AMD CPU speculation... and expert conjecture

[jdwii]

Just accept you were wrong juanrga and move on. Your predictions were based on 1 sentence from AMD about 30% better throughput and you took it as a 30% increase in IPC. Your predictions were way off base. You never said 10-20%. You said 20-30%.

Wait a minute juan stuck his neck out and used amd supplied numbers and slides to project performance. That is a lot more than most around here will do so give him some credit, his gpu numbers were very close. As paul and i found out in the 750k acticle what amd sends is not always indicative of what the consumer may buy We will have to wait until retail samples work there way down.

And just to see myself type it again,AMD has increased not only gpu performance with drivers but cpu performance with drivers. LLANO better with time
http://www.tomshardware.com/reviews/amd-llano-opencl,3284.html

Same with trinity and richland, kaveri will be no different. And i wll wager in 6months his numbers will be even closer.

Lets stop pointing fingers and move on. How much increase is kaveri getting from a88 over a75, My 750k a88 runs soooo much better than on a55 or a75, cooler and with less voltage.

 

[noob2222]

I must use juans math from now on. If I scored 5/10 on a test then I get a 95 instead of only being 50% correct.

I.e. "I predicted 15 and only got 8 so im only 7 off."

 

[jdwii]

So i made a quick build for 500$ i can build a AMD A10-7850K APU machine with 2133 8GB of ram and a ASUS A88XM-PLUS board. And for 400$ i can use that same ram and build a 400$ machine with a A10 6800K.
Could probably build the machine for 460$ if i get a cheaper board

 

[jdwii]

I rarly side with AnandTech but this time around however things seem to be different and i FULLY agree with this statement
"The reality is quite clear by now: AMD isn't going to solve its CPU performance issues with anything from the Bulldozer family. What we need is a replacement architecture, one that I suspect we'll get after Excavator concludes the line in 2015."
If Amd wants to pull a Intel and only make 4 cores for now on they have to deliver and a module can not do that when it was made to share resources, also the design is made to run at high clock rates something that can't be done. Now its not moar cores now its HSA and moar GPU power lets hope Amd makes a competitive CPU for gaming
Peace and happy gaming and juanrga no hard feelings everyone makes mistakes i did the same thing with bulldozer. Stay on the forum even if people disagree with you

 

[palladin9479]

So i made a quick build for 500$ i can build a AMD A10-7850K APU machine with 2133 8GB of ram and a ASUS A88XM-PLUS board. And for 400$ i can use that same ram and build a 400$ machine with a A10 6800K.
Could probably build the machine for 460$ if i get a cheaper board

The 7850K is priced at a horrible value. Use the 7600 instead at 65W (assuming it appears in another month or so). DDR3-2133 is about the same price as DDR3-1866 and the CPU will still use it, you'll just have to manually set the clock.

What would this theoretical system be used for? Any power / heat / sound requirements?

Without getting too creative with pricing I can already see a sub $500 system that runs off a 120~150W pico-PSU and is ultra quiet. Can go lower by moving to a cheap case / psu and 3.5 inch HDD.

 

[jdwii]

^ gaming machine -best bang for buck

 

[eidolon171]

It's time to put the rumors to rest. I just bought the a10 7850k on Amazon for $173.99 (waaay steeper than I'd hoped it would be). This is the hardware I already have...

ASRock FM2A88X Extreme4+

RAIDMAX Cobra RX-500AF-B 500W ATX 12V v2.3/EPS 12V SLI Ready CrossFire Ready 80 PLUS BRONZE Certified Power Supply

G.SKILL Ripjaws Z Series 8GB (2 x 4GB) 240-Pin DDR3 SDRAM DDR3 2400 CL 10 (PC3 19200)

SanDisk Extreme SSD 120 GB SATA 6.0 Gb-s2.5-Inch Solid State Drive SDSSDX-120G-G25

Gigabyte R7 260X GDDR5-1GB OC

LG 12X BD-ROM 16X DVD-ROM SATA

recycled alienware case, vantec fans, copper heatsink/liquid cooler

A total $620 build

Amazon says the APU will ship in 3-6 weeks. I have Cinebench R15, Geekbench, Passmark, and 3DMark, as well as Crysis 3, Battlefield 4, Batman Arkham Origins, Hitman Absolution, and Metro 2033. I'll be using Fraps to record and Cpuz/Gpuz, as well as a Sony Bloggie to deliver the results. The video will be posted to youtube and I will drop a link to it on this thread. I have a homemade liquid cooling solution that worked wonders overclocking a Phenom II 555 (to 4.2GHz), 955 (to 4.7Ghz), and an i5 2400 (to 4.6GHz), I will OC the 7850k and run Dualgraphics with the OC'd 260x, and then rerun the bench's and Games

I expect to get the same results as my i5 2400 with a GTX 560 installed, but that may be a little too hopeful.

