Some New AMD Carrizo Details

[carolkarine]

That's false. Sorry. TDPs don't lie. CPU tend to get hotter, but that's just because they don't have as much space for a heatsink (yep, it's the other way around, gpus get to have VERY long heatsinks, comaprable to nh-d15s in terms of volume, CPUs only start to get the advantge when we step into liquid cooling and 3x120mm radiators, which would hardly make sense on an APU) and fans and because they have a lower surface area (which would apply to an APU too), whereas a gpu can go balls to the wall with heatpipes and even triple 120mm fans. For reference, my core i7 2600 (95 watts tdp) never broke 75c on the silly stock cooler, which has a very small heatsink and a very weak fan. My R9 290 will reach that temperature even with two large fans and a ginormous heatsink with heatpipes everywhere. The 285 isn't quite as hot as the 290, but it's no refrigerator either. And together with a cpu it generates way more heat than is acceptable.

That's ALSO a problem. But heat would be the deal breaker for me, APUs are supposed to be semi-budget solutions, not something I need to buy a 100$ heatsink for just to run it at stock.

Nope, you're wrong. Cpus put off less heat, and run hotter despite having larger heatsinks. The cpu die has more concentrated heat- making it much more difficult to properly dissipate the lesser amount of heat a cpu generates. Gpus, namely the ones that put off large amounts of heat, are often twice the size physically than a cpu. Second, a cpu has an IHS, which, upon delidding, can often drop temps ten degrees. So because Gpus first are easier to extract heat from because the heat is more spread out within the larger die, and because there isn't an IHS, despite Gpus producing often twice as much heat, Cpus run hotter.

[Agosto]

Carizzo is a really weird name..... Sound Italian

 

It's Carrizo and it isn't an Italian word.

 

It's Spanish and means "reed"

[Sauron]

That cpu cooler is meant to be for a 100watt cpu. the 460 is a 160 watt gpu. It runs cooler on the gpu, than a 90 watt cpu

 

I don't see a check of the cpu temperatures with that thing on in that video. I'd bet you they're lower than 60c on full load. And still, bare in mind that VVVVV that also is a point; including a gpu in a package the size of  cpu would break all hell loose.

 

Nope, you're wrong. Cpus put off less heat, and run hotter despite having larger heatsinks. The cpu die has more concentrated heat- making it much more difficult to properly dissipate the lesser amount of heat a cpu generates. Gpus, namely the ones that put off large amounts of heat, are often twice the size physically than a cpu. Second, a cpu has an IHS, which, upon delidding, can often drop temps ten degrees. So because Gpus first are easier to extract heat from because the heat is more spread out within the larger die, and because there isn't an IHS, despite Gpus producing often twice as much heat, Cpus run hotter.

 

...which happens to be my point... I was saying gpus put off more heat than cpus. And although I didn't include the die size fact, I know it stands. On top of that gpus get to have larger stock heatsinks. And therefore even if cpus tend to get hotter, it's in a completely different scenario and hardly comparable. If we included a gpu of R9 285 caliber into a cpu, cooling it would be extremely hard, not because the cpu is hard to cool, but because the gpu would loose its natural advantage and still put out 190 watts of heat. If cooling a 95 watt cpu is hard, imagine a 230 watt one (or just think of the 9590, which would hardly make any sense for a budget solution if for anything).

[marldorthegreat]

I don't see a check of the cpu temperatures with that thing on in that video. I'd bet you they're lower than 60c on full load. And still, bare in mind that VVVVV that also is a point; including a gpu in a package the size of  cpu would break all hell loose.

 

 

...which happens to be my point... I was saying gpus put off more heat than cpus. And although I didn't include the die size fact, I know it stands. On top of that gpus get to have larger stock heatsinks. And therefore even if cpus tend to get hotter, it's in a completely different scenario and hardly comparable. If we included a gpu of R9 285 caliber into a cpu, cooling it would be extremely hard, not because the cpu is hard to cool, but because the gpu would loose its natural advantage and still put out 190 watts of heat. If cooling a 95 watt cpu is hard, imagine a 230 watt one (or just think of the 9590, which would hardly make any sense for a budget solution if for anything).

