"Mysterious" AMD launch (Mobile Kaveri APUs under FX)

[formulav8]

Meh, its probably something somewhat blah that AMD is promoting. Edge of space? Maybe Nasa has been using their chips or something. But I haven't kept up on the threads here or anything so I'm probably way off here of course

 

[ZGR]

If AMD can price the FX-7600P right, it could be a pretty good chip for it's TDP. Battery life is the biggest concern...

Anandtech' recent post of the 7600P show a pretty good improvement over Richland; plus it supports much faster RAM. The single-threaded performance is not competitive to Intel, but it's graphics performance is quite impressive given that Iris Pro is so expensive.

 

[Shivansps]

I still i dont get it why compare it to a I7 ULV that performs likely worse than a I3-4110M...

Iris Pro is better, but also non-existant on mobile.

 

[jj109]

I still i dont get it why compare it to a I7 ULV that performs likely worse than a I3-4110M...

Iris Pro is better, but also non-existant on mobile.

I'd guess that laptop manufacturers aren't as willing to send out 35W+ laptop samples because 15W mobile laptops are all the rage right now.

Either way, AMD better get that ULV part into a laptop before Broadwell hits or Kaveri will be sitting in the bargain bin like the rest of the mobile BD/PD parts.

 

[frozentundra123456]

The problem is that it is still very marginal for gaming, and loses badly in cpu performance, trading blows with an intel cpu at less than half the TDP. So again, as with the desktop apus, I can see a niche for light gaming, but overall it will have a hard time competing in either the low end light usage scenario or for "real" gaming where you still need a discrete card for 1080p demanding games. It also does not make sense to me that AMD brought out desktop Kaveri first. If they had brought out mobile Kaveri first, they would have had a longer run before having to compete against mobile Broadwell.

 

[Bubbleawsome]

13 hours now.

 

[USER8000]

The problem is that it is still very marginal for gaming, and loses badly in cpu performance, trading blows with an intel cpu at less than half the TDP. So again, as with the desktop apus, I can see a niche for light gaming, but overall it will have a hard time competing in either the low end light usage scenario or for "real" gaming where you still need a discrete card for 1080p demanding games. It also does not make sense to me that AMD brought out desktop Kaveri first. If they had brought out mobile Kaveri first, they would have had a longer run before having to compete against mobile Broadwell.

My mate got a Lenovo A10 5750M based laptop a while back and for the price at the time(it was £400 with an SSHD too),nothing could quite compare for a light gaming machine.

In fact in the UK many of the A8 and A10 based machines,used to be the cheapest if you wanted to have some sort of gaming ability in a laptop(to some degree it is still true now,although it appears stock is now being rundown in anticipation of Kaveri based ones I suspect),especially in the £300 to £400 range.

 

[Sweepr]

The problem is that it is still very marginal for gaming, and loses badly in cpu performance, trading blows with an intel cpu at less than half the TDP. ... If they had brought out mobile Kaveri first, they would have had a longer run before having to compete against mobile Broadwell.

My thoughs as well. My 2-year old 17W Ivy Bridge ULV scores 1.24 / 2.74 pts @ CB 11.5 (ST / MT x64) while AMD's fastest 2014 35W Kaveri scores 0.87 / 2.5 pts. They are still lagging badly in CPU performance, and then again this is the fastest SKU, their 17-19W models are either dual-cores or lower clocked quad-cores. iGPU performance is good, what you'd expect from the GCN transition (either similar or ~50% faster than Richland parts depending on game tested) but I wouldn't be surprised if Broadwell-Y/U somewhat closed the gap on <20W TDP models later this year. I bet Intel/AMD mobile iGPUs will have a hard time running the new X1/PS4 ports anyway, dGPU gaming machines based on Maxwell and future 20nm parts should be a lot faster. It would have been more impressive as an early 2014 product.

 

[sm625]

The giant unanswered question remains: what is power cosumption. How does it compare with intel 15W? What really gets me is that we still do not have good, solid numbers on richland mobile, even after a year. The Anandtech A10-5750M Review from last June also had nothing on power consumption, and they never updated it. This is extremely frustrating to say the least.

It's not just OEMs that are out to kill AMD. Without this critical information, we cant know how these chips actually compete, unless you are willing to place your trust in completely arbitrary TDP numbers. So we cant properly assess the value of these chips at all. You cant blame that on OEMs and their corner cutting.

Edit: I was able to find one useful piece of data on richland:

Considering how far ahead AMD's own discrete solution is vs Richland, it is safe to say kaveri wont close that gap. source The verdict isnt in just yet, but if it turns out that even in 2014, AMD still cannot produce an APU that beats a CPU+dGPU in FPS/watt at around the very popular 35W level, then they may as well just declare bankruptcy.

