AMD CPU speculation... and expert conjecture

Page 445 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.


didn't amd say they were going for the "pads" and "smart phones" i remember them saying something to that affect in some article.
that 7600 is quite the little beast,may just make a htpc with one of those.I would love to see the benches for base lines on beema and mullins, gotta be pretty impressive.
 


Might want to read the fine print on those roadmaps...especially the part that says "AMD reserves the right to modify or change this roadmap at any time without notice..."
 


That is a legal disclaimer that fits many purposes. E.g. in older roadmaps Kaveri appears as successor to Trinity. But AMD modified the older roadmap and launched Richland as successor whereas delayed Kaveri to 2014.

Carrizo appears as successor of Kaveri in the official 2015 roadmap. If some problem happens with Excavator, the extended HSA, the new GCN, or the new 20nm node, AMD could delay Carrizo and release some Kaveri refresh a la Richland. Legal disclaimers allow this.

You can believe that AMD will modify the roadmap and will release FX Steamroller in 2015. I maintain the chances of that happening are 0.001%.
 

Yea this is pretty much what I figured...although FX looks like it's dropped off the roadmap there is still a lot of investments taking place on the software end. To much effort in optimizing for multi-core, for a supposed dead line of CPU's.

They do need to work on support before they throw another chip out that has a max increase of performance of ~5% because they didn't get the developer backing for it.
 

amd is playing it really really safe when it comes to tablets. they don't have an x86 smartphone chip yet. i think that by omission, they admitted that they might go the arm (a57) way for smartphone and tablets should the opportunity present itself. at first they said that they're all for windows, then changed tune a few weeks later announcing support for android. amd doesnt have intel's big fat wallet to get oems to build temash tablets despite few design wins. although, i don't know why amd didn't push hard for temash adoption. may be console wins were good enough for them. may be intel's push intimidated them. because, if amd ever gains marketshare in ultramobile, it'll create an better opportunity for intel to bite a chunk from smaller amd than from bigger stronger competitors like qualcomm. may be that's why amd is waiting for a strong product that can withstand anything intel can throw at amd. although.... going arm will open up a chance for another competitor - nvidia. i think that's why amd is playing it safe.
 
^^ that was really dumb, noob. anyone who has read and/or posted in this thread for a while knows very well what juanrga is and what he does. refusing to feed juanrga is not a gesture of being supportive of his actions. it's a tried and true method of dealing with people like him. you should have let it go a long time ago.

and seriously, kaveri isn't exciting? it's amd's most complex mainstream chip. imo, only console socs and kabini/temash come close. there's a crap load of stuff built into kaveri, all under the same power limit and die area, working together. i'd even call it amd's own semi-custom chip due to the true audio logic and arm cores for trustzone.
 
@truegenius, thanks for the suggestion...yes lenovo e4325 is not far from what i need.
but, a little extra cost and it could have been equipped with one of the win versions. i probably would have borne the marginal cost for the pre-installed Win.
and a quad core and 4gb ram would have helped.
All these might have made it cost around 30000 instead of the 21000 it is now.
but the marginal changes would have made it a lot more desirable as a laptop.
i understand u don't use virtual mem, and 8 GB is not enough for u...
Wow, had no idea Win had become so bloated.

@Juanrga, here where I live (in my best knowledge) an insignificant %age of people use FreeDos.
I might be wrong though.
But of course glad to hear there are other initiatives out there trying to reduce the Win monopoly and increase choice for us.

@de5_roy, I am (very) eagerly waiting for more AMD tablets to come out in my market (India), be it with x86 or with the ARM licenses
Currently there is an AMD-powered tablet (named Xolo Win) that is supposed to be released by Jan end.
http://www.techtree.com/content/news/5326/ces-2014-xolo-win-tablet-first-amd-apu-powered-tablet.html
Hope the other components of the tablet are designed well, as opposed to what you and I (and may others) fear
(that link is of a good Indian tech website I have been following since 2003, the year I discovered AMD Athlon)
 

i think that the xolo tablet can be easily overwhelmed with basic task like web browsing or hd movie playback. a 4c version with 4GB lpddr3 ram or just 4c cpu would be much better. windows, antimalware and other background programs together will occupy at least 1GB, igpu will take up some more, leaving less than 1GB for the user. but the worst bit is the meager 32GB internal storage. one unknown is the display type. if xolo decides to cut further corners, they could end up using a TN display. afaik, indian markets are swarmed by mediatek-powered tablets.