 

[FALC0N]

Kaveri's fine...
The opening prices are not.

The 65W part is really cheap, $119 USD is perfect for gaming on a small box in the living room. The 95W is WAY overpriced, it should be $150 USD, $160 at most. It's going to require a bigger / louder cooling solution though.

Really don't see what people are getting worked up over, it's a fx4 chip without L3 cache and with a bolted on low ~ moderate power GPU. It's a big upgrade to the GPU side not the CPU one, which is what I was expecting honestly.

My thoughts exactly, Palladin. We knew coming in this was targeted at a mainstream audience. I it looks like it mostly hit its mark.

I was a little disappointed with the IPC numbers though. A fairly big deal was made about the additional decoders solving an issue with core bottlenecking. I thought 20% IPC improvement was quite a realistic expectation. It ended up being about the same level of gain as piledriver. Not bad, but not great either.

 

[palladin9479]

I was a little disappointed with the IPC numbers though. A fairly big deal was made about the additional decoders solving an issue with core bottlenecking. I thought 20% IPC improvement was quite a realistic expectation. It ended up being about the same level of gain as piledriver. Not bad, but not great either.

Most of that internal work was for multi-thread IPC, when more then one thread is scheduled per module. It won't do much for single threaded performance as that's primary bottle-necked by ALU and cache prefetch accuracy / latency.

 

[Master-flaw]

Kaveri's fine...
The opening prices are not.

The 65W part is really cheap, $119 USD is perfect for gaming on a small box in the living room. The 95W is WAY overpriced, it should be $150 USD, $160 at most. It's going to require a bigger / louder cooling solution though.

Really don't see what people are getting worked up over, it's a fx4 chip without L3 cache and with a bolted on low ~ moderate power GPU. It's a big upgrade to the GPU side not the CPU one, which is what I was expecting honestly.

I agree..the $119 model is basically the model that's going to attract the majority...
I'm thinking the 6800K may still be my best option from first glance, though I haven't looked into the other models much.

 

[Ags1]

The AnandTech review is quite interesting. It looks like the 100W Kaveri barely holds the line compared to the previous generation in terms of CPU performance (as had been predicted by 8350rocks etc, based on bulk not favoring performance parts), while the 45W Kaveri shows large performance jumps in the region predicted by Juanrga... everyone's a winner

http://www.anandtech.com/show/7677/amd-kaveri-review-a8-7600-a10-7850k/15

With the significant boost to low power performance, AMD may be able to take a larger share of the laptop market with Kaveri. Shame the mobile parts are launching later.

(Originally posted this under the IVB thread by mistake...)

 

[juanrga]

Umm juanrga that's not her percentages work.

If you state X is 30% of Y, and it ends up being 25% of Y, then you are not off by 5%. Instead you are about 17% over estimation. Your working with percentages here, it's actual/predicted (25/30).

The absolute error of your example is 5% = 30% - 25%

I think you are referring to percent error, but the definition is not what you give and your (25/30) is nothing. The relative error is (5/30).

I must use juans math from now on. <snip>

I.e. "I predicted 15 and only got 8 so im only 7 off."

I am honored, but the math is not from mine.

http://mathworld.wolfram.com/AbsoluteError.html

The absolute error in your example is 7.

Peace and happy gaming and juanrga no hard feelings everyone makes mistakes i did the same thing with bulldozer. Stay on the forum even if people disagree with you

The problem is not disagreement, but some (few) people here lying and making personal attacks one day and the other as well. E.g., I like as one poster takes my prediction about x264 using one concrete benchmark but the compares it to another completely different x264 on hardwarecacnuck, and mixes numbers from both, whereas ignores my prediction about what hardwarecacnuck would measure.

 

[juanrga]

Review of Kaveri is under development. Preliminary results are next. First what I predicted and then the measured results

Measured value for Kaveri: 94.45

Measured value for Kaveri: 3631

Measured value for Kaveri: 37.00

Measured value for Kaveri: 958.10

Work in progress...

http://openbenchmarking.org/result/1401145-PL-AMDA1078505

Note that my original predictions assumed a 4GHz Kaveri. If I correct the predicted value for x264 for 3.7Ghz I obtain (102x3.7/4) 94.35, which is very close to the measured value of 94.45. In other cases the error is bigger.

 

[gamerk316]

It seems like the iGPU on the APUs just can't reach its full potential. Slow memory is the issue perhaps?

