The Official Kaveri Review Thread (A10-7850K, etc)

[Shivansps]

Because performance and power is way too close to the 95W part, i smell something like Intel "2nd TDP" on it, but im not sure.

 

[ElFenix]

Because performance and power is way too close to the 95W part, i smell something like Intel "2nd TDP" on it, but im not sure.

in some of the tests it looks more like the 95W part isn't going all the way up to 95W. you also have to keep in mind that pretty much every measurement is going to be at the wall. in the case of anandtech's, power draw is something like 10% of the power supply's capable output where the supply itself is wasting 20-25W of its draw. (why is anything but top of the line SLI done with a 1250 watt supply?)

http://www.techpowerup.com/reviews/OCZ/ZX_1250W/5.html

 

[coercitiv]

Because performance and power is way too close to the 95W part, i smell something like Intel "2nd TDP" on it, but im not sure.

Let's consider measurements taken by Hardware Canucks and Tech Report. Peak power consumption for a system equipped with the 45W part and no dGPU was around 75W. We know peak power consumption is probably misleading (not average), but let's use it anyway.

Both sites used 80 Plus Gold rated power supplies, so we can expect above 80% efficiency even at low loads of 60-80W. Let's say 85% efficiency.We also know idle power consumption to be around 21W. So,if we factor in PSU efficiency, peak system power consumpton is around 75*0.85W = 64W. From this we subtract a 45W power budget for the CPU, which leaves us with 19W.

Now, if the completely idle system uses around 21W, would it be plausible for all the components except CPU (mobo, ram, ssd, fan) to use 19W while system is under load?

 

[ElFenix]

Let's consider measurements taken by Hardware Canucks and Tech Report. Peak power consumption for a system equipped with the 45W part and no dGPU was around 75W. We know peak power consumption is probably misleading (not average), but let's use it anyway.

Both sites used 80 Plus Gold rated power supplies, so we can expect above 80% efficiency even at low loads of 60-80W. Let's say 85% efficiency.We also know idle power consumption to be around 21W. So,if we factor in PSU efficiency, peak system power consumpton is around 75*0.85W = 64W. From this we subtract a 45W power budget for the CPU, which leaves us with 19W.

Now, if the completely idle system uses around 21W, would it be plausible for all the components except CPU (mobo, ram, ssd, fan) to use 19W while system is under load?

techpowerup's power supply is an antec HCP-750, which is less than 75% efficient when drawing 75 watts from the wall.

http://www.techpowerup.com/reviews/Antec/HCP_750W/5.html

 

[coercitiv]

techpowerup's power supply is an antec HCP-750, which is less than 75% efficient when drawing 75 watts from the wall.

http://www.techpowerup.com/reviews/Antec/HCP_750W/5.html

Well, tech report got 78W peak power draw, hardware canucks got 73W, and I took an average of 75 Watts, so that compensates a bit.

[Later Edit] although judging by the idle power draw in the tech report test, they probably had a dGPU installed.

 

[naukkis]

Both sites used 80 Plus Gold rated power supplies, so we can expect above 80% efficiency even at low loads of 60-80W. Let's say 85% efficiency.We also know idle power consumption to be around 21W. So,if we factor in PSU efficiency, peak system power consumpton is around 75*0.85W = 64W. From this we subtract a 45W power budget for the CPU, which leaves us with 19W.

There's also cpu-power regulation from 12V to cpu vcore which efficiency might be just 80% or even lower.

 

[coercitiv]

There's also cpu-power regulation from 12V to cpu vcore which efficiency might be just 80% or even lower.

I know, the point I'm trying to make is even if we consider the powering circuitry to have higher than normal efficiency, the power consumption figures still show the A8 7600 to be a 45W capable chip.

 

[AtenRa]

Clock To Clock at 3GHz against Richland from Techspot review

http://www.techspot.com/review/770-amd-a8-7600-kaveri/

 

[Dresdenboy]

I do not belive the A8-7600 to be working in a 45W TDP window.

I'm atheist. But you could measure it, or look at reviews. Maybe first you should check the definition of TDP used.

 

[jpiniero]

Seems like AMD needs to work on the chipset power consumption.

 

[littlebluesmurf]

Clock To Clock at 3GHz against Richland from Techspot review

http://www.techspot.com/review/770-amd-a8-7600-kaveri/

That looks rather promising, will be interesting to see what sort of clocks mobile Kaveri will have and how competitive it will be.

 

[CHADBOGA]

6% from BD to PD and 11% from PD to SR.

I understood the point you were making.

Your English is pretty good.

 

[Homeles]

Seems like AMD needs to work on the chipset power consumption.

That's basically tied to the process node. They may be using an older one for their chipsets as well.

 

[Fox5]

Clock To Clock at 3GHz against Richland from Techspot review

http://www.techspot.com/review/770-amd-a8-7600-kaveri/

The most impressive thing to me is Intel's cache bandwidth. Wowzers, if I could write software that ran directly from cache, it seems like you could get crazy performance.

