Qualcomm shows working MSM8x60 at CES.

[Wishmaster]

Qualcomm has indicated Adreno 3xx would be using multiple cores in the high-end, and I'd tend to believe that's not just marketing.

New architecture or modified(to support multiple cores) adreno 2xx? OpenGL ES Halti support should indicate new architecture but everything is possible...

 

[Deleted member 13524]

My recollection of Adreno cores is a bit rusty, but if memory serves well the 220 is a dual core GPU each core with a single TMU. Xenos has 16 TMUs amongst other things let alone that it clocks at 500MHz. Just as much as the NGP can match the PS3 I guess.

I wasn't talking about the Adreno 220, I was talking about the Adreno 320 in Krait, which should be 3-4x faster than 220 (assuming A220 is 4-5x A200, though I think, given some games I've seen for WP7, the A200's driver in Android is really under-developed).

BTW, the Adreno 320 is a quad-core GPU, as stated by Qualcomm themselves.
It wouldn't be surprising if it turns out a 4*Adreno 220 (and Adreno 225 being a 2*Adreno 220).


Nonetheless, as the Adreno 220 seems to surpass the higher-clocked SGX540, the Adreno 225 should compete with SGX543MP2 and the Adreno 320 would point its guns at SGX543MP4.



Given the supposed architectural similarity between Xenos and Adreno, I guess Microsoft could be trying to stick to Qualcomm CPUs as a requirement for future WP devices and\or a handheld console. Promoting "easy" ports from X360 to handhelds would surely be a good way to atract AAA developers for their platform.

 

[Ailuros]

I wasn't talking about the Adreno 220, I was talking about the Adreno 320 in Krait, which should be 3-4x faster than 220 (assuming A220 is 4-5x A200, though I think, given some games I've seen for WP7, the A200's driver in Android is really under-developed).

BTW, the Adreno 320 is a quad-core GPU, as stated by Qualcomm themselves.
It wouldn't be surprising if it turns out a 4*Adreno 220 (and Adreno 225 being a 2*Adreno 220).

I still don't see that one coming close to Xenos.

Adreno 320 delivers similar graphics performance to today’s latest game consoles, but for mobile devices.

 

[Deleted member 13524]

I still don't see that one coming close to Xenos.

Ok, let's not say "coming close" to Xenos but rather "coming to the same order of magnitude in performance".
Just as 4*A9+543MP4 approaches the PS3. It's not "more than 10x faster" anymore.

 

[Ailuros]

Ok, let's not say "coming close" to Xenos but rather "coming to the same order of magnitude in performance".
Just as 4*A9+543MP4 approaches the PS3. It's not "more than 10x faster" anymore.

It's definitely not an elephant vs. mouse difference; but a 10 pound rat is still not an elephant

 

[ltcommander.data]

Considering that those probably aren't final drivers I would say it compares quite well with other SoC's(especially with SGX540 on OMAP4). Impressive how much higher performance it achieves when Vsync is turned off. It should stop people from bashing snapdragon platform.

So it would seem that their PR talk of 2x the performance was accurate.

There's still the question of how much of this is realizable in the real world. Anand said these results are based on a development platform running at maximum clock speeds with no power management and run powered by a USB port. This is compared to shipping devices, optimized for battery life, presumably running on battery power with power management attempting to down clock things at every opportunity.

 

[Ailuros]

Development platforms are usually a lot closer to tablet than smart-phone designs. The frequency comment above is right on spot.

 

[Deleted member 13524]

Development platforms are usually a lot closer to tablet than smart-phone designs. The frequency comment above is right on spot.

On the other hand, driver performance may still improve, so the off-the-shelf performance could be both higher and lower.

 

[Ailuros]

On the other hand, driver performance may still improve, so the off-the-shelf performance could be both higher and lower.

I don't expect to see the dual CPUs to be clocked at 1.5GHz in a smart-phone either. Smart-phones (which would bank with the 800*480 resolution of most other devices tested from Anand) with the MSM8x60 will have the CPUs most likely clocked at 1.2GHz.

Devices with the MSM8x60 aren't that far off - we're looking forward to getting our hands on the HTC Pyramid and EVO 3D which both will likely have 1.2 GHz MSM8x60 inside.

http://www.anandtech.com/show/4243/...mance-1-5-ghz-msm8660-adreno-220-benchmarks/3

However I've skimmed through the articles a couple of times and couldn't find anything that mentions what resolution the MSM8x60 dev platform Anand received was running at. Either there's nothing or I'm too tired to find it

 

[Deleted member 13524]

However I've skimmed through the articles a couple of times and couldn't find anything that mentions what resolution the MSM8x60 dev platform Anand received was running at. Either there's nothing or I'm too tired to find it

The first page mentions it a couple of times.

It's WVGA, 800*480, like all the others (except Optimus One, which has a 480*320 display and a MSM7227 (ARM11+Adreno200))

 

[metafor]

I wasn't talking about the Adreno 220, I was talking about the Adreno 320 in Krait, which should be 3-4x faster than 220 (assuming A220 is 4-5x A200, though I think, given some games I've seen for WP7, the A200's driver in Android is really under-developed).

Note that the first Krait-based SoC (MSM8960) will likely not ship with a 3xx GPU but rather a 225.

 

[Lazy8s]

The drivers for that SGX540 in the OMAP4430 benchmark score weren't incorporating all of the latest substantial SGX performance improvements that were recently applied to the Nexus S either, so OMAP4's jump above Hummingbird is actually a bit more than measured there.

