What is going to happen to AMD?

[coercitiv]

Why do they use Prime95 + Furmark as a stress test? Of course it's going to throttle. They're trying to murder the poor chip. Even my desktop Kaveri tried defaulting to p4 clockspeed in horror when I pulled that stunt.

It throttles under Cinbench too. It should score a bit over 2 in Cinebench 11.5, only does 1.8. But it's a strange throttle, since as we saw from monstercameron's example, the chip stayed at 2.1Ghz for most of the test.

 

[AtenRa]

http://www.notebookcheck.net/HP-EliteBook-745-G2-Notebook-Review.125934.0.html

Why do we care about throttling with prime + Furmark ??

Lets see something more interesting for many people,


Kaveri A10-7300 1.9GHz/3.2GHz 19W TDP x264 benchmark 4.0
first pass = 62,1fps
second pass = 11,9fps

Core i3 4005U 1,7GHz 15W TDP x264 benchmark 4.0
first pass = 57,3fps
second pass = 10,9fps

Core i3 4030U 1.9GHz 15W TDP x264 benchmark 4.0
first pass = 65fps
second pass = 12,1fps


Core i3 5005U 2.0GHz 15W TDP x264 benchmark 4.0
first pass = 69,7fps
second pass = 13fps


A 2GHz Carrizo at 15W TDP will be very close to Broadwell Core i3 in x264.

 

[ShintaiDK]

Why do we care about throttling with prime + Furmark ??

Lets see something more interesting for many people,


Kaveri A10-7300 1.9GHz/3.2GHz 19W TDP x264 benchmark 4.0
first pass = 62,1fps
second pass = 11,9fps

Core i3 4005U 1,7GHz 15W TDP x264 benchmark 4.0
first pass = 57,3fps
second pass = 10,9fps

Core i3 4030U 1.9GHz 15W TDP x264 benchmark 4.0
first pass = 65fps
second pass = 12,1fps


Core i3 5005U 2.0GHz 15W TDP x264 benchmark 4.0
first pass = 69,7fps
second pass = 13fps


A 2GHz Carrizo at 15W TDP will be very close to Broadwell Core i3 in x264.

Why are you comparing turbo with non turbo speeds and then make up some false conclusion based on that? Not to mention MT scores only. Why dont you post some ST scores as well.

 

[krumme]

The best I can do is quote these. Listen to this man.

Yes that is the best you can do.

Where do you think amd got their cash the last years when it was needed?
That pretty much answers where it will come from next time.

 

[ShintaiDK]

Yes that is the best you can do.

Where do you think amd got their cash the last years when it was needed?
That pretty much answers where it will come from next time.

You cant increase your debt forever when your revenue is collapsing.

And those interest rates AMD already borrow to is quite high because they are rated junk. Its about the same rates as average joe can borrow to without any security.

 

[AtenRa]

Why are you comparing turbo with non turbo speeds and then make up some false conclusion based on that? Not to mention MT scores only. Why dont you post some ST scores as well.

You asked about throttling, i gave you integer MT load that is not throttling. ST performance is not necessary to investigate throttling.

 

[ShintaiDK]

You asked about throttling, i gave you integer MT load that is not throttling. ST performance is not necessary to investigate throttling.

Really, so how did you determine if its throttles or not? Because you didnt have any clock and you didnt have any ST performance to compare with.

4 Kaveri cores at 19W with 1.9/3.2Ghz more or less performs as 2 Haswell cores at 15W in MT locked at 1.9Ghz arcording to your own numbers.

 

[AtenRa]

Really, so how did you determine if its throttles or not? Because you didnt have any clock and you didnt have any ST performance to compare with.

4 Kaveri cores at 19W with 1.9/3.2Ghz more or less performs as 2 Haswell cores at 15W in MT locked at 1.9Ghz arcording to your own numbers.

2 Haswell cores (no turbo) at 1.9GHz is faster than four kaveri cores at 1.9GHz (no turbo). The fact that the Kaveri at 1.9GHz is equal to Haswell at 1.9GHz (no turbo) means the Kaveri can turbo higher than 1.9GHz thus no throttling.

