AMD Carrizo Pre-release thread

[monstercameron]

I think that going htpc with Kaveri was not a good idea... I can't overclock stably that much on a 275w case psu. I might just go Godavari and build for high overclocks. It's a shame that we won't see desktop Carizzo.

 

[Abwx]

Gar reply got deleted. I'll replace it with tl;dr version:

So that total score looks pretty good for the Carrizo.

Looks to be as good as a discrete 4510U + GT840m combo, for the record thoses two devices TDPs are 28 and 30W.

http://www.3dmark.com/3dm11/8611687

Now let s wait for some leaks to see if thoses scores are real...

 

[maarten12100]

If they turn out to be true then I'm pleasantly surprised very much so!
But if the gpu perf is that good then where is the efficiency during those tests. Is it consuming 20W or 35W pushing out those numbers that is the question.

 

[Abwx]

NDA is lifted, a few infos about the design :

One module area 14,48mm².

102 millions transistors,(86 millions for a Steamroller module)

L1 data cache increased to 32 KB from 16KB

10% higher GPU frequency at same comsumption or 20% lower comsumption at same frequency.

Dedicated supply voltage for the GPU (VDDGFX)

AVFS Response time 1ns

Rest of the goodies (and there s a collection) can be checked here :

http://www.hardware.fr/news/14085/amd-leve-voile-carrizo-cote-technique.html

Or here for in depth analysis at Anandtech and SA.

http://www.anandtech.com/show/8995/amd-at-isscc-2015-carrizo-and-excavator-details

http://semiaccurate.com/forums/showthread.php?p=230073#post230073

 

[mrmt]

If they turn out to be true then I'm pleasantly surprised very much so!
But if the gpu perf is that good then where is the efficiency during those tests. Is it consuming 20W or 35W pushing out those numbers that is the question.

Let's see... Carrizo brings 20% more performance but only in low wattage mobile parts, and it seems that its power management is still inferior to intel's. All they have to show is gaming, but in a TDP low enough to be meaningful. The CPU performance is still atrocious, and all they have to show is a 5% increase in IPC and maybe a bit of clocks.

Wake me up when Zen flops.

 

[Abwx]

If they turn out to be true then I'm pleasantly surprised very much so!
But if the gpu perf is that good then where is the efficiency during those tests. Is it consuming 20W or 35W pushing out those numbers that is the question.

The max TDP of Carrizo is 35W, to get the same level of perfs with Intel you have to follow the advice we hear here and there when it comes to Desktops, that is, adding a discrete GPU, looking at the numbers below of such a solution it is clear that with Carrizo AMD has the perf/watt crown for mobile segment.

http://www.notebookcheck.net/Acer-Aspire-E5-771G-Notebook-Review.136653.0.html

55W power comsumption for an equivalent Intel solution helped by a Nvidia GPU.

Some explanation at Anandtech :

It seems a little odd applying a ‘high-density’ design to a ‘high-performance’ process node, but AMD is stating that part of this has been driven by the GPU team sharing its experiences and knowledge of small, efficient die components with the CPU team, allowing the lessons learned there to benefit AMD's CPU designs

It was time that there s such synergies but at last it s here and it seems that the results are getting some people in full panic mode...

 

[Vesku]

Let's see... Carrizo brings 20% more performance but only in low wattage mobile parts, and it seems that its power management is still inferior to intel's. All they have to show is gaming, but in a TDP low enough to be meaningful. The CPU performance is still atrocious, and all they have to show is a 5% increase in IPC and maybe a bit of clocks.

Wake me up when Zen flops.

i3-4010U is atrocious CPU performance and efficiency? If AMD's presentation is accurate there should be a Carrizo SKU that roughly matches the 4010Us CPU performance and battery life.

 

[Enigmoid]

The max TDP of Carrizo is 35W, to get the same level of perfs with Intel you have to follow the advice we hear here and there when it comes to Desktops, that is, adding a discrete GPU, looking at the numbers below of such a solution it is clear that with Carrizo AMD has the perf/watt crown for mobile segment.

http://www.notebookcheck.net/Acer-Aspire-E5-771G-Notebook-Review.136653.0.html

55W power comsumption for an equivalent Intel solution helped by a Nvidia GPU.

Some explanation at Anandtech :

It was time that there s such synergies but at last it s here and it seems that the results are getting some people in full panic mode...

55W Furmark + Prime system power consumption (360 nit 1080p 17" display is not helping). 2.5-2.6 on the CPU and 1.1 ghz on the gpu (no throttling).

Secondly Carrizio is not going to beat an i7 BW-U in CPU performance. The singlethread performance of that mobile chip can go head to head with an overclocked steamroller and the multithread is no slouch either.

That CPU has a 3dmark 11 physics score of ~4000 vs. the 3100 of the carrizio prototype. Maybe an i5 version.

