Imagination Technologies Announces PowerVR Series6XT Successor to Apple's A7 Chip Graphics

Series 6 was a major IP shift. 6XT is comparatively much more modest. I'd compare it to SGX540, 545, 554 being added to the series 5 line after the 535 already existed. Their times to market were shorter.

554 announce to market for Apple's Ipad was 22 months. Though unlikely, I wouldn't totally rule out an Apple A8 with a 6XT graphics IP block in it.

It's important to remember that Apple is around a 10% owner in ImgTec. If anyone has early IP access, it's them.

edit: Most importantly, G6430 announce to A7 inclusion: 15 months.

50% increase from current chip huh ? well thats gonna help power the bigger display and new screen res for the iphone 6

Apple make sure you do a really good job with the next gen iPhone. My phone upgrades next December

DirectX 10. So, can it run Crysis?

PowerVR, sounds like a cheap 90s live action super hero series.

50% increase seems modest considering that Apple has always doubled GPU performance YOY. Especially with the A4 to A5 chip, the GPU was 7-9X faster (depending on the device; iPhone vs iPad).

Also, Imagination Technologies was boasting performance >1tflop with series 6 eventually, guess thats gonna take many years.

ghost187 said:

50% increase seems modest considering that Apple has always doubled GPU performance YOY. Especially with the A4 to A5 chip, the GPU was 7-9X faster (depending on the device; iPhone vs iPad).

Also, Imagination Technologies was boasting performance >1tflop with series 6 eventually, guess thats gonna take many years.

Click to expand...

They needed big performance jumps previously when going from non Retina to Retina iOS devices. Now, they're less necessary (but still nice!)

If they're genuinely planning a 12" iPad they'll need some extra GPU grunt.

Lots of speculation in the comments of the Anandtech article that Apple will be switching to their own custom GPUs (which has long been rumoured anyhow). Waiting until next year for just a 50% increase doesn't seem that appealing.

Tankmaze said:

50% increase from current chip huh ? well thats gonna help power the bigger display and new screen res for the iphone 6

Click to expand...

The iPhone 5s runs the same exact A7 as the iPad mini which is twice the size of the 5s.

Cloudsurfer said:

PowerVR, sounds like a cheap 90s live action super hero series.

Click to expand...

PowerVR has been around since the early 90's. Maybe 90's live action super hero series sound like a graphics chip.

This versus the new Nvidia K1. How do they compare?

I'd love to see one of those in the new Apple Soc(s). However Ryan from anandtech writes

Ryan Smith said:

Finally, while Imagination doesn’t provide a timeframe for consumer availability (since they only sell designs to chipmakers), based on the amount of time needed to integrate these designs into new products and then get those products in the hands of consumers, we should be looking at a timetable similar to the original Series6 designs. In which case Series6XT equipped SoCs would start appearing in 2015, likely in the latter half.

Click to expand...

jonAppleSeed said:

This versus the new Nvidia K1. How do they compare?

Click to expand...

Hard to say, at least if you mean the Tegra K1 introduced at CES this week. A lot of PR ignores practical TDP/battery life considerations. It will take a while before we see apples to apples, but both chips will be pretty awesome. Next year's ARM series will see a major node-1.5 node shrink which will be all the more awesome.

FWIW, current A7 is around a 4W TDP but generally runs closer to 1-2W most of the time especially on battery power. You don't notice a performance drop because the CPU is rarely the bottleneck for most operations. The value of faster and faster chips in the same power envelope is that they need less and less power to get the same performance. That means less heat, smaller battery, lighter machine, longer battery life.

MacLC said:

Hard to say, at least if you mean the Tegra K1 introduced at CES this week. A lot of PR ignores practical TDP/battery life considerations. It will take a while before we see apples to apples, but both chips will be pretty awesome. Next year's ARM series will see a major node-1.5 node shrink which will be all the more awesome.

FWIW, current A7 is around a 4W TDP but generally runs closer to 1-2W most of the time especially on battery power. You don't notice a performance drop because the CPU is rarely the bottleneck for most operations. The value of faster and faster chips in the same power envelope is that they need less and less power to get the same performance. That means less heat, smaller battery, lighter machine, longer battery life.

Click to expand...

Anand's back of the envelope numbers on K1 show it should be around 2W, which is in line with what's in the A7.

However, a K1 product will presumably ship H1 2014. We may not see 6XT until H2 2015.

An increase in power doesn't have much practical benefit right now. Battery life is the big issue.

Once we are ready to dock phones and use them as a primary computer - then this increased power will make sense again.

Now if only Apple included a decent amount of RAM then we could have a nice gaming handheld.

Jimmy James said:

An increase in power doesn't have much practical benefit right now. Battery life is the big issue.

Once we are ready to dock phones and use them as a primary computer - then this increased power will make sense again.

Click to expand...

Specifically with this architecture, it has more fine grain power gating (specifically clock gating in this instance from what I've read). That means it can shut off the parts of the GPU that aren't doing anything. Also, in general, fast is good in a mobile environment because it means you can turn off the processor quicker and return to a low power idle state. The important metrics in a mobile space are the idle power in addition to performance per milliwatt and performance per mm^2.

sexiewasd said:

Now if only Apple included a decent amount of RAM then we could have a nice gaming handheld.

Click to expand...

Their cadence is to double it ever other generation. That means iPhone 6 would be 2GB.

MacLC said:

Hard to say, at least if you mean the Tegra K1 introduced at CES this week. A lot of PR ignores practical TDP/battery life considerations. It will take a while before we see apples to apples, but both chips will be pretty awesome. Next year's ARM series will see a major node-1.5 node shrink which will be all the more awesome.

