AMD Carrizo Pre-release thread

[maarten12100]

D: KILL IT WITH FIRE

I really feel for anyone who ends up with a PC like that in 2015. Slower than a Pentium 4...

I don't know that thing is basically half a Beema. If it boosts both cores to 2.1GHz that is an easy multi score of 1-1.2 in CB right there and that is enough for regular usage. Might even play old games like fallout on that. I mean for the price it isn't all bad.

 

[AtenRa]

I don't know that thing is basically half a Beema. If it boosts both cores to 2.1GHz that is an easy multi score of 1-1.2 in CB right there and that is enough for regular usage. Might even play old games like fallout on that. I mean for the price it isn't all bad.

No turbo, its only 1.35GHz Dual Core 10W TDP, this is not supposed to be in Desktop PCs, it was meant for 2in1 or entry laptops.

 

[cbn]

Kaveri is TDP constrained.

Have in mind that in kaveri CPU and GPU fight for each W.

Bollocks. That is just not true. Something in Windows is pegging the chip to its p5 state under iGPU load, period (probably a combination of ACPI drivers and microcode in the CPU itself, but still, I haven't been able to get iGPU-induced throttling to the p5 state to work under Linux. The Linux ACPI driver won't allow it/won't do it, even without me using cpufrequtils).

All the noise about Kaveri being TDP or cTDP constrained stem from people trying to push past 4.5 ghz on Gigabyte and/or ASRock boards, the p5 state throttling, and some people drawing erroneous conclusions.

Here is the bottom line: Under Windows, set your p5 state to be where you want it to be and the CPU will not throttle under heavy iGPU load, even if the CPU is also under load. I can push my 7700k to 4.7 ghz and 1028 mhz iGPU, load up y-cruncher or Prime95 or whatever and run Furmark in the background and not see any loss in performance. It'll run hotter, sure, and it might even crash under all that strain, but the chip isn't going to just start to slow down arbitrarily on account of a TDP constraint.

If you want to go past 4.5 ghz on Kaveri, you pretty much need an a88x-Pro or Crossblade Ranger (or maybe the A88x-Plus, but I make no guarantees.).

The major complaints I have seen stem from throttling under stock conditions.

So while a person can add better cooling and adjust p states to remove throttling, what happens when the stock HSF (rated either 65 watts or 95 watts) is used?

 

[DrMrLordX]

There are also some Kaveri chips out there with really, really, really poor TIM applications under the IHS which just beg for a delid. Perhaps that was the cause of Ian's 7850k throttling at speeds over 4.3 ghz or so.

The issue that I can see there is mostly related to cooling, namely the infamous thermal margin, which is difficult to track since most temperature monitoring programs in common use aren't specifically set up to track that (though it can be done).

Now, if you're running @ stock, there's no reason why that stock HSF shouldn't work. I'm pretty sure the software trick will render the CPU quite capable of operating properly under heavy iGPU load. Most people don't know about the trick or otherwise don't use it, and personally I have yet to see anyone with a Kaveri @ stock that could not make it work.

Are you sure you've seen people with stock configs run amdmsrtweaker and have the CPU throttle due to thermal margin? Most people just watch it throttle to 3 ghz (or whatever other speed depending on the exact CPU) and assume it's a cooling or TDP issue, which is an inaccurate assumption. They don't even try to fix the problem using a technique that's been known about since . . . January 2014?

I'll bet you that a stock 7850k with a proper p5 state and a motherboard that doesn't resort to insane overvolting (also a common problem, Kaveri doesn't take 1.3v @ stock! Even for turbo!) won't throttle to 3 ghz with Furmark running in the background. It'll stay under 80c thermal margin and will be all hunky dory.

 

[VirtualLarry]

Something like this?

http://www.newegg.com/Product/Produc...82E16883266592

We already have a thread about that 1.35Ghz AMD dual-core desktop.
http://forums.anandtech.com/showthread.php?t=2416273

No need to pollute the current thread.

 

[Abwx]

AMD Bristol Ridge to bring excavator on desktop PCs

http://www.computerbase.de/2015-01/amd-bristol-ridge-soll-excavator-auf-desktop-pcs-bringen/

 

[III-V]

Yeah, in Q3 2016, according to the article. Might as well not come at all. Perhaps it was a typo and they meant 2015.