How come the gtx 640, with a tdp of 75 watts, has this heatsink, which keeps it at 60c under load, but the Intel® Core™ i7-3770  (75 watt cpu's stock heatsink is larger, and is louder and runs hotter

[Sauron]

How come the gtx 640, with a tdp of 75 watts, has this heatsink, which keeps it at 60c under load, but the Intel® Core™ i7-3770  (75 watt cpu's stock heatsink is larger, and is louder and runs hotter

 

Because of die size limitations and IHS. With the same TDP, it IS harder to dissipate a cpu, but when it's twice as much the gpu gets hotter.

[rentaspoon]

<p>

There hasn't been much talk lately about Carrizo so why not.

In a recent video of Lisa Su talking about the future of the semiconductor technology she is found playing with a Carrizo engineering sample.

Revealing a close up shot of the top of the chip and it's strings.

The identification string is as follows.

ZM18B8C1Y4382

We can break down some of the string using the Bulldozer family identification information.

• Z = Brand
• M = Mobile
• 18B8 = Model (ES 1.8 GHz Base 3.4 GHz Turbo)
• C1 = TDP (15w/35w)
• Y = Package (FP4)
• 4 = Cores (4 Cores)
• 3 = Cache (1 MB L2 Per Module)
• 82 = Stepping

so basicly this is the same chip that's in my laptop already?

[Guest]

It's Carrizo and it isn't an Italian word.

 

It's Spanish and means "reed"

Oh yeah? Well you're a Reed!

Ha!

[vm'N]

*AMD is claiming a 30% performance increase at 15w with Excavator. If that relates to IPC increase over Steamroller the chip should be really impressive (40% increase in IPC over Piledriver). We know so far that the chip runs at a 1.8 GHz base clock and turbo's up to 3.4 GHz (compared to the 2.1 GHz base and 3.3 GHz turbo of the FX-7500 at 19w). With such a drastic increase in IPC Carrizo will out perform the desktop 750k in single thread performance easily while in its boost states. If that all adds up a desktop variant of Carrizo would essentially destroy the sales of their own FX microprocessors. Not many would buy an FX-8350 when Carrizo performs like Sandy Bridge. With that being said I still think it's a possibility tho just not practical for AMD to release an APU of that magnitude to the consumer market. They have bigger problems at hand to address such as feeding the iGPU memory bandwidth (e.g. the incorporation of on package HBM). *AMD has moved away from the desktop APU ecosystem to focus on high performance CPUs.

 

The PS4 APU has 1152 SP's (18 compute units). Tho supposedly die shots of the APU show there are 20 compute units on board (2 disabled).

* It will most likely be a IPC improvement (15%) and clock-speed improvement, when compared towards a kaveri at same TDP (only for low end carrizo).

I expect excavator to hit a performance wall around 3GHz, and its 3.4GHz turbo will be mostly unreachable without having other parts been idle (like the iGP).

 

Excavator might even start to lose against kaveri when you go over 35w TDP, hence why we see no part after the 35w TDP range.

 

** That is a bold statement. AMD is just expanding their market instead of been sitting on a niche market.

[Agosto]

Oh yeah? Well you're a Reed!

Ha!

 

 

wut

[Guest]

wut

Just a joke

[Opcode]

* It will most likely be a IPC improvement (15%) and clock-speed improvement, when compared towards a kaveri at same TDP (only for low end carrizo).

I expect excavator to hit a performance wall around 3GHz, and its 3.4GHz turbo will be mostly unreachable without having other parts been idle (like the iGP).

 

Excavator might even start to lose against kaveri when you go over 35w TDP, hence why we see no part after the 35w TDP range.