 

[PPB]

My thoughs as well. My 2-year old 17W Ivy Bridge ULV scores 1.24 / 2.74 pts @ CB 11.5 (ST / MT x64) while AMD's fastest 2014 35W Kaveri scores 0.87 / 2.5 pts. They are still lagging badly in CPU performance, and then again this is the fastest SKU, their 17-19W models are either dual-cores or lower clocked quad-cores. iGPU performance is good, what you'd expect from the GCN transition (either similar or ~50% faster than Richland parts depending on game tested) but I wouldn't be surprised if Broadwell-Y/U somewhat closed the gap on <20W TDP models later this year. I bet Intel/AMD mobile iGPUs will have a hard time running the new X1/PS4 ports anyway, dGPU gaming machines based on Maxwell and future 20nm parts should be a lot faster. It would have been more impressive as an early 2014 product.

So you use your PC to run CB ? Because the only thing CB is representative of is itself. CB isnt even representative of the C4D in real world situations, and not even talking the rest of the renderers out there.

It is laughable people still use CB as an standard for CPU performance, but hey, if it floats your boat.... :biggrin:

 

[Sweepr]

It is laughable people still use CB as an standard for CPU performance, but hey, if it floats your boat.... :biggrin:

Well I'm sure you will find plenty of benchmarks showing 35W Kaveri models slower than some Haswell ULT 15W chips in CPU tasks very soon (not to mention 17-19W stuff), but right now I'm fine with CB. :biggrin:

 

[Arachnotronic]

The giant unanswered question remains: what is power cosumption. How does it compare with intel 15W? What really gets me is that we still do not have good, solid numbers on richland mobile, even after a year. The Anandtech A10-5750M Review from last June also had nothing on power consumption, and they never updated it. This is extremely frustrating to say the least.

It's not just OEMs that are out to kill AMD. Without this critical information, we cant know how these chips actually compete, unless you are willing to place your trust in completely arbitrary TDP numbers. So we cant properly assess the value of these chips at all. You cant blame that on OEMs and their corner cutting.

Edit: I was able to find one useful piece of data on richland:

Considering how far ahead AMD's own discrete solution is vs Richland, it is safe to say kaveri wont close that gap. source The verdict isnt in just yet, but if it turns out that even in 2014, AMD still cannot produce an APU that beats a CPU+dGPU in FPS/watt at around the very popular 35W level, then they may as well just declare bankruptcy.

Bingo...

 

[AtenRa]

The giant unanswered question remains: what is power cosumption. How does it compare with intel 15W? What really gets me is that we still do not have good, solid numbers on richland mobile, even after a year. The Anandtech A10-5750M Review from last June also had nothing on power consumption, and they never updated it. This is extremely frustrating to say the least.

It's not just OEMs that are out to kill AMD. Without this critical information, we cant know how these chips actually compete, unless you are willing to place your trust in completely arbitrary TDP numbers. So we cant properly assess the value of these chips at all. You cant blame that on OEMs and their corner cutting.

Edit: I was able to find one useful piece of data on richland:

Considering how far ahead AMD's own discrete solution is vs Richland, it is safe to say kaveri wont close that gap. source The verdict isnt in just yet, but if it turns out that even in 2014, AMD still cannot produce an APU that beats a CPU+dGPU in FPS/watt at around the very popular 35W level, then they may as well just declare bankruptcy.

Im sure at the same fps, Richland has higher fps/watt than HD4000.
Getting higher performance always decrease the performance/watt, that is why the lower-end GPUs and CPUs always have higher performance/watt although they are of the same architecture/design as their higher-End SKUs.

So, performance/watt is not what you want to have as a metric for power consumption unless you are looking at the same performance.

 

[Enigmoid]

Im sure at the same fps, Richland has higher fps/watt than HD4000.
Getting higher performance always decrease the performance/watt, that is why the lower-end GPUs and CPUs always have higher performance/watt although they are of the same architecture/design as their higher-End SKUs.

So, performance/watt is not what you want to have as a metric for power consumption unless you are looking at the same performance.

Sure and so does haswel

 

[AtenRa]

Sure and so does haswel

What does Haswell ??

 

[Phynaz]

Im sure at the same fps, Richland has higher fps/watt than HD4000.
Getting higher performance always decrease the performance/watt, that is why the lower-end GPUs and CPUs always have higher performance/watt although they are of the same architecture/design as their higher-End SKUs.

So, performance/watt is not what you want to have as a metric for power consumption unless you are looking at the same performance.

This makes no sense at all.

 

[Enigmoid]

What does Haswell ??

At the same performance as hd4000, hd4400 in HW ULV has significantly better efficiency. Both AMD and intel's newer products are more efficient. It's a fair comparison using older products.

 

[TrulyUncouth]

This makes no sense at all.

I think the point he is making is similar to how Intel will say "This processor can be up to twice as fast, or use 30% less power for the same performance."

He is saying that if we framerate locked all of those solutions and saw how all of them did at the same settings and framerate we would get better perf/W for the AMD solution. I am not going to say I agree, but he does make a valid point.

