AMD Carrizo Pre-release thread

[The Stilt]

Is the Carrizo voltage higher than Kaveri's at similar clocks?

It's just fits:

What will they do for BR to provide 4 GHz + dGPU? IVR?

The median between 7700K (3.8 / 3.4GHz) Kaveri and Athlon X4 845 (3.8 / 3.5GHz) for example are around 75mV lower in favor of Kaveri in turbo state.

The fastest Bristol Ridges for AM4 will run up to 1.5V... Based on how the FX-8800P is scaling voltage wise, I don´t expect there to be much of headroom either...

 

[ShintaiDK]

The Athlon X4 845 appears to be throttling even at stock, due TDP limit being reached It is no surprise really, since the FX-8800P already consumes 45-50W when the CPU cores are locked to the maximum speed (3400MHz).

Athlon X4 845 is rated for significantly higher voltages than the mobile Carrizo APUs, so the 65W TDP limit might be pretty maxed out even at stock.

And you would think by now they got the throttlemonster part sorted out. Yet now it even throttle at CPU loads only, worse than ever :\

It raises a legality question about the proclaimed performance and the delivery. Unless mobo makers simply add too much voltage.

 

[NTMBK]

I wonder how much of this is due to shoehorning it onto an FM2+ motherboard which wasn't designed for it. Hopefully AM4 should have better results.

 

[The Stilt]

There should be no ill-effects as long as the iGPU and iFCH are shut down, which obviously is the case (factory config). FM2+ is far more rugged platform than the FP4 it was originally designed for. FM2+ boards supporting these parts have been updated for CarrizoPI AGESA with backwards compatibility for older FM2+ parts, so it shouldn´t be a software issue either.

Based on my experience on Carrizo, I´d say the 65W TDP is just too low for the clocks. At this point Carrizo is significantly less efficient than Steamroller.

 

[USER8000]

I think people are reading way too much into one benchmark. If the CPU was throttling when all 4 threads were used,then HWBOT Prime and Wprime 1024M would also have poor scores - except it appears to not. The latter test is over six minutes long too. Cinebench R10 and R11.5 would also cause the CPU to throttle.

Flanker does have a point though - Carrizo only has 2MB of L2 cache,which is half the amount the previous generations have and less than a Core i3(if you include L3 cache too). That is not a lot for a modern quad core CPU.

 

[Yuriman]

Doing a quick ST comparison to Ivy Bridge:

My i5, at the same 3.8GHz, in Cinebench 11.5, scores 1.60 in the single threaded bench. Carrizo scores 1.11, according to AtenRa's benches. Even if it has made significant ground in IPC improvement, Intel's 4 generation old i5 is still 44% faster per clock in this specific benchmark.

Disclaimers: Meaningless comparison, you can't buy Ivy Bridge new anymore, Cinebench is not necessarily representative of overall performance, platform differences, cost, throttling, early bios, etc. etc.

However, we can gather from this that Carrizo should still be slower than the slowest locked Sandy Bridge quad... in Cinebench at least. It's downright depressing to see that AMD's 2015/2016 quad core is still not only slower per clock than Intel's 5 generation old quad core CPUs (and probably 6 generation/7 year old i5's as well), but also slower in absolute terms too, even with its clockspeed advantages. There was a time when all of my builds were AMD.

On the other hand, benches show that it's not too far behind Intel's modern chips in efficiency, even if it doesn't compete in absolute performance. Crossing my fingers for Zen.

 

[Sweepr]

Doing a quick ST comparison to Ivy Bridge:

My i5, at the same 3.8GHz, in Cinebench 11.5, scores 1.60 in the single threaded bench. Carrizo scores 1.11, according to AtenRa's benches. Even if it has made significant ground in IPC improvement, Intel's 4 generation old i5 is still 44% faster per clock in this specific benchmark.