More important question: When is Kaveri hitting laptops? That seems to be where its performance/watt will really make a difference. 45w and 65w seem like they'd fit fine in larger laptops.

Disappointing to see that even clock for clock, Kaveri struggles to match Richland.

 

[AtenRa]

Disappointing to see that even clock for clock, Kaveri struggles to match Richland.

Struggles ?? Kaveri is 10-15% faster on average than Richland clock to clock with 15-20% lower power consumption. A 35W TDP Laptop Kaveri could be up to 50% faster than Trinity. Its not far stretched to expect at least a 30% gain over Richland at the same TDP. I expect even more in iGPU performance.

 

[Enigmoid]

Struggles ?? Kaveri is 10-15% faster on average than Richland clock to clock with 15-20% lower power consumption. A 35W TDP Laptop Kaveri could be up to 50% faster than Trinity. Its not far stretched to expect at least a 30% gain over Richland at the same TDP. I expect even more in iGPU performance.

Your previous link is highly optimistic with 10-15% increases at the same clock. At lot of time Kaveri was slower than the underclocked a10-6800k. It appears that singlethread performance is no better clock for clock but the multithread performance has improved substantially.

The other thing is if they undervolted richland because that is tremendous frequency drop to be running at the same volts.

Mobile kaveri will be nowhere close to 50% faster than mobile trinity. 50% faster is in a10-6800k territory.

 

[Ajay]

Your previous link is highly optimistic with 10-15% increases at the same clock. At lot of time Kaveri was slower than the underclocked a10-6800k. It appears that singlethread performance is no better clock for clock but the multithread performance has improved substantially.

That makes allot of sense given the architectural changes AMD has admitted to making. I guess we'll be waiting for Excavator to see significant SP IPC improvements. The only problem is that I wouldn't expect to see clock speeds increasing with the current top TDP of 65 W on Carrizo (if @ 20nm). It's really too bad that AMD can't afford to subsidize GF R&D to get a real top of the line HP node.

 

[el etro]

I understood the point you were making.

Your English is pretty good.

Thanks!

 

[Shivansps]

I'm atheist. But you could measure it, or look at reviews. Maybe first you should check the definition of TDP used.

Look TDP and Turbos where already hard to understand by everyone before AMD started to use it, i actually had to do my own tests to verify.

On Intel, you have 2 TDP settings on bios, 1 set by the CPU used and could also be changed by manufacturer, and 2nd TDP setting that it may or may not be used.

So, on Intel you will find that, for example a 35W mobile cpu using 35W (or less) as 1st TDP, and 45W as 2nd TDP (some OEMs to reduce cooling use like 25W 1st/35W 2nd).
The 2nd TDP works as long the cooling allows it, so it allows the CPU to turbo boost to it, until a pre-defined temperature is reached.

The 1st TDP setting is the one that the cooling must support in order to keep temperature below critical, here temperature is secondary as turbos work on a calculared power, Open Hardware Monitor reads this value as "CPU Package Power", and turbos ajust to the maximum possible number, higher temperature also means you can reach that number faster, but it does not really matter until temperature hits critical (85°C).

So to make it short, on Intel, 2nd TDP allows to turbo boost well over the CPU max TDP as long the cooling allows it, 1st TDP ties up turbos to it and temperature is secondary.

I smelling something like that on the A8-7600, maybe a 1st TDP of 45W, but a hidden 2nd tdp of 65W, them for 65W as 1st, a 2nd tdp of 85W, as the cooler allows it and no bios info on it, you will have no way to know, and that will explain a lot of what im seeing on A8-7600 numbers in general.

As its a Desktop chip, it does not really matter, but on mobile it will.

 

[AtenRa]

Your previous link is highly optimistic with 10-15% increases at the same clock. At lot of time Kaveri was slower than the underclocked a10-6800k. It appears that singlethread performance is no better clock for clock but the multithread performance has improved substantially.

I wasn't talking about single thread performance or IPC but performance in general. Kaveri at lower TDPs have a lot of frequency headroom, A8-7600 has base frequency of 3.1GHz vs Richland A8-6500T with base 2.1GHz and A10-6700T with base frequency of 2.5GHz.

Mobile kaveri will be nowhere close to 50% faster than mobile trinity. 50% faster is in a10-6800k territory.

I said up to 50% faster than Trinity and 30% on average against Richland.

Just compare 45W Richland A8-6500T against Kaveri 45W A8-7600.

According to AT review, CPU only benchmarks

Handbrake film = 40% faster than 6500T
ProvRay = 53% faster than 6500T
Agisoft PS v1.0 = 18% faster than 6500T
Faststone Image viewer = 22% faster than 6500T
7zip = 24% faster than 6500T

~31% faster than 6500T on average.