To add insult to injury, the 3GHz Haswell Pentium G3220 is $69 on newegg right now. It's effectively a slightly slower clocked i3-4130 without hyperthreading. Add a $100 Radeon card (like 7730), or better the 7750 for $120, and I think it will have better frame rate than the $189 Kaveri A10-7850K APU.

But on the bright side, the A10-6800k dropped to $130-140 online, so same price as i3-4150, but the A10-6800K is slightly better APU. You get two extra integer cores as well as Richland iGPU, which is still better than HD 4600 by a noticeable margin.

A very good point, actually. Would be interested to see how the two platforms compared...

 

[gamerk316]

I was a little disappointed with the IPC numbers though. A fairly big deal was made about the additional decoders solving an issue with core bottlenecking. I thought 20% IPC improvement was quite a realistic expectation. It ended up being about the same level of gain as piledriver. Not bad, but not great either.

Most of that internal work was for multi-thread IPC, when more then one thread is scheduled per module. It won't do much for single threaded performance as that's primary bottle-necked by ALU and cache prefetch accuracy / latency.

Which was the point I was making at the end of last week: Optimizing something that won't always be used won't give a flat increase to performance. Sure, fixing the decoders will increase threaded throughput, but if only half the cores get loaded, and only get loaded to 20%, how much are you really gaining? The answer, as we now see, is not much.

Hence why I've always favored single core IPC, then scaling upward to more cores, then adding more cores, then trying to fix IPC. Intel's in a far better position if they decide to start throwing on more cores. They don't, mainly because they don't need too. Yet, at least.

 

[logainofhades]

Well, I was considering adopting the APU for my itx/OC the crap out it build. Looks like I will have to wait until prices come down and updated bios out of the box arrive. Now if only my microcenter can stock some ITX mobos.

In the meantime. Team Vulcan 2133 RAM. Anybody know of them? Quality?

Overclock the crap out of it all you want, it still won't match the cpu performance of an i3. You will be able to match a PhII at 4.0 or so at best. You may as well go with a 760k and HD 7770 and have better overall performance for nearly the same cost.

 

[etayorius]

I think Juanrga did a great job with all his work on his blog, the only ones to blame was GloFo and possibly AMD for being to stubborn to stick with them.

Juanrga numbers where rather high in the end, but his work done on comparisons and estimations were good enough, at the end of the day it was not juanrga estimated numbers, but the numbers provided by AMD as claim that Kaveri will have X% of improvements where just wrong.

 

[pangolin_user]

Kaveri's fine...
The opening prices are not.

The 65W part is really cheap, $119 USD is perfect for gaming on a small box in the living room. The 95W is WAY overpriced, it should be $150 USD, $160 at most. It's going to require a bigger / louder cooling solution though.

Really don't see what people are getting worked up over, it's a fx4 chip without L3 cache and with a bolted on low ~ moderate power GPU. It's a big upgrade to the GPU side not the CPU one, which is what I was expecting honestly.

is it posible the high asking price for 7850k and 7700 k because BF4 bundle? then will the price drop to $150 when the bundle expired?

 

[$hawn]

So how come Toms hasn't put out an article benchmarking the Kaveri chips yet?

 

[de5_Roy]

bits and pieces that seemed interesting from different kaveri reviews:

When we asked AMD's CTO Joe Macri if a better solution would have been six CPU cores and fewer GPU cores, the answer was a resolute 'no'. Macri said that very little software is able to take advantage of more than four cores. The software that does work well with multiple cores, such as photo and video editing programs, works even better with GPU cores, since algorithms can run even faster and more efficient. Macri does have a good point, but as long as all multi-threaded software hasn't been optimised to efficiently use GPUs, AMD will keep lagging behind Intel.

http://uk.hardware.info/reviews/5156/3/amd-a10-7850k-kaveri-review-amds-new-apu-steamroller-cpu-10-faster

Let's make it clear: AMD is making a play into the low power space rather than attempting to pit the A10-7850K against the likes of Intel's Core i5 Haswell processor lineup and this is likely because it falters in a lot of areas. It was spelled out to me that, with the move to 28nm on GlobalFoundries, there were some production differences that made the higher power and higher voltage parts less flexible (and less scalable) than expected. But, in contrast, the lower end parts like the A8-7600 shown here today can use less power and hit higher performance levels than was previously possible.