as for amd and arm tablet soc, i'd like to add a bit more speculation. the biggest problem may be amd's competitors, regardless of uarch. if amd does x86, intel will go for the low hanging fruit that is amd's marketshare. if amd does arm, then qualcomm, mediatek (which has spread like malware on unprotected pcs, (i skipped the vd analogy :ange:)), nvidia will do the same. amd's best chance may be with niche segments where no one is bothering to compete yet e.g. firefox o.s., ubuntu and tizen (or whatever rises from it's ashes after it's predictable demise a la meego) powered devices and with windows and windows RT.
 


I am beginning to see a lot of software support being pushed around by AMD, adobe,java trying to push mantle, hsa, opencl. seems amd is trying to take away intels optimized software/compiler advantages. this could play well in amds favor if they can pull it off. maybe RR can get manufactures to start selling more of their product with what they are doing too. they got the consoles which probably isn't a huge profit margin and those will be selling for a bit. Kaveri looks very promising since they will most likely improve over time.

AMD still needs a good HPDT chip, maybe we will see it sooner than 2015, steamroller a,vs steamroller b?
sounds like they had 2 chip designs to me, only since global is out of room and they opted kaveri first which uses steamroller b? and steamroller a is a HPDT version?
 

Thing is they are getting SUCH bad press over the HP lines unless they make decent gains.
Even Vishera got torn out in some reviews. As an owner, I can say it's not bad at all, but AMD needs to literally match or beat Intel in almost all benchmarks to keep from being called a disappointment(especially in gaming, where are the majority of enthusiasts). Steamroller doesn't look like it's going to do that with the software being used now.

I think they are way better off holding it until they get the support they need, and then releasing something. Almost think they're stupid for releasing PD without getting the support they are now, alongside of it.
 

exactly my point
when they went from 65nm to 45nm they got ( comparing quad core versions )
1.1GHz higher clock
27mm^2 less die area
same tdp
in short, they gave 40% higher performance (40% is clock rate only, so if i count other improvements too then it is 50% increase ) which costs 10% less to amd to manufacture ( both on soi )
http://www.cpu-world.com/Compare/325/AMD_Phenom_II_X4_980_vs_AMD_Phenom_X4_9950_(125W).html

if i compare 1100 vs 9950 then clock and core advantage turns out to be 90% but more die area

so if they can provide this much gain just by die shrink then they can provide this by another die shrink ( don't know what was the cause of low clocks on k10 in fm1 apu, maybe tom and jerry ( cartoon ) did some damage to their tweaked chart/design :sarcastic: )

they should have added more l2 and l3 cache and should have tweaked the memory controller ( by clocking my ram at 1800mhz and NB at 2.7Ghz i get 5-8% higher score in cinebench at same cpu speed (1090t@3.6GHz), which means 3.6Ghz ( nb and ram tweak) behaves like 3.8-3.9GHz (stock nb and ram))

my estimation/calculation is that they can get 50-66% more multithreaded performance ( over 1100t) by die shrinking, adding more cores and clock speed keeping same 125w tdp ( means it will be better than 8350 which got only 15-20% performance increase (CB11.5) by increasing raw power ( clock and core ) by total of 66% )

i suspect that even steamroller won't be able to match these hypothetical 32nm phenoms ( 8 k10 cores at 3.6 - 4GHz )

intel is doing die shrink and little tweaking to their sandy/ivy/has/broad, so why not amd ( we all know that it will take miracle for amd to catch intel in per core ipc, so better is to save R&D cost )
intel did good thing when they ditched the pentium 4, but amd is carrying bulldozer stupidity



this is the first amd based tab that i have seen which will be launched for Indian market
but it will have window ! :heink: nokia tried to sell windows based phone and then it went down in market share, i would have bought lumia 525 asap but won't buy it due to windows on it, no other brand provide this much specification at that price range (not even these cheap indo-chinese manufacturer)
i am sure that they will price this xolo win tab in INR20-30k range :pfff:


