AMD Carrizo Pre-release thread

[Fjodor2001]

Such wishful thinking, its almost funny.

Based on past history, yes. But I really don't get why the OEMs keep screwing up their AMD based products, often pairing them with poor screens, keyboard, and cheap components giving them poor battery life. What's the point for an OEM of releasing a product if you don't actually want it to sell?!

 

[monstercameron]

Based on past history, yes. But I really don't get why the OEMs keep screwing up their AMD based products, often pairing them with poor screens, keyboard, and cheap components giving them poor battery life. What's the point for an OEM of releasing a product if you don't actually want it to sell?!

People will buy them because its cheap but won't help amds rep.

 

[Fjodor2001]

People will buy them because its cheap but won't help amds rep.

Yeah, and there are of course valid reasons to cover that segment too. But why don't at least some OEMs cover the segment where it's paired with other high quality components (display, keyboard, ...)?

For the Intel CPUs, we get both cheap models, mid range, and high end models. So why can't we get the same with AMD based laptops?

 

[monstercameron]

Yeah, and there are of course valid reasons to cover that segment too. But why don't at least some OEMs cover the segment where it's paired with other high quality components (display, keyboard, ...)?

For the Intel CPUs, we get both cheap models, mid range, and high end models. So why can't we get the same with AMD based laptops?

I don't know, but if I were to hazard a guess, I would think it doesn't sell well enough to invest in amds platform. Asus u38n comes to mind.

 

[Fjodor2001]

I don't know, but if I were to hazard a guess, I would think it doesn't sell well enough to invest in amds platform. Asus u38n comes to mind.

Then why not skip it altogether?

 

[monstercameron]

Then why not skip it altogether?

Another guess is that they keep and as a second source in case something goes awry with Intel like delays or low stocks.

 

[Fjodor2001]

Sorry, still don't see the point of making it crappier than it has to be.

Also, why no 13.3" or below?

 

[maarten12100]

Sorry, still don't see the point of making it crappier than it has to be.

Also, why no 13.3" or below?

Because otherwise nobody is going to give them their bribe mo... I mean contra revenue check.

Carrizo is shapping up to be a good product an amazing product at the 15W spectrum. And cheap to put into a design versus the 300 dollar intel chips. Yet it isn't being used to the fullest in the highest quality ULV machines. I wonder why...

Actually I don't really I just hope there will be a 14" 1080P notebook that fits my needs. That's all I'm asking for. :ninja:

 

[podspi]

Okey so this is a very very good performing product - but who needs that gpu power in a 15w tdp envelope?

I mean i would like it but how many wants to game on their ultra portable 13.3 machine?

I can say for sure a zen machine with hbm on 14nm gets my money. This carizo sets my expectations high and i just want it now !

I do, though usually it is streaming something from the desktop in the basement. :biggrin:

Wouldn't be terrible to have some oomph when traveling, though.

 

[Abwx]

I don't know, but if I were to hazard a guess, I would think it doesn't sell well enough to invest in amds platform. Asus u38n comes to mind.

Kaveri couldnt compete at those low power, Haswell hasnt specialy better perf/Watt than Kaveri at 15W, but like Beema it has noticeably lower idle power, and that s this detail that made all the difference, Carrizo is in another league as it has undoubtly the perf/Watt crown for both CPU and GPU usages while idle power seems competitive.

 

[Enigmoid]

Yes, i would actually prefer 35watts to make lite gaming more viable. 15 watts is still going to be pretty mediocre for gaming, and i think you will get better battery life and cpu performance from intel, albeit probably at a higher price. Intel actually seems to be abandoning the quad core, higher tdp laptop market, so it would seem like a good opportunity for amd to move in tbere.

47W intel chips are going strong. The market as a whole is moving toward thin and light.

Yeah, and there are of course valid reasons to cover that segment too. But why don't at least some OEMs cover the segment where it's paired with other high quality components (display, keyboard, ...)?

For the Intel CPUs, we get both cheap models, mid range, and high end models. So why can't we get the same with AMD based laptops?

AMD lacks brand image to compete well at higher prices.

 

[pablo87]

Very impressive numbers for Carrizo - 7750 is listed at 3160.

 

[frozentundra123456]

Are there 47 watt quad broadwell chips launched yet? Actually, lookint at Arc Intel, there are a lot of 47 watt quads, a lot with iris pro actually, but I havent seen any tests. I would be really interested to see what iris pro can do in a laptop.

Edit: sorry, was thinking this was the broadwell thread, dont mean to derail it.

 

[Abwx]

Are there 47 watt quad broadwell chips launched yet? Actually, lookint at Arc Intel, there are a lot of 47 watt quads, a lot with iris pro actually, but I havent seen any tests. I would be really interested to see what iris pro can do in a laptop.

Edit: sorry, was thinking this was the broadwell thread, dont mean to derail it.