 

[frozentundra123456]

Yeah, in Q3 2016, according to the article. Might as well not come at all. Perhaps it was a typo and they meant 2015.

Still on 28nm in third quarter of 2016. Seriously, why even bother? This doesnt make sense anyway. Why would they bring this out at the same time as supposedly Zen is coming out?

 

[Spawne32]

Still on 28nm in third quarter of 2016. Seriously, why even bother? This doesnt make sense anyway. Why would they bring this out at the same time as supposedly Zen is coming out?

Yeh doesn't make any sense at all, only thing I can figure from it is that this is the last death throws of AMD as it winds down into oblivion. The only thing we can really be hopeful for is that someone like samsung comes in and buys them out within the next 3 years in an effort to bring back competition against intel. By the end of 2016 we will probably be looking at 12nm parts from intel and they will be over 5 years ahead in technological advancement.

 

[Spawne32]

D: KILL IT WITH FIRE

I really feel for anyone who ends up with a PC like that in 2015. Slower than a Pentium 4...

Ugh, tell me about it. I won't even work on E1 based laptops or pc's when they come in for repair. I did one for a friend as a favor, its was an E1-2100 processor. Literally took 3 continuous days for a windows re-install and update. When people ask me to repair them I tell them to throw them in the trash and buy something intel based, they arent worth investing the money in repairing whatever defective part (typically hard drives) has gone bad.

 

[AtenRa]

ehm, 28nm excavator in Q3 2016 is a no go. Perhaps its a typo and they meant Q3 2015.
And now that im thinking about it, a Q3 2015 Excavator Desktop part goes hand in hand with Q3 2016 ZEN release.

 

[Spawne32]

ehm, 28nm excavator in Q3 2016 is a no go. Perhaps its a typo and they meant Q3 2015.
And now that im thinking about it, a Q3 2015 Excavator Desktop part goes hand in hand with Q3 2016 ZEN release.

Didn't AMD make specifically clear that excavator was only going to be in a OEM bga form for desktop parts?

 

[AtenRa]

Didn't AMD make specifically clear that excavator was only going to be in a OEM bga form for desktop parts?

They only said they would be OEM bga desktop products.

edit: im not saying there will be socketed Carrizo Desktop SKUs either.

 

[Shehriazad]

Meh...I'll be happy if there would even be cherry picked Kaveri chips during 2015. Like some 4.5 Ghz Athlon and a 1000+Mhz A10(gpu) for FM2+.

And Carrizo...well...if their laptops happen to be able to play casual games like League of Legends or Warframe maxed out on an ultra light notebook/laptop... and thanks to it being a SoC have loooong battery life while playing...I'll seriously think about getting one.

 

[DrMrLordX]

Didn't AMD make specifically clear that excavator was only going to be in a OEM bga form for desktop parts?

If I recall correctly, what they said was that Carrizo was going to be BGA-only. Bristol Ridge != Carrizo.

If I had to guess, I'd say Bristol Ridge is Q3 2015 release, 28nm planar (no HDL libraries), and no on-die FCH/PCH.

 

[redzo]

Ugh, tell me about it. I won't even work on E1 based laptops or pc's when they come in for repair. I did one for a friend as a favor, its was an E1-2100 processor. Literally took 3 continuous days for a windows re-install and update. When people ask me to repair them I tell them to throw them in the trash and buy something intel based, they arent worth investing the money in repairing whatever defective part (typically hard drives) has gone bad.

It is not a fair comparison. E1 is a low power low performance netbook class core. It competes with atom, which is also a sh*ty option from a performance perspective. In fact, atom's integrated gpus are even trashier. Don't miss the fact that on cheap platforms like those, being AMD or intel, everything is trashy from a performance perspective. By recommending them to switch to intel you do them no good.

 

[podspi]

Q3 2016 isn't a typo, since they also state 15 months. I'd say it's just another (probably untrue) rumor. Hard to imagine anybody being interested unless they can still them very very cheap. And who wants to sell a large die 8-core for peanuts?

 

[Spawne32]

It is not a fair comparison. E1 is a low power low performance netbook class core. It competes with atom, which is also a sh*ty option from a performance perspective. In fact, atom's integrated gpus are even trashier. Don't miss the fact that on cheap platforms like those, being AMD or intel, everything is trashy from a performance perspective. By recommending them to switch to intel you do them no good.

