% faster for an i7-3970X Extreme with two R9 290X cards.

Your predictions about MANTLE are destroyed once again. Myself predicted MANTLE would bring 30--50% gains to BF4. Kaveri with an AMD Radeon™ R9 290X falls in the middle with 41%. The 58% for an i7-3970X Extreme with two R9 290X cards is ~8% on top of my maximum predicted value. The 58% destroys your "almost nothing on the high end" prediction.

Finally, I believe that you also missed this pre-build Kaveri plus a 290

http://www.tomshardware.com/news/cyberpowerpc-zeus-mini-kavari-haswell-pc-gaming,25890.html

Did you not learn your lesson with "best case" benchmarking yet?

The 3970x case is interesting, since its unlikely to be a pure CPU bottleneck even with two GPU's; I wonder if HTT is affecting the results, similar to how they affect Args1's benchmarks. I'd be interested to see HTT On versus HTT Off results, to see if the graphics driver is getting stuck occasionally on a HTT core, which could KILL performance...I'd like to think game developers are smarter then that though...

And IMO, Mantle becomes moot if MSFT tackles the driver overhead in DX12, like it is said they are doing. No dev is going to do a OpenGL EX/libgcm backend for the PS4, DirectX 11.1/Native backend for the XB1, DirectX+Mantle backend for the Windows, and OpenGL/Mantle backend for Linux; they aren't going to re-write their backend on four separate occasions.

 

[$hawn]

Power consumption and OC easiness

http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/9

Kaveri can be OC from 3.7GHz up to Richland base freq. without changing the voltage. The increase in power consumption at load is small (single digit watts). This data seems to agree with my original explanation that the reduction in CPU frequencies was mainly due to the reduction in TDP rating (100W-->95W) and not from SOI-->bulk. A hypothetical Kaveri APU rated at 100W, like Richland, could be clocked at 4.0--4.1 GHz. The difference would be on the OC capacity, with Richland OC better.

The above table also shows that you don't need increase voltages heavily to hit 4.4Ghz. Personally I would prefer stop at 4.3GHz with 1.275.

Except your ignoring ~50% of the silicon on the CPU. The GCN iGPU eats up more power then the previous VLIW4 design's. That translates into more TDP having to be used for the iGPU and so less is available for the CPU to clock up. When AMD design's these chips they have to take both components into account when designing specs. If you have a dGPU then you can safely ignore the iGPU's thermal usage, but then why would you be using an APU in the first place.

I think you mean ~50% of the silicon on the APU. But you are missing my point. I am comparing hypothetical 100W Kaveri (4.0--4.1GHz) against 95W Kaveri (3.7GHz), with the iGPU being the same in both.

I will explain it again: A hypothetical 100W Kaveri would have the CPU clocked at 4.0--4.1GHz using the same voltage (iGPU unchanged); the reduction in CPU clocks to 3.7GHz is not due to bulk but due to the reduction in TDP.

That last line, iit shows how little you know about CPUs and power consumption!!
400MHz extra for just 5watts? Are you f**kin kidding me !! This is one good chip that AMD has cherry picked and provided to the reviewer. Most retails chips won't be so good.

You think AMD would've bothered with an insignificant 5 watts for a Desktop APU, when it could have gotten ~10% higher clocks?!!

 

[gamerk316]

Kaveri Memory Scaling:

http://www.hardcoreware.net/kaveri-memory-speed-performance-scaling/

So as expected, diminishing returns after DDR3-1866. At that point, GPU is a larger bottleneck the RAM speeds.

 

[logainofhades]

Fortunately, you can often find 2133 gskill ares kits for the same price or less than many 1866 and some 1600 kits. I recently grabbed the 2133 cas 10 ares 8gb kit. It was one of the least expensive kits on PC partpicker.

 

[8350rocks]

Power consumption and OC easiness

http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/9

Kaveri can be OC from 3.7GHz up to Richland base freq. without changing the voltage. The increase in power consumption at load is small (single digit watts). This data seems to agree with my original explanation that the reduction in CPU frequencies was mainly due to the reduction in TDP rating (100W-->95W) and not from SOI-->bulk. A hypothetical Kaveri APU rated at 100W, like Richland, could be clocked at 4.0--4.1 GHz. The difference would be on the OC capacity, with Richland OC better.

The above table also shows that you don't need increase voltages heavily to hit 4.4Ghz. Personally I would prefer stop at 4.3GHz with 1.275.

Except your ignoring ~50% of the silicon on the CPU. The GCN iGPU eats up more power then the previous VLIW4 design's. That translates into more TDP having to be used for the iGPU and so less is available for the CPU to clock up. When AMD design's these chips they have to take both components into account when designing specs. If you have a dGPU then you can safely ignore the iGPU's thermal usage, but then why would you be using an APU in the first place.

I think you mean ~50% of the silicon on the APU. But you are missing my point. I am comparing hypothetical 100W Kaveri (4.0--4.1GHz) against 95W Kaveri (3.7GHz), with the iGPU being the same in both.

I will explain it again: A hypothetical 100W Kaveri would have the CPU clocked at 4.0--4.1GHz using the same voltage (iGPU unchanged); the reduction in CPU clocks to 3.7GHz is not due to bulk but due to the reduction in TDP.

That last line, iit shows how little you know about CPUs and power consumption!!
400MHz extra for just 5watts? Are you f**kin kidding me !! This is one good chip that AMD has cherry picked and provided to the reviewer. Most retails chips won't be so good.

You think AMD would've bothered with an insignificant 5 watts for a Desktop APU, when it could have gotten ~10% higher clocks?!!

Honestly, I am not sticking up for juan here at all, but that is typical. AMD advertises what they can guarantee, there will be fluctuation, but the stock configuration will be 100% achievable on any CPU, with any hardware compatible with the design. Most PD CPUs could hit 10% overclock without bumping voltage, my FX8350 is one of numerous such examples.

Really it shows how little you know about historical trends with a given architecture.

 

[Sigmanick]

As much as we have gotten off topic in the past, I started a separate thread for Star Swarm (free on steam) and Mantle (14.1 driver not out yet) results.

http://Toms/starswarmresults.Tada!

SO, shall this thread stay open till a time when software is benchmarked for the full capabilities of the chip with regards to HUMA and such?

 

[Reynod]

I think we would prefer you leave all of the AMD stuff in the one thread here which is a sticky for the present.

We have enough stickies to manage ... have at it over in drivers for a mantle thread but this one will stay up as a main sticky thanks.

$hawn ... abusing other users will get you sent on a holiday ... first and last warning.