 

[ShintaiDK]

2 Haswell cores (no turbo) at 1.9GHz is faster than four kaveri cores at 1.9GHz (no turbo). The fact that the Kaveri at 1.9GHz is equal to Haswell at 1.9GHz (no turbo) means the Kaveri can turbo higher than 1.9GHz thus no throttling.

Where did you see no turbo for Kaveri? And what about the throttle? You havent provided anything at all to backup your statements.

 

[coercitiv]

Where did you see no turbo for Kaveri? And what about the throttle? You havent provided anything at all to backup your statements.

His inference was straightforward: since Kaveri obtained equal scores to an Haswell i3 @ 1.9Ghz, and we know HW has higher IPC than Steamroller, then the only logical explanation would be that Kaveri sustained clocks higher than 1.9Ghz throughout those tests.

 

[Paul98]

They don't need ground breaking, they just need responsibly competitive up to date products.

 

[DrMrLordX]

It throttles under Cinbench too. It should score a bit over 2 in Cinebench 11.5, only does 1.8. But it's a strange throttle, since as we saw from monstercameron's example, the chip stayed at 2.1Ghz for most of the test.

Right, but we are not going to learn anything about inexplicable throttling behavior during Cinebench 11.5 from a Prime95 + Furmark beatdown. Nor will we learn anything meaningful about any other oddball behavior that mobile Kaveri chips might exhibit.

 

[Yuriman]

AtenRa, out of curiosity, how would you rate the (average) per-core performance of Kaveri vs Haswell right now, in integer and in FPU loads? Roughly how far ahead is Intel in those metrics? In power-constrained vs non-constrained loads?

I want to say that, taking HT out of the picture, each Haswell core is a minimum of 45% faster in integer loads and probably a lot more in FPU loads.

Also, I wonder if AMD's SMT implementation will be as good as Intel's? It's possible that AMD will close the gap in single-threaded performance (mostly) only to lose in multithreaded loads in their first-generation Zen chips due to a somewhat immature SMT design.

 

[imaheadcase]

They have 800 million in cash on hand... In other words, they aren't going bankrupt.

Cash on hand does not mean what you think it means for a company. Cash on hand is to pay creditors, not a profit.

 

[krumme]

Cash on hand does not mean what you think it means for a company. Cash on hand is to pay creditors, not a profit.

Amd havnt turned a profit the last 30 years except a few years. Yet we have this idiotic thread each week.
Have you wondered - perhaps you dont know what profit means for a company?

 

[krumme]

Its hillarious. Some newbee comes to at forum ones a week and ask when amd is going belly up. Then the usual 4 financial geniuses says is probably within a year or two. All said with great joy. The same as for some its great to pronounce Intel lost 4b on mobile. Wtf is up with you? Do you lack self confidence?

 

[positivedoppler]

Step 1 wait for AMD to go belly up
Step 2 Wait for Intel to beat everyone else in cpu, server, gpu, mobile, and process technology
Step 3 Make it rain with all the money made by buying Intel and shorting AMD
Step 4 Get bottle of lotion
Step 5 turn off lights
Step 6 After completing all prior steps, shrink from 32 to 22 to 14 and finally to 10 nm

 

[AtenRa]

AtenRa, out of curiosity, how would you rate the (average) per-core performance of Kaveri vs Haswell right now, in integer and in FPU loads? Roughly how far ahead is Intel in those metrics? In power-constrained vs non-constrained loads?

I want to say that, taking HT out of the picture, each Haswell core is a minimum of 45% faster in integer loads and probably a lot more in FPU loads.

From my testing(x264), MT Integer throughput per module of Kaveri at 55W TDP is almost equal to Haswell Core throughput (Core + HT) at 55W TDP.

Same can be said for the FP MT throughput (Pov-Ray), 55W TDP Kaveri is very close to Haswell.