If you had actualy no intention to thread crap this thread you would had edited your post this way for instance :

Edit: sorry, wrong thread

 

[Abwx]

And it's still on 28 nm. Imagine what they can do next year on 14 nm and completely new uArch.

The 28nm process is obviously not up to the 14nm of the competition but the concept of CMT start to pay off since they increased IPC substancialy and got some kind of 1/2 node power efficency improvement at the expense of 10% lower max frequency than Kaveri.

Now with GF announced, and quantified, improvements i would say that on the mobile side they dont even need Zen to be more than competitive, a simple shrink would get a chip almost twice as efficient at 2.4GHz, and likely up to Carrizo s higher frequencies as GF/Samsung 14nm LPP s process is more efficient than Intel s 14nm by a substancial 50% margin, there s interesting times to come...

 

[Arachnotronic]

as GF/Samsung 14nm LPP s process is more efficient than Intel s 14nm by a substancial 50% margin, there s interesting times to come...

Proof?

 

[Abwx]

Proof?

It did surely require only a handfull of seconds for the enginers working in the field to get to such a conclusion once they saw GF s numbers.

If absolute values are generaly hardly computable relative electrical differences can be extracted quite easily, as an exemple let s use intel s
22 to 14nm transition, the comparison is possible because the compared designs are close.

First they said that parasistic capacitance was reduced by 0.65, from here if the rest is unchanged they should have a CPU 54% faster at same efficency or 2.36x more efficient at same frequency, wich of course didnt occur because there s a second parameter to consider, that is the (trans)conductance of the device, this latter parameter was degraded.

The 14nm transistors conduct less and to get them conduct more you ll have to increase the supply voltage, the amplitude of this excess voltage along with said 0.65 ratio is all we need to deduce the ratios of transconductances, in our case it s within 1.135 for the voltage, so the number is 1.135/0.65 = 1.746, or said otherwise 43% less transconductance.

The energy efficency improvement can be evaluated this way :

0.65 x (1.135)^2 = 0.84, that is BDW consume 16% less than HW, or said otherwise has 19.5% better perf/Watt.

That s not taking account of the IPC improvement that will take a toll in this number, given their official 5% number this should yield about 10-15% better efficency.

Likewise GF publication was a comparison of a same design using their 28nm and their licenced 14nm LPP, from their numbers they have either 50% higher frequency (up to 2.4GHz) at same efficency, or twice the efficency at same frequency (up to 2.4GHz) compared to what Intel achieved at 14nm.

 

[Arachnotronic]

It did surely require only a handfull of seconds for the enginers working in the field to get to such a conclusion once they saw GF s numbers.

If absolute values are generaly hardly computable relative electrical differences can be extracted quite easily, as an exemple let s use intel s
22 to 14nm transition, the comparison is possible because the compared designs are close.

First they said that parasistic capacitance was reduced by 0.65, from here if the rest is unchanged they should have a CPU 54% faster at same efficency or 2.36x more efficient at same frequency, wich of course didnt occur because there s a second parameter to consider, that is the (trans)conductance of the device, this latter parameter was degraded.

The 14nm transistors conduct less and to get them conduct more you ll have to increase the supply voltage, the amplitude of this excess voltage along with said 0.65 ratio is all we need to deduce the ratios of transconductances, in our case it s within 1.135 for the voltage, so the number is 1.135/0.65 = 1.746, or said otherwise 43% less transconductance.

The energy efficency improvement can be evaluated this way :

0.65 x (1.135)^2 = 0.84, that is BDW consume 16% less than HW, or said otherwise has 19.5% better perf/Watt.

That s not taking account of the IPC improvement that will take a toll in this number, given their official 5% number this should yield about 10-15% better efficency.

Likewise GF publication was a comparison of a same design using their 28nm and their licenced 14nm LPP, from their numbers they have either 50% higher frequency (up to 2.4GHz) at same efficency, or twice the efficency at same frequency (up to 2.4GHz) compared to what Intel achieved at 14nm.

Intel claimed a 1.6x perf/watt improvement in going from 22nm -> 14nm, though. If you're going to use GloFo/Samsung marketing, you should surely also consider Intel's marketing claims

 

[Abwx]

Intel claimed a 1.6x perf/watt improvement in going from 22nm -> 14nm, though. If you're going to use GloFo/Samsung marketing, you should surely also consider Intel's marketing claims

So if if Intel said that it was 10x you would be selling the idea that it s possible.?.

Actualy it s you that are relying on misleading marketing slides to try to argue...



The number about intel are not from GF, that s the prove that you didnt understand my explanations, they can be deduced from the very infos published by Intel, really i ve no time to waste commenting their non sense when their products numbers say completely otherwise.

The 1.6 improvement would had been possible if they managed to keep the transconductance as good as with their 22nm and this should have required to keep the same voltages or so, wich is obviously not the case and is proved fact.

Ask the guys at EE Times if you want, and try to contradict them if ever they do not agree with Intel s 1.6x fairy tale, and i hope that you ll present better arguments than this marketing slide, although it could be a good idea to relax the atmosphere there, those people generaly are dead serious, i m sure that they would appreciate some unexpected and creative tech fun...