You are aware that Atom CPU's are faster then E1 CPU's by a significant margin, right? Even a 45nm Atom from 08-10 puts down better single threaded performance. I have one sitting right next to me. lol

 

[monstercameron]

You are aware that Atom CPU's are faster then E1 CPU's by a significant margin, right? Even a 45nm Atom from 08-10 puts down better single threaded performance. I have one sitting right next to me. lol

not even close.
http://www.notebookcheck.net/Intel-Atom-N450-Notebook-Processor.23722.0.html [atom from 09]
vs.
http://www.notebookcheck.net/AMD-E-Series-E1-6010-Notebook-Processor.115407.0.html

atom n450 vs beema e1-6010
591 983

and beema is clocked lower.

 

[Abwx]

You are aware that Atom CPU's are faster then E1 CPU's by a significant margin, right? Even a 45nm Atom from 08-10 puts down better single threaded performance. I have one sitting right next to me. lol

A big lol, really, Kabini has 20% better IPC, roughly, in integer and 30-35% in FP than Intel s baytrails, thoses latters largely trails in IPC, as the name suggest so i let you imagine what are worth the old Atoms, to the point that i wonder if you re not voluntarly trying to mislead the public.

Edit : even old bobcat has significantly better IPC than Baytrails...

 

[elemein]

A big lol, really, Kabini has 20% better IPC, roughly, in integer and 30-35% in FP than Intel s baytrails, thoses latters largely trails in IPC, as the name suggest so i let you imagine what are worth the old Atoms, to the point that i wonder if you re not voluntarly trying to mislead the public.

Edit : even old bobcat has significantly better IPC than Baytrails...

I dont think that's right. I'd have to check the Agners guide again, but IIRC, they have very similar integer throughput with Jag having much better SIMD due to more SIMD EUs and the scheduler being OoOE (for SIMD. Silvermont is OoOE for Int but not SIMD IIRC.)

Either way, Bay Trail tends to clock much higher, hence the performance discrepancy mentioned.

 

[ShintaiDK]

not even close.
http://www.notebookcheck.net/Intel-Atom-N450-Notebook-Processor.23722.0.html [atom from 09]
vs.
http://www.notebookcheck.net/AMD-E-Series-E1-6010-Notebook-Processor.115407.0.html

atom n450 vs beema e1-6010
591 983

and beema is clocked lower.

You could have used a better Atom, even if you stick to the name. And the Beema is alsmost twice the TDP.
http://ark.intel.com/products/family/29035/Intel-Atom-Processor#@All

Not that it changes much. However its not exactly impressive either unfortunately. 2 snails in a snailrace for the recycling yard.

 

[Abwx]

I dont think that's right. I'd have to check the Agners guide again, but IIRC, they have very similar integer throughput with Jag having much better SIMD due to more SIMD EUs and the scheduler being OoOE (for SIMD. Silvermont is OoOE for Int but not SIMD IIRC.)

Either way, Bay Trail tends to clock much higher, hence the performance discrepancy mentioned.

The numbers i provided are correct, as for clocks Beema tops at 2.4 and the 4.5W Mullins at 2.2 on one core, Baytrails is up to 2.66 or so but only with one core, as said even Bobcats have noticeably higher IPC than Baytrail, actualy a shrinked Bobcat based 4C would be better than Baytrail since it has currently better GPU perfs.

 

[elemein]

The numbers i provided are correct, as for clocks Beema tops at 2.4 and the 4.5W Mullins at 2.2 on one core, Baytrails is up to 2.66 or so but only with one core, as said even Bobcats have noticeably higher IPC than Baytrail, actualy a shrinked Bobcat based 4C would be better than Baytrail since it has currently better GPU perfs.

Re-checked Agners guide and the discrepancy you list would be due to the more versatile memory pipeline Jag has. That'd definitely help both SIMD and Int performance.

Apologies, been awhile since I checked

 

[monstercameron]

You could have used a better Atom, even if you stick to the name. And the Beema is alsmost twice the TDP.
http://ark.intel.com/products/family/29035/Intel-Atom-Processor#@All

Not that it changes much. However its not exactly impressive either unfortunately. 2 snails in a snailrace for the recycling yard.

yeah but I was factually correct. It matters. Also snails depend on the context.