However, we can gather from this that Carrizo should still be slower than the slowest locked Sandy Bridge quad... in Cinebench at least. It's downright depressing to see that AMD's 2015/2016 quad core is still not only slower per clock than Intel's 5 generation old quad core CPUs (and probably 6 generation/7 year old i5's as well), but also slower in absolute terms too, even with its clockspeed advantages. There was a time when all of my builds were AMD.

Agreed. Adding to this, my 2.8-3.0 GHz ULV Ivy Bridge (from 2012) scores ~1.25 in the single-threaded bench. You would expect AMD's desktop APUs from 2016 to top that, especially after all the hype about Carrizo's IPC improvement.

 

[monstercameron]

Alot of new embedded chips with odd configurations was just paper launched. Seems interesting.

http://www.anandtech.com/show/10083...mbedded-socs-with-excavator-microarchitecture

 

[Abwx]

Doing a quick ST comparison to Ivy Bridge:

My i5, at the same 3.8GHz, in Cinebench 11.5, scores 1.60 in the single threaded bench. Carrizo scores 1.11, according to AtenRa's benches. Even if it has made significant ground in IPC improvement, Intel's 4 generation old i5 is still 44% faster per clock in this specific benchmark.

Disclaimers: Meaningless comparison, you can't buy Ivy Bridge new anymore, Cinebench is not necessarily representative of overall performance, platform differences, cost, throttling, early bios, etc. etc.

However, we can gather from this that Carrizo should still be slower than the slowest locked Sandy Bridge quad... in Cinebench at least. It's downright depressing to see that AMD's 2015/2016 quad core is still not only slower per clock than Intel's 5 generation old quad core CPUs (and probably 6 generation/7 year old i5's as well), but also slower in absolute terms too, even with its clockspeed advantages. There was a time when all of my builds were AMD.

On the other hand, benches show that it's not too far behind Intel's modern chips in efficiency, even if it doesn't compete in absolute performance. Crossing my fingers for Zen.

Perhaps that the same comparison should be done with POVRay....

 

[deasd]

No review yet. Won't draw any conclusion. Still wait for unlock Excavator(if it exist).

As for throttling, I think maybe the mobo and BIOS need a hotfix.

 

[The Stilt]

I think people are reading way too much into one benchmark. If the CPU was throttling when all 4 threads were used,then HWBOT Prime and Wprime 1024M would also have poor scores - except it appears to not. The latter test is over six minutes long too. Cinebench R10 and R11.5 would also cause the CPU to throttle.

Flanker does have a point though - Carrizo only has 2MB of L2 cache,which is half the amount the previous generations have and less than a Core i3(if you include L3 cache too). That is not a lot for a modern quad core CPU.

So as soon as you plug in the die in FM2+ package the L2 becomes a huge restriction? :sneaky:

Athlon X4 845 should score at least 300 in Cinebench R15 when running fully stock (3.5GHz).

255 score means that the chip is clocking down to < 3GHz during the run.

 

[Headfoot]

I'll wait for the anandtech article before I draw conclusions. While forum tests can be good, early forum tests can be plagued by all sorts of undisclosed issues that a site like Anandtech discovers through its due diligence -- bios issues, etc. Some of the scores are unusual and I trust anandtech to discover the reason best when they have the chip in their hands.

 

[Dresdenboy]

disclaimers: Meaningless comparison, you can't buy ivy bridge new anymore, cinebench is not necessarily representative of overall performance, cb performance is compiled dependent, cb favors fast cache subsystems, cb doesn't reflect gaming nor office performance, platform differences, cost, throttling, early bios, etc. Etc.

scnr

 

[The Stilt]

Removing the automatically added CPU dispatcher (by ICL) nor spoofing the CPU vendor, model under VM makes no difference in Cinebench 11.5 or 15 Feel free to try it yourself.

Cinebench 11.5 tests mostly FP performance and it utilizes MMX, SSE and SSE2 instructions. Cinebench 15 meanwhile is purely a FP test and it utilizes MMX, SSE, SSE2 and SSE3 instructions.