 

[geoxile]

NDA is lifted, a few infos about the design :

One module area 14,48mm².

102 millions transistors,(86 millions for a Steamroller module)

L1 data cache increased to 32 KB from 16KB

10% higher GPU frequency at same comsumption or 20% lower comsumption at same frequency.

Dedicated supply voltage for the GPU (VDDGFX)

AVFS Response time 1ns

Rest of the goodies (and there s a collection) can be checked here :

http://www.hardware.fr/news/14085/amd-leve-voile-carrizo-cote-technique.html

Or here for in depth analysis at Anandtech and SA.

http://www.anandtech.com/show/8995/amd-at-isscc-2015-carrizo-and-excavator-details

http://semiaccurate.com/forums/showthread.php?p=230073#post230073

What are these mid to late 2015 devices that have integrated voltage regulation? I thought it was going to be Carrizo but it doesn't look that way.

 

[richaron]

Wake me up when Zen flops.

I think we'd all like to see less FUD and general negativity around here for a while...

I'm currently typing on a 45W Kaveri desktop (a8-7600) and I don't ever notice the "attrocious" CPU performance. Plus I can use fancy 3d desktop effects (linux) and boot up CS every now and then.

This or better in a small/cheap notebook would be pretty sweet.

 

[AnandThenMan]

http://www.anandtech.com/show/8995/amd-at-isscc-2015-carrizo-and-excavator-details

Looks extremely promising. :thumbsup:

 

[Abwx]

55W Furmark + Prime system power consumption (360 nit 1080p 17" display is not helping). 2.5-2.6 on the CPU and 1.1 ghz on the gpu (no throttling).

Secondly Carrizio is not going to beat an i7 BW-U in CPU performance. The singlethread performance of that mobile chip can go head to head with an overclocked steamroller and the multithread is no slouch either.

That CPU has a 3dmark 11 physics score of ~4000 vs. the 3100 of the carrizio prototype. Maybe an i5 version.

I didnt notice the display.

As for the single threaded perf granted but this no more relevant, otherwise there wouldnt be 4T CPUs, the perfs is about the one of an i5 4210U wich score 3256 on the physics test, the i7 above has obviously more available TDP than this i5 given the 4000 score, we can even evaluate it to 40% more TDP given the score delta, BDW is not more efficient than HW, the i7 5600U laptop below is at 48.8W :

http://www.notebookcheck.net/First-Impressions-Lenovo-ThinkPad-T450s.136641.0.html

A lush base clock rate of 2.6 GHz makes the Core i7 5600U to the currently fastest CPU in the economic 15 watt category.

Another review with "only" 44.3W but a 12.5" display.

http://www.notebookcheck.net/Lenovo-ThinkPad-X250-Ultrabook-Review.136646.0.html

GPU score is in the 1100-1200 region, that s not enough to get the perf/Watt advantage for the whole item.

 

[Abwx]

What are these mid to late 2015 devices that have integrated voltage regulation? I thought it was going to be Carrizo but it doesn't look that way.

Look like they learned from Intel s Haswell...?..

They have a separate regulation for the GPU, though.

Looks extremely promising. :thumbsup:

This is even more remarkable when accounting for their limited possibilities process wise.

 

[beginner99]

Looks extremely promising. :thumbsup:

It does. But then these are marketing slides and I will wait for real-world numbers to verify the claims.

 

[DrMrLordX]

At least AMD has an upcoming product that they are willing to discuss in public. Carrizo looks pretty good. It definitely looks better than mobile Kaveri.

Anandtech's article made brief mention of the differences in Carrizo's HSA Compliance vs. Kaveri's HSA "Readiness". What are the implications of GPU context switching? Is this a feature that is sorely lacking in Kaveri?

 

[Vesku]

At least AMD has an upcoming product that they are willing to discuss in public. Carrizo looks pretty good. It definitely looks better than mobile Kaveri.

Anandtech's article made brief mention of the differences in Carrizo's HSA Compliance vs. Kaveri's HSA "Readiness". What are the implications of GPU context switching? Is this a feature that is sorely lacking in Kaveri?

Should improve touch interfaces and, if I'm reading correctly, better system responsiveness when running multiple GPU compute programs.

6. GPU compute context switch and GPU graphics pre-emption: GPU tasks can be context switched, making the GPU in the APU a multi-tasker. Context switching means faster application, graphics and compute interoperation. Users get a snappier, more interactive experience. As UI's are becoming increasing more touch focused, it is critical for applications trying to respond to touch input to get access to the GPU with the lowest latency possible to give users immediate feedback on their interactions. With context switching and pre-emption, time criticality is added to the tasks assigned to the processors. Direct access to the hardware for multi-users or multiple applications are either prioritized or equalized

From http://www.anandtech.com/show/5847/...geneous-and-gpu-compute-with-amds-manju-hegde

 

[Shehriazad]

Am I the only one that is impressed by the performance/watt ratio of those 28nm pieces?

