AMD - Piledriver chitchat

[CAT-THE-FIFTH]

I saw this on SA forums:

http://translate.google.cz/translate...14_635132.html

http://forwardthinking.pcmag.com/chi...ighlight-isscc

[watercooled]

It looks like desktop Kabini might carry Athlon and Sempron brand names: http://www.cpu-world.com/news_2014/2..._AM1_APUs.html

[CAT-THE-FIFTH]

I saw this on SA forums:

http://www.linkedin.com/in/saketdoshi

It seems AMD Carrizo is still on the GF 28NM bulk process,but is meant to be an SOC too.

I do wonder whether the power consumption improvements are coming from chipset improvements,like with Haswell.

There is mention of 20NM SOCs too.

[watercooled]

Makes sense IMHO - it would be unusual to use a node for only one product line, unless GF 20nm were to arrive very early with good yields etc.

GF are apparently the fab making a lot of these ASIC Bitcoin miners. If the business is big enough, they could potentially offset some of the early-adopter cost for 20nm.
https://www.semiwiki.com/forum/conte...cture-you.html

Also, I wonder what's happening after Excavator? AMD released details 4 generations into the future, albeit only codenames and some very basic goals, but now we're approaching the end of that timeline.

[DanceswithUnix]

Apparently the 7850K on an A88X-PRO motherboard can have the TDP turned right down to 45W with lots of steps in between.

http://www.phoronix.com/scan.php?pag...veri_tdp&num=1

[CAT-THE-FIFTH]

I was just reading the GTX750TI review from TH. Wow,they have taken a downturn even by their standards:

In absolute terms, the GeForce GTX 750 Ti is just about as fast as Radeon R7 265. But again, that’s a 150 W GPU. The new GeForce uses just 40% of its power.

It does make me wonder at times,why such rubbish is written.

[DanceswithUnix]

It does make me wonder at times,why such rubbish is written.

Because they have the time and resource to write benchmarks, and we don't.

OEMs are going to love that though. Just think of the tiny PSUs they can put in!

KHashes per watt might be a reasonable metric for places where leccy is cheaper than here in the UK.

[CAT-THE-FIFTH]

Because they have the time and resource to write benchmarks, and we don't.

I don't understand what you mean??

Looking at their own review and indeed many others,the GTX750TI is not practically R7 265 level performance. Plus 40% of the power?? What type of nonsense is that?? The R7 270 is faster in all metrics then a GTX750TI,sometimes as much as 30% to 40% faster in both latency and framerates.

The R7 270 is consuming around 120W and the GTX750TI around 60W. The GTX750TI is using half the power of a R9 270 while being upto 30% to 35% slower from my quick estimates in more intensive games.

Now looking at the fact the HD7850 has been out for two years,we know very well it consumes less power than an HD7870,and yet TH magically has no HD7850 or R7 265 power numbers. Plus the HD7850 and R7 265 numbers are weird for latency measurements. Both are the same cards and should be close together - yet we see massive variation for all of a 65MHZ clockboost,and the R7 265 is still a faster card.

If the GTX750TI by their own numbers is consuming 50% of the power of a much faster R9 270,how can it consume 40% of the power of a lower power consumption graphics card??

What this sounds like,is the Nvidia reviewing guide in action. Nvidia knows the R7 265 will be the new competition for the GTX750TI,so lets understate the performance of the card and overstate the power consumption advantage of the GTX750TI.

My Xeon E3 1220 and GTX660 based system consumes no more than 200W at the wall running Skyrim or Tombraider.

Plus lets look at their own numbers again:

http://media.bestofmicro.com/2/W/422...70-seconds.png



Power consumption peaks at around 141W.

I suspect the cards are hitting very high clockspeeds due to the non-deterministic boost. The HD7850 does not boost and the R7 265 looks like it runs at 925MHZ continously when running games. They are less likely to throttle over extended periods.

We saw what happened with the GTX760 with throttling over extended periods - I hope the GTX750TI does not do that,but only an extended test will say so. The same happened with Geforce Titan. The HT4U GTX760 was eye opening to say the least,with the cards hitting significantly lower clockspeeds after only 15 minutes,and it was boosting clockspeeds by around 137MHZ during benchmarks. Geforce Titan showed the same problems too.

Yet,we see none for this card and AMD cards get all the extended testing to prove they are running OK,not Nvidia ones. Only European sites have down extended testing on Nvidia cards.

Don't get me wrong this is a great bus powered card.

However,the amount of "Jesus second coming" hype is just typical with any Nvidia card launch. It is as bad as Apple.

OEMs are going to love that though. Just think of the tiny PSUs they can put in!

KHashes per watt might be a reasonable metric for places where leccy is cheaper than here in the UK.

It looks a good fit for those kind of systems,but OTH it depends on how much Nvidia charges in the OEM channels.

[DanceswithUnix]

I don't understand what you mean??

Simply that their grasp of scientific methods seems poor at best. I know I could do better, I'm sure you could too. But they are the ones that have access to a lab full of kit and the time to put a review together, I have a full time job and a family. So we just get to shake our heads and stare in disbelief.

Oh the 40% source is easy. Check out the card specs: http://www.scan.co.uk/products/2gb-e...dport-dvi-hdmi
GPU max power usage is 60W. Toms said the R7 265 is 150W. 60/150=0.4 or 40%. One figure was written on the Internet, and the other sounds like an educated guess, so what could be wrong with the figure?

[watercooled]

Yeah it looks like they've used the scientific method of comparing TDPs.

When it comes to GPU-only power measurements, TPU to the rescue: http://www.techpowerup.com/reviews/N...750_Ti/23.html

The AIB cards have higher power draw so far, but the ones they've tested aren't stock clocked so it's hard to make a comparison. Still, a lot of the cards on sale will be pre-overclocked ones, which generally does hurt efficiency.

[CAT-THE-FIFTH]

I would like a better test of the longterm effects of the GPU boosting mechanisms,to see how consistent performance is. So far we have seen the GTX660TI,GTX760,Geforce Titan,R9 290 and R9 290X have issue.

[DanceswithUnix]

Reading the Anandtech article I'm actually interested that the 750 isn't that far off the 750ti despite only having 1GB of ram.

Looks ideal for a Steambox if coupled with an Athlon X4 FM2 setup. Shame I am broke this month

[DanceswithUnix]

I would like a better test of the longterm effects of the GPU boosting mechanisms,to see how consistent performance is. So far we have seen the GTX660TI,GTX760,Geforce Titan,R9 290 and R9 290X have issue.

Sadly I don't think they would send you review hardware to investigate that

[CAT-THE-FIFTH]

Sadly I don't think they would send you review hardware to investigate that

Probably not,but its quite funny that review sites have gone to all the complication about latency testing and yet something far simpler to test,is not done.

However,I still don't understand why the R7 260X consumes so much more power than an HD7790:

http://tpucdn.com/reviews/AMD/R7_260...er_average.gif

It is only a 100MHZ bump to the CPU and RAM frequency too,and the HD7850 consumes the same amount of power.

I can only think that is the Tensilica audio SOC,but it still seems way to high.

[kompukare]

HT4u.net have a review of the 750Ti out too:
http://ht4u.net/reviews/2014/nvidia_...st/index16.php

Have to say I do not look forward to Nvidia gaining even more laptop marketshare but then I am still boycotting them after their response (or lack thereof) for all those 65nm Nvidia parts which failed on me.

[CAT-THE-FIFTH]

It still looks like compute performance is behind the AMD cards. One of the reasons,Apple have starting using more AMD cards is due to their improved OpenCL performance.

It will be interesting to see how AMD responds to all of this.