 

** That is a bold statement. AMD is just expanding their market instead of been sitting on a niche market.

With whatever amount of IPC increase over Bdver3 validations on SiSoft are putting Carrizo around the GFLOPS of a mobile Sandy Bridge i3 @ 2.0 GHz while at 15w TDP. While not competing clock for clock (not important) the power consumption could show impressive in comparison. With Intel not yielding any major IPC increases from Sandy Bridge all the way through Broadwell. Carrizo is shaping up to possibly be a bigger success in the mobile market than any of their past mobile generations. Intel undoubtedly still easily has the upper hand tho it's nice to finally possibly nice see an AMD CPU capable of competing within range of Sandy Bridge in performance at much lower power consumption. The 3.4 GHz I would imagine is only effective within thermal limits (iGPU would have to be idle).

[Deletive]

With whatever amount of IPC increase over Bdver3 validations on SiSoft are putting Carrizo around the GFLOPS of a mobile Sandy Bridge i3 @ 2.0 GHz while at 15w TDP. While not competing clock for clock (not important) the power consumption could show impressive in comparison. With Intel not yielding any major IPC increases from Sandy Bridge all the way through Broadwell. Carrizo is shaping up to possibly be a bigger success in the mobile market than any of their past mobile generations. Intel undoubtedly still easily has the upper hand tho it's nice to finally possibly nice see an AMD CPU capable of competing within range of Sandy Bridge in performance at much lower power consumption. The 3.4 GHz I would imagine is only effective within thermal limits (iGPU would have to be idle).

Are we talking core per core or a 4 core compared to 2 cores? 

 

[Opcode]

Are we talking core per core or a 4 core compared to 2 cores? 

4 threads vs 4 threads

[Faa]

Because of die size limitations and IHS. With the same TDP, it IS harder to dissipate a cpu, but when it's twice as much the gpu gets hotter.

The die size has nothing to do with other than it could block you from soldering the die to the HSP which is the case for Intels TIM'ed CPU's. Feel free to compare a 5960x (has effectively 8 cores) to a 12 core xeon at the same clocks/voltage/corecount, you won't see any difference.

S2011 chips run a lot cooler than lga1150/lga1155 at the exact same power draw because the die is soldered and also the pre-installed backplate is making a huge difference though. When I moved from a 2600K @ 5GHz to my 3930K, turning 2 cores off and setting it to 5GHz it was literally running 20° cooler.

 

 

How come the gtx 640, with a tdp of 75 watts, has this heatsink, which keeps it at 60c under load, but the Intel® Core™ i7-3770  (75 watt cpu's stock heatsink is larger, and is louder and runs hotter

TDP is a cooling spec, they all define their TDP values differently because they feel different about the max safe operating temperatures which is one out of the so many variables.

[marldorthegreat]

so basicly this is the same chip that's in my laptop already?

Why quote the ENTIRE post?????

[Agosto]

 

The die size has nothing to do with other than it could block you from soldering the die to the HSP which is the case for Intels TIM'ed CPU's. Feel free to compare a 5960x (has effectively 8 cores) to a 12 core xeon at the same clocks/voltage/corecount, you won't see any difference.

S2011 chips run a lot cooler than lga1150/lga1155 at the exact same power draw because the die is soldered and also the pre-installed backplate is making a huge difference though. When I moved from a 2600K @ 5GHz to my 3930K, turning 2 cores off and setting it to 5GHz it was literally running 20° cooler.

 

 

TDP is a cooling spec, they all define their TDP values differently because they feel different about the max safe operating temperatures which is one out of the so many variables.

 

 

 

The 2600K is soldered. The temperature difference you noticed is due to the bigger size of the 3930K die ( which is a 12 core cut down to 6 ) that makes heat spread better.

[Opcode]

 

The die size has nothing to do with other than it could block you from soldering the die to the HSP which is the case for Intels TIM'ed CPU's. Feel free to compare a 5960x (has effectively 8 cores) to a 12 core xeon at the same clocks/voltage/corecount, you won't see any difference.

S2011 chips run a lot cooler than lga1150/lga1155 at the exact same power draw because the die is soldered and also the pre-installed backplate is making a huge difference though. When I moved from a 2600K @ 5GHz to my 3930K, turning 2 cores off and setting it to 5GHz it was literally running 20° cooler.