Since it they are FP benchmarks it is obvious that all 15h CPUs perform poorly in them.

 

[dark zero]

Doing a quick ST comparison to Ivy Bridge:

My i5, at the same 3.8GHz, in Cinebench 11.5, scores 1.60 in the single threaded bench. Carrizo scores 1.11, according to AtenRa's benches. Even if it has made significant ground in IPC improvement, Intel's 4 generation old i5 is still 44% faster per clock in this specific benchmark.

Disclaimers: Meaningless comparison, you can't buy Ivy Bridge new anymore, Cinebench is not necessarily representative of overall performance, platform differences, cost, throttling, early bios, etc. etc.

However, we can gather from this that Carrizo should still be slower than the slowest locked Sandy Bridge quad... in Cinebench at least. It's downright depressing to see that AMD's 2015/2016 quad core is still not only slower per clock than Intel's 5 generation old quad core CPUs (and probably 6 generation/7 year old i5's as well), but also slower in absolute terms too, even with its clockspeed advantages. There was a time when all of my builds were AMD.

On the other hand, benches show that it's not too far behind Intel's modern chips in efficiency, even if it doesn't compete in absolute performance. Crossing my fingers for Zen.

Make it worse?
The FX 4350 at 4.2 GHz does 1.13 on CB 11.5 at ST...

On CB 15 does... 331.... Yeah...
The older chip from 2011 is doing better than the crappy Carrizo...

Even worse than it? The old 45 nm Phenom X4 965 BE does 345 at 4.0Ghz... Yeah.. AMD did a MASSIVE disaster with Carrizo on desktop.

The current Carrizo Athlon X4 845 does as good as an OCed Phenom X3 B50 at 3.7 Ghz..
Yeah, it has lower power consumption, but the improvement was really zero compared to Vishera and Kaveri. The lack of L3 and the massive nerf of L2 were really a big involution from them.

AMD screwed up big time with Carrizo on desktop. They tries to sell something that is on the Core2 Duo era at Core i3 prices. Hope that Bristol Ridge returns to 4MB of L2 and ditches the HDL (High Density Libraries) which was a big mistake.

 

[Yuriman]

Let's wait until we have some proper reviews before drawing conclusions, these early numbers are all over the place.

 

[DrMrLordX]

Interesting. I would not have expected throttling from the 845. Can changes in clockspeed be monitored by tools like CPU-z during the R15 benchmark runs?

And what happens to the clockspeed of the 845 doing something like Prime95?

 

[USER8000]

So as soon as you plug in the die in FM2+ package the L2 becomes a huge restriction? :sneaky:

Athlon X4 845 should score at least 300 in Cinebench R15 when running fully stock (3.5GHz).

255 score means that the chip is clocking down to < 3GHz during the run.

Right,so if a laptop with less cooling and a more limited motherboard and have all the SOC and IGP functions active,is getting close to 300 and the X4 845 isn't that can't mean its throttling,and by extension all those other benchmarks like the 6 minute Prime calculation put a heavy load on the CPU meaning it should throttle.

 

[USER8000]

Make it worse?
The FX 4350 at 4.2 GHz does 1.13 on CB 11.5 at ST...

On CB 15 does... 331.... Yeah...
The older chip from 2011 is doing better than the crappy Carrizo...

Even worse than it? The old 45 nm Phenom X4 965 BE does 345 at 4.0Ghz... Yeah.. AMD did a MASSIVE disaster with Carrizo on desktop.

The current Carrizo Athlon X4 845 does as good as an OCed Phenom X3 B50 at 3.7 Ghz..
Yeah, it has lower power consumption, but the improvement was really zero compared to Vishera and Kaveri. The lack of L3 and the massive nerf of L2 were really a big involution from them.