Since they're SoCs they will probably end up being very cheap for notebooks/"ultrabooks".

The power draw should also be insanely low for the entire system...thus "yay battery life". (Doesn't it have the power to beat intel in its' respective TDP area? Some intel savvy people can feel free to enlighten me on this)

Well...gotta wait for some actual products, I guess.

 

[maarten12100]

It does. But then these are marketing slides and I will wait for real-world numbers to verify the claims.

ISSCC is marketing not aimed at the consumer but at the industry. Putting lies on those slides will get you called out and trashed later on. :awe:

 

[Enigmoid]

I didnt notice the display.

As for the single threaded perf granted but this no more relevant, otherwise there wouldnt be 4T CPUs, the perfs is about the one of an i5 4210U wich score 3256 on the physics test, the i7 above has obviously more available TDP than this i5 given the 4000 score, we can even evaluate it to 40% more TDP given the score delta, BDW is not more efficient than HW, the i7 5600U laptop below is at 48.8W :

http://www.notebookcheck.net/First-Impressions-Lenovo-ThinkPad-T450s.136641.0.html

Another review with "only" 44.3W but a 12.5" display.

http://www.notebookcheck.net/Lenovo-ThinkPad-X250-Ultrabook-Review.136646.0.html

GPU score is in the 1100-1200 region, that s not enough to get the perf/Watt advantage for the whole item.

You can't compare haswell and broadwell like that under max load because Haswell is throttling and broadwell is not (in most reviews, sometimes it is).

Performance is impressive too. 3.3 point in CB multi is about equivalent to an i3 ivy desktop chip.

 

[RandSec]

Anandtech's article made brief mention of the differences in Carrizo's HSA Compliance vs. Kaveri's HSA "Readiness". What are the implications of GPU context switching? Is this a feature that is sorely lacking in Kaveri?

GPU context switching and interruptability are HSA hardware features, and are indeed lacking in Kaveri, along with hardware micro-task priority queues and dispatch. Of course, they are also lacking in every other chip made, and their absence has not been particularly "sore" thus far. The features would be "sorely" lacking if the intent was to maximize performance in very low level HSA coding. But the micro-task hardware also could have more conventional uses, such as real time coding.

 

[Valantar]

Looks to be as good as a discrete 4510U + GT840m combo, for the record thoses two devices TDPs are 28 and 30W.

http://www.3dmark.com/3dm11/8611687

Nope, the 4510U is a 15W part, the Futuremark specs are wrong.
http://ark.intel.com/m/products/810...o-3_10-GHz?q=I7-4510u#@product/specifications

Still, an okay CPU and a decent GPU in a 35W TDP is pretty great. I'd love that on an ITX board for my living room PC.

Also, stick something like this in a reasonably thin and light laptop with a good screen, decent battery life, a not too shabby build and a suitable price, and it should sell like hotcakes to the DotA (and other such games) crowd.

 

[Abwx]

Nope, the 4510U is a 15W part, the Futuremark specs are wrong.
http://ark.intel.com/m/products/810...o-3_10-GHz?q=I7-4510u#@product/specifications

3D Mark is 100% right.

What is wrong is Intel sTDPs wich are obviously hugely underspecced, same as BDW 5600U, that is 15W TDP at ark.intel but roughly 30W+ once ypu test such a product.

44W at notebookcheck for a 4510U equiped laptop :

http://www.notebookcheck.net/Dell-Inspiron-17-5748-Notebook-Review.122422.0.html

People should definitly pay attention to measurement and stop using official specs that are obviously completely wrong.

 

[Shehriazad]

Alone based on the fact that it's a SoC should make it very competetive on the total system power draw compared to some intel products, aye?

But still...gotta wait until we actually get to buy such products. I'm really looking forward to some benchmarks/reviews of finished products.

 

[NTMBK]

3D Mark is 100% right.

What is wrong is Intel sTDPs wich are obviously hugely underspecced, same as BDW 5600U, that is 15W TDP at ark.intel but roughly 30W+ once ypu test such a product.

44W at notebookcheck for a 4510U equiped laptop :

http://www.notebookcheck.net/Dell-Inspiron-17-5748-Notebook-Review.122422.0.html

People should definitly pay attention to measurement and stop using official specs that are obviously completely wrong.

It's a freaking 17" laptop, how many watts do you think go into driving that display?

 

[frozentundra123456]

It's a freaking 17" laptop, how many watts do you think go into driving that display?

Dont confuse him with facts. In his universe, every intel cpu draws 2x its TDP and AMD's draw 1/2.