Gaming

Bioshock = 41% faster than 6500T
Tomb Raider = 49% faster than 6500T
F1 2012 = 39% faster than 6500T
Company Of Heroes = ~50% faster than 6500T
Sleeping Dogs = 40.5% faster than 6500T

~44% faster than 6500T on average

And Compute

Luxmark = 174% faster than 6500T
CL Benchmark = 164% faster than 6500T

No need to get the average here :whiste:

Now, imagine a 35W Mobile Kaveri against 35W Mobile Richland. And that's only with cheap 28nm bulk.

 

[Enigmoid]

I wasn't talking about single thread performance or IPC but performance in general. Kaveri at lower TDPs have a lot of frequency headroom, A8-7600 has base frequency of 3.1GHz vs Richland A8-6500T with base 2.1GHz and A10-6700T with base frequency of 2.5GHz.

I said up to 50% faster than Trinity and 30% on average against Richland.

Just compare 45W Richland A8-6500T against Kaveri 45W A8-7600.

According to AT review, CPU only benchmarks

Handbrake film = 40% faster than 6500T
ProvRay = 53% faster than 6500T
Agisoft PS v1.0 = 18% faster than 6500T
Faststone Image viewer = 22% faster than 6500T
7zip = 24% faster than 6500T

~31% faster than 6500T on average.

Gaming

Bioshock = 41% faster than 6500T
Tomb Raider = 49% faster than 6500T
F1 2012 = 39% faster than 6500T
Company Of Heroes = ~50% faster than 6500T
Sleeping Dogs = 40.5% faster than 6500T

~44% faster than 6500T on average

And Compute

Luxmark = 174% faster than 6500T
CL Benchmark = 164% faster than 6500T

No need to get the average here :whiste:

Now, imagine a 35W Mobile Kaveri against 35W Mobile Richland. And that's only with cheap 28nm bulk.

You are competing with mobile richland.

A10-5750m 2.5 ghz base. 3.5 ghz turbo. In most reviews its running at 3.2 ghz on lightly threaded workload and 2.7 ish under four core heavy (prime) loads.

A8-6500T is a good bit weaker than mobile richland. less shader cores and lower frequency on the CPU.

And don't compare it to the a8-6500T. That CPU has basically 0 availability.

Toms and techreport showed less efficiency too.

 

[AtenRa]

You are competing with mobile richland.

A10-5750m 2.5 ghz base. 3.5 ghz turbo. In most reviews its running at 3.2 ghz on lightly threaded workload and 2.7 ish under four core heavy (prime) loads.

5750M is an A10, not A8 like 7600. You have to compare apples to apples. You dont compare Core i3 to Core i5 dont you ??

A8-6500T is a good bit weaker than mobile richland. less shader cores and lower frequency on the CPU.

A8-6500T is exactly the same as A8-5557M only with higher iGPU clocks. A 35W A8 Kaveri will have almost the same performance characteristics the desktop A8-7600 exhibits against A8-6500T.

And don't compare it to the a8-6500T. That CPU has basically 0 availability.

The comparison was to show the performance advantage Kaveri has over Richland at the same TDP. It is irelevant if the 6500T has zero availability in retail or not.

 

[Enigmoid]

5750M is an A10, not A8 like 7600. You have to compare apples to apples. You dont compare Core i3 to Core i5 dont you ??

A8-6500T is exactly the same as A8-5557M only with higher iGPU clocks. A 35W A8 Kaveri will have almost the same performance characteristics the desktop A8-7600 exhibits against A8-6500T.

The comparison was to show the performance advantage Kaveri has over Richland at the same TDP. It is irelevant if the 6500T has zero availability in retail or not.

A10 or A8 is irrelevant. All that matters is the clocks. Its very likely that what passes as an A8 on the desktop is going to pass as A10 on mobile, similar to how i5m ~ <= i3 desktop.

At the same tdp is not equal to the same power consumption.

I expect to see reasonable gains on mobile but 50%? Nope not a chance.

 

[AtenRa]

A10 or A8 is irrelevant. All that matters is the clocks.

No it is not irrelevant. A10 is a full Chip while A8 is castrated.

A8-6500T only has 256 Radeon Cores, A10-6700T has 384. Both are 45W TDP SKUs

Mobile A8-5557M only has 256 Radeon Cores, A10-5750M has 384. Both are 35W TDP SKUs

Kaveri A8-7600 has 384 Radeon Cores, A10-7850K has 512.

We have to assume that AMD will also release a mobile A10 with 512 Radeon Cores.

 

[Enigmoid]

No it is not irrelevant. A10 is a full Chip while A8 is castrated.

A8-6500T only has 256 Radeon Cores, A10-6700T has 384. Both are 45W TDP SKUs

Mobile A8-5557M only has 256 Radeon Cores, A10-5750M has 384. Both are 35W TDP SKUs

Kaveri A8-7600 has 384 Radeon Cores, A10-7850K has 512.

We have to assume that AMD will also release a mobile A10 with 512 Radeon Cores.

I was looking at CPU performance.

And I can't see any appreciable difference between A8-7600 and A10-7850k in GPU performance as an effect of shader number.