FWIW, current A7 is around a 4W TDP but generally runs closer to 1-2W most of the time especially on battery power. You don't notice a performance drop because the CPU is rarely the bottleneck for most operations. The value of faster and faster chips in the same power envelope is that they need less and less power to get the same performance. That means less heat, smaller battery, lighter machine, longer battery life.

Click to expand...

A7 can really suck down the power under certain conditions.



CPU power virus. No GPU.

A7 is more power hungry than A6.



(Same total power draw when the platform power for the ipad air is significantly lower).

A7 advantage is that for most tasks it is extremely efficient for user use senarios (a lot of this is due to apple optimization rather than SOC efficiency, easily seen on macbooks under OSX vs. Windows). But give it something like the Prime 95 equivalent and it will suck down power.

That said tegra 4 uses massively more power than the A7 and given nvidia's history of power hungry parts K1 is likely to continue this (though kepler was a major departure from fermi).

Jimmy James said:

An increase in power doesn't have much practical benefit right now. Battery life is the big issue.

Once we are ready to dock phones and use them as a primary computer - then this increased power will make sense again.

Click to expand...

The problem we have are, believe it or not, consumers who accept a 1 day battery life as the norm.
As people are buying many millions of phones with the daily recharge routine, there is no reason for them to give them better batteries.

If someone came to Apple and said, he guys we can give you double the lifespan in the current form factor, or due to our amazing new research we can give you the current lifespan but 2mm thinner.

Which one do you think Apple would do for?

Announce a double lifespan battery (yawn)
Announce the new iPhone NOW 2MM THINNER THAN BEFORE - Whoop Whoop yay yay rounds of applause

I think you know which one would be seen as the selling point.
So really buyers who accept this and continue to buy, really only have themselves to blame.

cirus said:

A7 can really suck down the power under certain conditions.

Image

CPU power virus. No GPU.

A7 is more power hungry than A6.

Image

(Same total power draw when the platform power for the ipad air is significantly lower).

A7 advantage is that for most tasks it is extremely efficient for user use senarios (a lot of this is due to apple optimization rather than SOC efficiency, easily seen on macbooks under OSX vs. Windows). But give it something like the Prime 95 equivalent and it will suck down power.

That said tegra 4 uses massively more power than the A7 and given nvidia's history of power hungry parts K1 is likely to continue this (though kepler was a major departure from fermi).

Click to expand...

Notice though on your power usage graph for Kraken the A7 gets the job done much quicker and powers back to idle. The A6 continues using a ton of power to finish the job. Hense the one of the reasons the A7 is much more of a power miser because it gets things done quickly and then powers back to idle.
This gets overlooked most of the time but it is a huge power saving part of the A7.

So are we going to see this in the iPhone 6S then and not the iPhone 6? If it's not available till mid 2015

Mechanic said:

Notice though on your power usage graph for Kraken the A7 gets the job done much quicker and powers back to idle. The A6 continues using a ton of power to finish the job. Hense the one of the reasons the A7 is much more of a power miser because it gets things done quickly and then powers back to idle.
This gets overlooked most of the time but it is a huge power saving part of the A7.

Click to expand...

Yep. But I was talking about TDP and max power usage. Which is up (also up for baytrail vs. clovertrail and S800 vs predecessors).

GPU Cores or Compute Clusters?

ImageTec is talking about compute clusters http://www.imgtec.com/powervr/sgx_series6.asp
Here is always mentioned the term GPU cores. I would assume that this is meaning the same?

I know that comparing GLFOPs may not compare the real performance, but:
NVIDIA Tegra K1: 384 GFLOPs
PowerVR G6400: 180+ GFLOPs

Does anybody have some information about the PowerVR Series6XT performance?

openglpabst said:

ImageTec is talking about compute clusters http://www.imgtec.com/powervr/sgx_series6.asp
Here is always mentioned the term GPU cores. I would assume that this is meaning the same?

I know that comparing GLFOPs may not compare the real performance, but:
NVIDIA Tegra K1: 384 GFLOPs
PowerVR G6400: 180+ GFLOPs

Does anybody have some information about the PowerVR Series6XT performance?

Click to expand...

Saying cores is an inaccurate term when referring to ImgTec's series 6 IP. However, in this case, cluster and core are interchangeable.

Series6XT is supposed to be a 50% better performer, but that has no context such as memory bandwidth or frequency comparison.

Similarly, you can't compare the K1 and G6400 on GF alone because things like memory bandwidth play a huge part. Apple is usually on the forefront of memory bandwidth to their SoCs compared to their competitors. Tile-based deferred rendering (TBDR) used in ImgTec's parts is an extremely efficient method for graphics computation, but not so well suited for higher performance desktop environments. The K1 has a desktop derivative part (Kepler), and I don't know how they compare in terms of efficiency/utilization. That would tell you how realistic that peak GF number is. ImgTec has always prided themselves on not inflating their performance numbers when quoting them, but we'll have to see if that continues. If the large 4MB SRAM cache on the A7 is used by the GPU, it will have an advantage over its competitors when comparing raw SDRAM bandwidth as well. This should improve with a 6XT implementation because it has even more compression utilization than the 6 series implementation found in the A7.

The K1 incarnation coming to market also has a dated CPU comparatively. Apple is already use a wide ARMv8 design while Samsung and Qualcomm will be doing so this year too. We'll have to wait until the second half of next year to see Nvidia's custom ARMv8 implementation. By then, Apple and others will be on their second ARMv8 iteration and will have also updated their graphics capabilities too.

From what I've seen, openGL ES 3.0 use in android is very limited (support is relatively new) so the question always becomes whether the faster GPU will even manifest in games with better graphics on the platform anyway. Until GPU compute becomes a big deal in mobile, it may be a spec sheet bragging rights type of win.