 

 

TDP is a cooling spec, they all define their TDP values differently because they feel different about the max safe operating temperatures which is one out of the so many variables.

 

Disabled cores act as passive heatsinks for the cores next to them. Creating a much larger contact surface area with the IHS thus better heat dissipation. This has been proven by AMD with their APU's. If it's one thing about heat dissipation that none of us can argue is that surface area is everything. Thus why we have massive coolers like the NH-D15.

[Faa]

Disabled cores act as passive heatsinks for the cores next to them. Creating a much larger contact surface area with the IHS thus better heat dissipation. This has been proven by AMD with their APU's. 

Where's the proof then? As usual you're lacking proof.

 

If it's one thing about heat dissipation that none of us can argue is that surface area is everything. Thus why we have massive coolers like the NH-D15.

 

Except they're not fully utilized unlike AIO's, heatsinks never feel as hot as a radiator. Surface area isn't everything, using a pump to circulate the liquid faster makes a world of difference.

 

 

The 2600K is soldered. The temperature difference you noticed is due to the bigger size of the 3930K die ( which is a 12 core cut down to 6 ) that makes heat spread better.

Prove me a bigger die size makes a difference, gave you an example how you would prove it. The pre-installed backplate which really makes the fit a lot tighter will make a lot more difference than a theory you can't prove. Most coolers come with crappy backplates, with S2011 pre-installed backplate it completely fixes this.

[Opcode]

Where's the proof then? As usual you're lacking proof.

You guys should really learn how to Google.

 

Except they're not fully utilized unlike AIO's, heatsinks never feel as hot as a radiator. Surface area isn't everything, using a pump to circulate the liquid faster makes a world of difference.

Doesn't matter all an AIO does is move the heat away from the chip. Which ends up in a radiator where surface area again makes the world of difference.

[Faa]

You guys should really learn how to Google.

If you're linking marketing fluff, which apparently is advertising AMD's amazing TDP as proof then you're the guy who should work on his googling skills and I've criticized you enough for making a fact out of AMD's PR fluffs. This doesn't prove anything. Show me a temperature difference between a 8 core die vs & 12 core with the same corecount-enabled/frequency/vcore/sensors on the cores running FFT's fixed at 16K.

 

 

Doesn't matter all an AIO does is move the heat away from the chip. Which ends up in a radiator where surface area again makes the world of difference.

 

Pump > heatpipes. Moves heat much quicker so yes it does matter.

[Opcode]

If you're linking marketing fluff, which apparently is advertising AMD's amazing TDP as proof then you're the guy who should work on his googling skills and I've criticized you enough for making a fact out of AMD's PR fluffs. This doesn't prove anything. Show me a temperature difference between a 8 core die vs & 12 core with the same corecount-enabled/frequency/vcore/sensors on the cores running FFT's fixed at 16K.

Don't try twisting it when you're presented with concrete evidence. You have a hex-core CPU why don't you do your own testing because so far I have the upper hand in this debate.

 

Especially when you factor in your very own findings.

When I moved from a 2600K @ 5GHz to my 3930K, turning 2 cores off and setting it to 5GHz it was literally running 20° cooler.

I imagine VCORE isn't going to be that different among the two. Same architecture, both are soldered, how would you explain your 20C difference.

 

Pump > heatpipes. Moves heat much quicker so yes it does matter.

Either way the method of moving the heat isn't the subject. The subject is heat dispersion. Maybe you can explain as to why a 240mm radiator performs better than a 120mm?

[Agosto]

 

Prove me a bigger die size makes a difference, gave you an example how you would prove it. The pre-installed backplate which really makes the fit a lot tighter will make a lot more difference than a theory you can't prove. Most coolers come with crappy backplates, with S2011 pre-installed backplate it completely fixes this.

 

 

Looks like you don't know much about thermodynamics, do you? @Opcode and I are trying to explain it why the 3930K  ran cooler but you just don't want to accept the fact you are wrong.