My understanding is that if you take a 3ghz processor and underclock/undervolt it to 2ghz it will get much better perf/W but you would lose 30% of your performance. He is making that point.

 

[tential]

The giant unanswered question remains: what is power cosumption. How does it compare with intel 15W? What really gets me is that we still do not have good, solid numbers on richland mobile, even after a year. The Anandtech A10-5750M Review from last June also had nothing on power consumption, and they never updated it. This is extremely frustrating to say the least.

It's not just OEMs that are out to kill AMD. Without this critical information, we cant know how these chips actually compete, unless you are willing to place your trust in completely arbitrary TDP numbers. So we cant properly assess the value of these chips at all. You cant blame that on OEMs and their corner cutting.

Edit: I was able to find one useful piece of data on richland:

Considering how far ahead AMD's own discrete solution is vs Richland, it is safe to say kaveri wont close that gap. source The verdict isnt in just yet, but if it turns out that even in 2014, AMD still cannot produce an APU that beats a CPU+dGPU in FPS/watt at around the very popular 35W level, then they may as well just declare bankruptcy.

I didn't even read the review. Second you have 0 batterylife slides why should I care? What's the point of a mobile CPU review without batterylife? Imagine if the DC review came out tomorrow, and the only benchmarks were IGP benchmarks. We'd be enraged.

Performance only mobile benchmarks are pretty much pointless. I couldn't care less how fast Kaveri is without batterylife there. Not trying to be rude, but it really is a waste of time for a review to have no slides on power consumption/batterylife when it's a mobile chip.

 

[AtenRa]

This makes no sense at all.

Have a look at HD7850 performance per Watt against HD7950.
HD7850 has higher performance per Watt although they are the same mArchitecture and manufactured at the same process at the same fab.

The higher your performance the lower your performance per watt. It is simple to understand, you spend way more power to get another 10% of performance. And another 10% of performance will require more than 10% of power usage increase, so your Performance Per Watt falls down again. And will continue to fall the higher you raise your performance because the power you need to use to have 10% of performance is disproportionate to the performance you are getting.

So the same GPU can have different performance per watt characteristics depending of the performance you are willing to have. So actually HD7950 can have the same performance per watt as HD7850 but it will decrease its performance. On the other hand HD7850 can increase its performance but it will decrease its Performance per Watt.

In order to have a nice comparison so that you will understand how efficient each architecture or product is you must compare them at the same performance or at the same power. Like Intel did to demonstrate its 14nm Broadwell against 22nm Haswell, they measured the power consumption at the same performance between the two.

 

[AtenRa]

At the same performance as hd4000, hd4400 in HW ULV has significantly better efficiency. Both AMD and intel's newer products are more efficient. It's a fair comparison using older products.

Ahh yes i agree, but comparing performance per watt alone doesnt tell the whole story.

 

[TreVader]

The 7600P is performing at about the level of the Iris Pro part in my rMBP (HD5200), maybe even a little better. The 750M tends to double the framerate on the the Iris Pro (usually anything that gets 15fps on iris pro gets 30 on 750M).



Very impressive that CPU is still on the level of an i5 - U or i3 - U (Haswell).



The real question is battery life.

 

[Rickyyy369]

Page has been updated

 

[Enigmoid]

The 7600P is performing at about the level of the Iris Pro part in my rMBP (HD5200), maybe even a little better. The 750M tends to double the framerate on the the Iris Pro (usually anything that gets 15fps on iris pro gets 30 on 750M).

Very impressive that CPU is still on the level of an i5 - U or i3 - U (Haswell).
The real question is battery life.

Its not quite there. Its quite significantly slower than the 750m and and that chip is using DDR3-1800 mhz. The iris pro rmbp is slower than the 750m ddr5 on the retina (I'm assuming that you mean in your macbook). The 750m in that charts is quite BW bound. Iris also performs a lot better at lower resolutions and settings.

http://www.anandtech.com/show/7238/acer-v7482pg9884-review-everything-you-need/4

Comparison to 15W intel chips is problematic because they are so thermally bound. Even with mods (4k textures and such) the HD 4000 in my laptop gets around 30-32 fps in skyrim (riverwood). (That's the only game on my notebook at the moment).

The other thing to be extremely aware of is that the AMD kaveri reference design is using much faster RAM (at least 1866 mhz and probably 2133 mhz). That is responsible for a good chunk of the performance increase due to RAM speed that OEM notebooks easily may not use.

However, if used with high speed RAM this will easily eliminate a good chunk of the low end igps. Strong performance.

<snip>

In order to have a nice comparison so that you will understand how efficient each architecture or product is you must compare them at the same performance or at the same power. Like Intel did to demonstrate its 14nm Broadwell against 22nm Haswell, they measured the power consumption at the same performance between the two.

I see your point but Tahiti is incredibly inefficient compared to Pitcairn. Pitcairn vs. Cape Verde is different.

 

[AtenRa]

https://www.youtube.com/watch?v=umR_iyQ0RSE

all that time for that ???