http://www.pcper.com/reviews/Processors/AMD-A8-7600-Kaveri-APU-Review-HSA-Arrives/System-Setup-and-Review-Goals

a8 7600's memory bandwidth looks so... bad..
http://www.pcper.com/reviews/Processors/AMD-A8-7600-Kaveri-APU-Review-HSA-Arrives/SiSoft-Sandra
it's amazing that the apu performs so well despite that. it's a damn shame that it's constricted this way (<- after seeing how much haswell core i3 gets). injustice! oh, don't use ram slower than ddr3 1600 if you wanna game with the igpu.

a10 7850k o.c. with noctua nh-u14s. although i am not sure if the 28nm apu should be fed 1.440 vcore for 4.7ghz and 1.2v for 1ghz igpu. won't it degrade the a/pu?
http://www.hardcoreware.net/kaveri-review-a10-7850k/16/

will update this as i read moar reviewz. for now, the little(!) a8 7600 looks like the standout.

edit:
A technical look at AMD’s Kaveri architecture
Step 3 of 4 on the road to HSA brings Steamroller and Hawaii
http://semiaccurate.com/2014/01/15/technical-look-amds-kaveri-architecture/

 

[gamerk316]

When we asked AMD's CTO Joe Macri if a better solution would have been six CPU cores and fewer GPU cores, the answer was a resolute 'no'. Macri said that very little software is able to take advantage of more than four cores. The software that does work well with multiple cores, such as photo and video editing programs, works even better with GPU cores, since algorithms can run even faster and more efficient. Macri does have a good point, but as long as all multi-threaded software hasn't been optimised to efficiently use GPUs, AMD will keep lagging behind Intel.

Now, if only there were someone on this forum to argue this same exact point for the past four years...

...oh wait, that's me. Glad AMD finally came around.

Seriously, I've been arguing this same exact point FOR FOUR YEARS, been constantly attacked for doing so, and now the AMD CTO agrees with me. VINDICATION!

Also explains why no PD refresh; AMD views it as a dead-end arch.

 

[gamerk316]

You see the trend compared against Richland: ~27% in gaming, mostly due to the GPU. ~135% in compute, again, due to the GPU.

Less then 8% in multicore CPU performance, and the same single-core performance [about 60% IB].

So matches what everyone else is seeing.

 

[logainofhades]

Yea, it is pretty sad on the CPU side of things. I was hoping for something more worthwhile than this. Had they put the 7850k @ $125 or so, it would have been a great deal. I was hoping for a low cost, low power solution for my 4th WoW rig, but I just don't think the performance is there for the price they are wanting. While my PhII and HD 4870 will use more power, its performance will be better for my needs. The power savings do not justify the cost for me.

 

[de5_Roy]

Now, if only there were someone on this forum to argue this same exact point for the past four years...

...oh wait, that's me. Glad AMD finally came around.

Seriously, I've been arguing this same exact point FOR FOUR YEARS, been constantly attacked for doing so, and now the AMD CTO agrees with me. VINDICATION!

Also explains why no PD refresh; AMD views it as a dead-end arch.

tut tut, gamerk316. through red-colored reality distortion field/glasses your post reads as: "Now, if only there were someone on this forum to campaign for intel/dual core cpus for the past four years..."
"...oh wait, that's me."
i am pretty sure someone somewhere out there on the internet is prepping a writeup in reply to your post. err.. i'm not refering to the thg forum user "someone somewhere". darn forum ids. >_>

 

[-Fran-]

When we asked AMD's CTO Joe Macri if a better solution would have been six CPU cores and fewer GPU cores, the answer was a resolute 'no'. Macri said that very little software is able to take advantage of more than four cores. The software that does work well with multiple cores, such as photo and video editing programs, works even better with GPU cores, since algorithms can run even faster and more efficient. Macri does have a good point, but as long as all multi-threaded software hasn't been optimised to efficiently use GPUs, AMD will keep lagging behind Intel.

Now, if only there were someone on this forum to argue this same exact point for the past four years...

...oh wait, that's me. Glad AMD finally came around.

Seriously, I've been arguing this same exact point FOR FOUR YEARS, been constantly attacked for doing so, and now the AMD CTO agrees with me. VINDICATION!

Also explains why no PD refresh; AMD views it as a dead-end arch.

It's kind of a middle point, I'd say. You have always stated that "moar cores" makes no sense because scaling in software is almost non existant (games was your main argument here). What the CTO says is "the software that can be fully parallel will be better if we use the extra space in GPU for that instead of regular CPU cores". That's why AMD says Kaveri (7850K) has 12 general purpose cores: 4 CPU + 8 GPU. They're not backing out on "moar cores", they just shifted the effort from regular CPU cores to GPU "cores".

So, a second read would be "Kaveri is our first 12 core APU". Hardly a back up for your statements

Cheers!