It still won't come close to Adreno 220, but Adreno 220 is really competing with the 543MP2.

 

[Xmas]

It still won't come close to Adreno 220

What makes you think that?

 

[Deleted member 13524]

What makes you think that?

Among other things? Anand's benchmarks.

 

[Xmas]

Among other things? Anand's benchmarks.

I'm curious, what other things?

Anand only had one OMAP4 result to compare to, and that one is pretty much obsolete.

 

[Deleted member 13524]

I'm curious, what other things?

Release dates, power consumption, product orientation (tablets), marketing..

 

[Xmas]

Release dates, power consumption, product orientation (tablets), marketing..

I'm not sure what exactly you mean with each of these points. Could you elaborate?

 

[Ailuros]

Among other things? Anand's benchmarks.

Ok let's break them down then:

http://www.anandtech.com/show/4243/...mance-1-5-ghz-msm8660-adreno-220-benchmarks/2

LG Optimus 3D/SGX540@307MHz = 31,0 fps
Google Nexus S/SGX540@200MHz = 20,9 fps

If you do the math using the 20,9 score to scale it up to 307MHz you'll get 32,0 fps which is within the margin of error of any benchmark.

Now the Nexus S with the latest drivers looks like that:

http://www.glbenchmark.com/phonedetails.jsp?D=Google+Nexus+S&benchmark=glpro20

Google Nexus S/SGX540@200MHz = 26,8 fps

Scaling up to 307MHz:

LG Optimus 3D/SGX540@307MHz = 41,00 fps

Before you say now that there will be driver updates, favorable winds or a lottery win for Adreno, all IHVs constantly work on further driver improvements and that goes equally for Qualcomm, IMG, ARM, NVIDIA and whoever else. It still remains a valid point that the MSM8x60 smart-phone variant will run at somewhat smaller frequencies.

Adreno220 gives definitely a way better show then it's predecessors but considering that it truly runs at 800*480 after all in those benchmarks I can't say I'm blown out of my socks either.

 

[Deleted member 13524]

I'm not sure what exactly you mean with each of these points. Could you elaborate?

No, I won't elaborate self-explanatory terms.

Ok let's break them down then:

http://www.anandtech.com/show/4243/...mance-1-5-ghz-msm8660-adreno-220-benchmarks/2

LG Optimus 3D/SGX540@307MHz = 31,0 fps
Google Nexus S/SGX540@200MHz = 20,9 fps

If you do the math using the 20,9 score to scale it up to 307MHz you'll get 32,0 fps which is within the margin of error of any benchmark.

Now the Nexus S with the latest drivers looks like that:

http://www.glbenchmark.com/phonedetails.jsp?D=Google+Nexus+S&benchmark=glpro20

Google Nexus S/SGX540@200MHz = 26,8 fps

Scaling up to 307MHz:

LG Optimus 3D/SGX540@307MHz = 41,00 fps

Before you say now that there will be driver updates, favorable winds or a lottery win for Adreno, all IHVs constantly work on further driver improvements and that goes equally for Qualcomm, IMG, ARM, NVIDIA and whoever else. It still remains a valid point that the MSM8x60 smart-phone variant will run at somewhat smaller frequencies.

Adreno220 gives definitely a way better show then it's predecessors but considering that it truly runs at 800*480 after all in those benchmarks I can't say I'm blown out of my socks either.

Honestly.. it doesn't seem to be coming from someone as knowledgeable as you to assume that a driver update that brings a performance bump to a single synthetic benchmark is the same as increasing that same bump to the overall performance across the board.

Sure, the new driver boosts the Egypt score by roughly 28% on SGX540. Looking at the Pro bench, the driver boost is only 10%.

And we do we even know if that driver is going to be implemented in all OMAP4 devices? Or just some? Which ones?


Most tests are showing Adreno 205 trailing the SGX540 by 40 to 15%, 3D Mark Mobile shows the Adreno 220 having between 2x and 4.5x the performance of a 205.
That seems like a competitor to SGX543MP2 to me, but we'll only know for sure when a (preferably decent) benchmark is able to compare both GPUs in an apples-to-apples situation (or at least the same resolution, given it seems there won't be Android devices with SGX543MP2).

And why should we trust the Egypt test the most? It's a test where the graphics look bad and all GPUs across the board show poor performance.

 

[metafor]

Honestly.. it doesn't seem to be coming from someone as knowledgeable as you to assume that a driver update that brings a performance bump to a single synthetic benchmark is the same as increasing that same bump to the overall performance across the board.

Sure, the new driver boosts the Egypt score by roughly 28% on SGX540. Looking at the Pro bench, the driver boost is only 10%.

Note that because of the 60 fps cap on these devices, one shouldn't expect perfect scaling when approaching 60 fps.

Most tests are showing Adreno 205 trailing the SGX540 by 40 to 15%

Against SGX540 at 200MHz. The margin compared to SGX540 at 307MHz is larger though.

3D Mark Mobile shows the Adreno 220 having between 2x and 4.5x the performance of a 205.
That seems like a competitor to SGX543MP2 to me, but we'll only know for sure when a (preferably decent) benchmark is able to compare both GPUs in an apples-to-apples situation (or at least the same resolution, given it seems there won't be Android devices with SGX543MP2).

Possibly. However, given the iPad 2's scores on GLBenchmark even with 2x the pixels to push, there are at least some parts of SGX543MP2 that outperform the Adreno 220 significantly. Of course, we don't know how high Apple clocked the GPU. Perhaps they really pushed the clockspeed up.