At 15W TDP Haswell has a lead because of the mArchitecture difference but mostly due to the 22nm FF vs 28nm Planar. It will be very interesting to see how Carrizo (28nm HDL) will do against 22nm FF Haswell and vs 14nm FF Broadwell.
I will say that Carrizo Integer/FP throughput per module will be extremely close if not faster at 15W TDP vs Haswell Core (+HT).

Also, I wonder if AMD's SMT implementation will be as good as Intel's? It's possible that AMD will close the gap in single-threaded performance (mostly) only to lose in multithreaded loads in their first-generation Zen chips due to a somewhat immature SMT design.

Well its not the same but, Excavator will have a fourth Generation SMT implementation of FP unit per Module. So AMD already has some expertise in SMT technology and I believe they could reach 20-30% from it (depending on the application) with ZEN.

 

[krumme]

Then at last you deserve a nap
...then rinse and repeat

 

[nismotigerwvu]

Its hillarious. Some newbee comes to at forum ones a week and ask when amd is going belly up. Then the usual 4 financial geniuses says is probably within a year or two. All said with great joy. The same as for some its great to pronounce Intel lost 4b on mobile. Wtf is up with you? Do you lack self confidence?

That and the whole AMD death rumors have literally been going around since the early 90's. Today's Bulldozer is yesterday's K5. When you're constantly being compared to Intel, life is going to seem bleak for just about everyone outside of maybe Apple. Beyond that, competition is a good thing for consumers. It is in out best interests for Intel to succeed in mobile as that pushes Qualcomm, Samsung and the like to push the performance/price balance. The same is true for AMD in x86. It's simply econ 101.

 

[coercitiv]

Right, but we are not going to learn anything about inexplicable throttling behavior during Cinebench 11.5 from a Prime95 + Furmark beatdown. Nor will we learn anything meaningful about any other oddball behavior that mobile Kaveri chips might exhibit.

I completely ignored the Prime95+Frumark part of this conversation. From my point of view it was just taunting.

If there's anything to learn from this oddball behavior, is to know what to check first when Carrizo becomes available.

 

[krumme]

That and the whole AMD death rumors have literally been going around since the early 90's. Today's Bulldozer is yesterday's K5. When you're constantly being compared to Intel, life is going to seem bleak for just about everyone outside of maybe Apple. Beyond that, competition is a good thing for consumers. It is in out best interests for Intel to succeed in mobile as that pushes Qualcomm, Samsung and the like to push the performance/price balance. The same is true for AMD in x86. It's simply econ 101.

The expectations to amd is not proportional to their size. Perhaps its because they oftens brings highly innovative tech to the market be it dx12/mantle/vulcan or hsa, hbm whatever. Technologically they behave like they are qcom size and amd brings much good to us the last 10-15 years (prior to k7 only minor). Its crazy how they shape the future.

But for the shareholders it have been a letdown. Its just a gamble. And right now because of mubadala no potential upside imo.

 

[DrMrLordX]

I completely ignored the Prime95+Frumark part of this conversation. From my point of view it was just taunting.

Taunting or not, notebookcheck.net decided that such a test was a good measure of maximum heat/power consumption, which is . . . different, to say the least.

If there's anything to learn from this oddball behavior, is to know what to check first when Carrizo becomes available.

Agreed. Kaveri is nearing the end of its limited run in the notebook market. From this point forward, it is a desktop processor.

 

[Enigmoid]

NBC always uses Prime + Furmark to also test the power brick as well as the maximum load the platform can handle (its a good test for down the road when the cooling system gets gummed up and loads that were not previously a problem start throttling the machine). Sometimes you have laptops reaching the limits of the power supply. Case in point on the surface 3 (cherry trail) the tablet shut off under such a load, possibly because the power brick was overworked.

 

[DrMrLordX]

I guess if you want to torture the power brick, hey, go ahead. That's not going to happen from dust accumulation, though. You will get inferior thermals, but you won't get a situation where the power brick, socket, or anything else are being pushed beyond their power delivery limits.