AMD screwed up big time with Carrizo on desktop. They tries to sell something that is on the Core2 Duo era at Core i3 prices. Hope that Bristol Ridge returns to 4MB of L2 and ditches the HDL (High Density Libraries) which was a big mistake.

So,what a Core i7 950 or Core i7 980X still is faster than a Core i3 6100 or i3 4130 in many benchmarks.

Oh wait,the Core i3 6100 or Core i3 4330 is crap. Weird logic.

This chip was leaked recently in the UK for the massive sum of £50 - or nearly half the price of a Core i3 6100,and close to the cost of Athlon II X2 and Sempron chips during the Phenom II era.

 

[frozentundra123456]

I agree, lets wait for tests from reliable bench-marking sites. Too soon to say anything, but I expect very similar results to Kaveri with maybe lower power usage. I certainly dont expect it to be close to intel in single thread performance.

AMD has actually done a fairly good job of tweaking as much as they can out of a basically flawed architecture on a lagging process node, but they need Zen desperately and they need it to be more than just a low cost alternative.

 

[Dresdenboy]

Removing the automatically added CPU dispatcher (by ICL) nor spoofing the CPU vendor, model under VM makes no difference in Cinebench 11.5 or 15 Feel free to try it yourself.

Cinebench 11.5 tests mostly FP performance and it utilizes MMX, SSE and SSE2 instructions. Cinebench 15 meanwhile is purely a FP test and it utilizes MMX, SSE, SSE2 and SSE3 instructions.

Since it they are FP benchmarks it is obvious that all 15h CPUs perform poorly in them.

I know, you made the tests.

CB or raytracers in general (POVRay etc.) usually use a way of calculation, which benefits a bit from SIMD (4x4 matrices), but the individual calcuations for a single pixel aren't that easily scalable and depend a lot on short FP instruction latencies, often in combination with mem operands. This results in the known ST performance.
Rendering a whole picture can easily be parallelized (the well known CB tiling) -> good MT scaling even while sharing the FPU.

 

[Shehriazad]

If it really runs at 1.45V it seems massively overvolted like all the other Kaveri chips....

Except it's an even more extreme case since it's only a 65W chip. MB makers really need to hotfix that stuff....but it seems like they never cared about it.

If a 860K can run at 1.19v @ 4Ghz then it absolutely nonsensical for a 845 to run at 1.45.


But I'm waiting for the 845 to be available in my region...if it throttles...I expect this to be the actual reason as to WHY it does so. The majority of FM2+ boards have voltage issues like this...Asrock even claims that the chips run at 1.45-1.5 for "stability reasons" (Asrock support mailed that back as an answer after I asked them what was up with the massive overvoltage)...yet I cannot possibly understand how this makes sense.

 

[dark zero]

So,what a Core i7 950 or Core i7 980X still is faster than a Core i3 6100 or i3 4130 in many benchmarks.

Oh wait,the Core i3 6100 or Core i3 4330 is crap. Weird logic.

This chip was leaked recently in the UK for the massive sum of £50 - or nearly half the price of a Core i3 6100,and close to the cost of Athlon II X2 and Sempron chips during the Phenom II era.

You can buy an Athlon 860K at £65 and is FAR better than that.

Also you are seeing something wrong. The i7 950 and 980X has 8 Threads and 4 REAL Cores compared to the i3 6100 and i3 4130 who trade blows with the i5 of the 2nd and the 1st generation respectively.

Sorry, but failed scam like that doesn't deserve to be sold.
Even VIA X1 is far more stable than that chip, LOCKED and at supposed higher wattage.

 

[monstercameron]

1 month later...
http://www.amazon.com/HP-Touchscree...8&qid=1459272513&sr=8-1&keywords=amd+fx-8800p

15.6" tablet with...1366*768 display.

 

[NTMBK]

1 month later...

15.6" tablet with...1366*768 display.

Don't forget the weird mismatched 6GB RAM configuration, I'm sure that will really help graphics performance