Edit : Gf didnt do marketing claims, they publicly displayed electrical caracteristics in an enginers dedicated conference, not in a marketing event, never Intel provided such numbers at any event or conferneces.

 

[IntelUser2000]

The 1.6 improvement would had been possible if they managed to keep the transconductance as good as with their 22nm and this should have required to keep the same voltages or so, wich is obviously not the case and is proved fact.

While Intel has been sly in their marketings the competitors are no different. I wholeheartedly agree with this statement from RWT:

From 0.25um to 0.13um: golden age of Dennard scaling
From 90nm to 32nm: silver age of strain engineering for high performance logic.
From 22nm to 14nm: bronze age of tri-gate.

More exotic materials, and less gains. Welcome to world of semiconductors 2015. I doubt either 14nm Zen or 10nm III-V transistor based CPUs or even carbon nanotube based ones will make any meaningful difference. By the time those technologies(CNT, III-V) are available, we might be lucky to get 5% IPC on a tock.

 

[IntelUser2000]

The fanny thing is the 35W TDP FX8800P 3D Mark 11 score is close to 65W TDP Broadwell HD6200 and faster than 95W Kaveri.

2750 in 3DMark 11 means the hailed Iris Pro 6200 with "eDRAM" is only 10% faster than 28nm Carrizo.

That's amazing. If we assume only 20% gains to 14nm and add single HBM chip in there, its almost for sure that at least for GPU it'll beat anything Intel has there at a much cheaper price. And since Intel is at diminishing returns to CPU gains and AMD will gain a one time big gain with Zen, that means we'll have a true "APU".

 

[Silverforce11]

2750 in 3DMark 11 means the hailed Iris Pro 6200 with "eDRAM" is only 10% faster than 28nm Carrizo.

That's amazing. If we assume only 20% gains to 14nm and add single HBM chip in there, its almost for sure that at least for GPU it'll beat anything Intel has there at a much cheaper price. And since Intel is at diminishing returns to CPU gains and AMD will gain a one time big gain with Zen, that means we'll have a true "APU".

It'll certainly be nice to have some competition on the x86 market, hopefully Zen lives up to its hype, if it does, I'm building a tiny gaming rig on it.

 

[coercitiv]

2750 in 3DMark 11 means the hailed Iris Pro 6200 with "eDRAM" is only 10% faster than 28nm Carrizo.

We should keep in mind 3DMark scores usually favor Intel more than real world performance.

Do we know if the iGPU in Carrizo has some improved bandwidth saving mechanisms relative to Kaveri?

 

[DrMrLordX]

I don't think anyone knows for sure if Carrizo uses Tonga-like color compression. It might be in some of the Carrizo release slides. Regardless, it seems like something like that has come into play.

Intel and AMD are definitely raising the profile of the APU as gaming and computational devices. The real loser here (if any) is Kaveri/Kaveri refresh. That APU is being left behind.

 

[Idontcare]

It did surely require only a handfull of seconds for the enginers working in the field to get to such a conclusion once they saw GF s numbers.

If absolute values are generaly hardly computable relative electrical differences can be extracted quite easily, as an exemple let s use intel s
22 to 14nm transition, the comparison is possible because the compared designs are close.

First they said that parasistic capacitance was reduced by 0.65, from here if the rest is unchanged they should have a CPU 54% faster at same efficency or 2.36x more efficient at same frequency, wich of course didnt occur because there s a second parameter to consider, that is the (trans)conductance of the device, this latter parameter was degraded.

The 14nm transistors conduct less and to get them conduct more you ll have to increase the supply voltage, the amplitude of this excess voltage along with said 0.65 ratio is all we need to deduce the ratios of transconductances, in our case it s within 1.135 for the voltage, so the number is 1.135/0.65 = 1.746, or said otherwise 43% less transconductance.

The energy efficency improvement can be evaluated this way :

0.65 x (1.135)^2 = 0.84, that is BDW consume 16% less than HW, or said otherwise has 19.5% better perf/Watt.

That s not taking account of the IPC improvement that will take a toll in this number, given their official 5% number this should yield about 10-15% better efficency.

Likewise GF publication was a comparison of a same design using their 28nm and their licenced 14nm LPP, from their numbers they have either 50% higher frequency (up to 2.4GHz) at same efficency, or twice the efficency at same frequency (up to 2.4GHz) compared to what Intel achieved at 14nm.

I agree with Abwx, for whatever the electrical engineering reasoning, when it comes to Intel's 14nm the promised goods have simply yet to show up. Not a good sign when they are well more than a year into HVM on the node.

It suggests they have/had a lot of reasons to go quiet on 10nm. Sure they have successfully spun it as a strategic decision, competitive advantage, etc etc...but the 14nm situation is proving out to be a bit of a 90nm déjà vu thus far.