 

 

p.s.: I'm studying energy engineering, I think I have a decent preparation about thermodynamics

[Faa]

Don't try twisting it when you're presented with concrete evidence. You have a hex-core CPU why don't you do your own testing because so far I have the upper hand in this debate.

If AMD claims the cores act as a passive heatsink that also means they can heat up the GPU and vice versa. And we're missing the performance differences which I asked you to prove, not a freaking theory you'll never have any use of.

Right, lets bring up a test;

 

A 5820K/5930K/5960x are all physically octocores except 2 are lasercut on the 5820K/5930K -> 356mm² big; http://www.techpowerup.com/cpudb/

A 4960x (nothing is laser cut, just 6 cores on the die) is 260mm² big; http://www.techpowerup.com/cpudb/1569/core-i7-4960x.html

The die size of the 4960x is like 40% bigger and doesn't even have the FIVR and yet it's running warmer than the 5930K.

 

I imagine VCORE isn't going to be that different among the two. Same architecture, both are soldered, how would you explain your 20C difference.

Already told you, the pre-installed backplate that's allowing for a much tighter fit. I don't need you to prove that most backplates cooling manufacturers are offering are shit. I couldn't keep my 2600K below 90° in small fft's at 5GHz with 1.5Vcore pulling around 200W from the 4+4 cable while the water temp was only a tiny bit warmer than the ambient (25°). Air coming from the rad never felt hot, rather just cold but on my 3930K the heat was being dumped out was just massive.

 

Either way the method of moving the heat isn't the subject. The subject is heat dispersion. Maybe you can explain as to why a 240mm radiator performs better than a 120mm?

 

Surface area means shit if the problem is between the die & HSP or eventually even between the HSP/HSF - which is the case for Intels TIM'ed CPU's or AIO's with a crappy backplate. With lga115x CPU's (except SB) it's all about how quickly you can move heat away so a pump will be better than heatpipes, how well the contactbase of your cooler is designed, delidded or not, the backplate etc.

If you go through many nh-d15 reviews who mostly use lga115x, you basically see it being on par with 240mm AIO's or even performing better, use S2011 as a test platform a 240mm AIO will wipe the floor with a nh-d15.

The NH-D14 didn't pass the test. http://www.overclock3d.net/reviews/cases_cooling/fractal_design_kelvin_s36_review/7

 

 

 

Looks like you don't know much about thermodynamics, do you? @Opcode and I are trying to explain it why the 3930K  ran cooler but you just don't want to accept the fact you are wrong.

 

 

p.s.: I'm studying energy engineering, I think I have a decent preparation about thermodynamics

 

Oh, the guy who's studying energy engineering not being able to prove a bigger die size would make a noticeable difference (which is what I'm asking you to do). A HSP growing 0.5mm² in area isn't going to make a difference unless it got billion times bigger. We both know the die alone isnt going to be more effective than a HSP.

If I tell you I got my master in energy engineering, then you're wrong if we play your game.

What did Opcode prove anyways? Has to prove a bigger die size makes a difference, has to prove that cooling is all about surface area. Which I did in his place.

[Opcode]

-snip-

Nice long post that doesn't backup anything. You're disputing based on your own bullshit theory's ("ermagahd my block wasn't applied with 120 foot pounds").

 

Cores sitting next to other cores will act as passive heatsinks creating more surface area contact with the IHS. This is common sense and doesn't even need to be proven on a wide spread scale. With Kabini the CPU cores (Jaguar) run at a much lower power (cooler) than the GPU.

 

They are actually leveraging this concept for maintaing higher boost states. With the GPU idle it too can draw heat away from the CPU cores.

[Faa]

Cores sitting next to other cores will act as passive heatsinks creating more surface area contact with the IHS. This is common sense and doesn't even need to be proven on a wide scale. With Kabini the CPU cores (Jaguar) run at a much lower power (cooler) than the GPU.

Am I denying it? I'm saying putting more lets say disabled cores to get a bigger die size is not going to make a noticeable difference to your temperatures, as I proven above a bigger die size doesn't do anything to your temperatures.