absolutely correct
now every Indian phone manufacturer is providing mediatek's quad and dual core which are based on cortex-a7 paired with crappy gpu
to top it up, they are launching every good speced phone in phablet segment, even their mini version comes in phablet segment ( like canvas turbo mini ), every manufacturer is making this phablet insanity, only Apple is not doing such stupidity

and now what, we are seeing true octa core based phablets, which have ( as per my expectations ) mediatek's chipset (mt6592), Y U No Providing Krait :fou: ( cheapest krait based android available here is xperia M and maybe Hawaii hw-01e too ( not sure about this one ))

the one that have phone's form factor ( 4" ) have very low specs like only 512MB ram even my almost 2 year old phone (which costed me same as these new 512MB phones) have 512MB ram ( when i purchased my phone, nokia was selling lumia 510 which comes with 256MB ram, but now nokia's lowest end lumia series ( series 5 ) have 1GB ram which is priced in lumia 510's price bracket, but these Indian manufacturers got stuck at 512MB only )
 
^^

thats what i think they are doing, they are producing kaveri while waiting for applications/programs to come into play. before they start production of a HPDT chip. if the kaveri 7850k cpu side gained up to 9% on a crappy bulk node, whats to say that Steamroller A would on a better process node perform 2 times that at 18% and more if the statements made by AMD reps are true these would be ivy/haswell performance lvl or better.

I admit I am not the smartest techy ,but from what I am seeing, AMD is playing the field and biting Intel back at their own games. RR seems to know what he is doing, as we are seeing some smarter moves from AMD.


 
@truegenius: amd's microarchitecture has to be able to clock high first, i read that bd had been designed that way, but k10 was probably not. may be that's the reason llano's stock clockrate didn't go higher than 3.0 ghz, same with core scaling. that's why i don't think a 32nm shrink of thuban would yield much benefit. i don't think it would even lower power consumption since thuban's turbo was not that good... same with power management. imo kaveri, richland and kabini have amd's most advanced power management system. then there're the new instruction set support like avx, aes-ni and fma3/4. L2 and L3 caches get added per design requirement and being made of SRAM, adding cache is quite costly. for better or worse, amd is stuck with bd until excavator/carrizo comes out.

@juggernautxtr: RR is actually enjoying the results of years of r&d. his own efforts didn't pay off much. he miscalculated how far the pc market would shrink or the impact of tablet sales on netbooks which use amd's low power apus/socs.

 


AMD released a 32nm upgraded K10 CPU, it was known as Llano. Seriously go check it out, the A8-3850 and A8-3870 were the best they could do. Trinity and Richland are both BD uArch design's and do much better. People are worshiping K10 because it was the last time AMD was competitive with Intel. AMD CPU's didn't get worse, Intel started investing heavy amounts into R&D and created SB which was an amazing design. AMD, having a much lower R&D budget, couldn't do the same and instead went a different route.
 
To the people talking about AMD and OEM releases. HP seems to be the ones really offering design's, but it seems to be a regional thing. From what I've seen most OEM's are deliberately handicapping AMD models with less options as to protect their higher margin offerings. This is the same thing car manufacturers do. An 1xx model won't have the same options as a 3xx model even though the two share compatibility and the same engine and transmission. They do this to give an incentive for car salesmen to "sell up" customers.
 

phenom out ! :??::heink:
1521501_782622398433221_1319098744_n.jpg



atleast they said this ""Thuban processors are today still solid chips that do not need to upgrade yet"" :pt1cable:

palladin9479 said:
AMD released a 32nm upgraded K10 CPU, it was known as Llano. Seriously go check it out, the A8-3850 and A8-3870 were the best they could do.
Click to expand...
their best effort resulted in poor clock rates :pfff:
this can't be their best
1557531_654987691219920_546926955_n.jpg
 


I was talking with a bud of mine a while back who used to work with AMD. He said the smartest thing AMD ever did was buy ATI. He proceeded to tell me the stupidest thing they've done is not invest in their own fabs when they had the $$, instead of trusting that to Global Screwanddries. I think AMD should just do something smart and skip multi-core steamroller chips, and do an early release on excavator. I look forward to buying a multi-core excavator chip. I'm just going to upgrade my PSU, grab a 240Gb SSD and possibly Windows 9 while I wait! Haha! I don't see hasfail'd 8 core chips running at an efficient GHz.
 


Again stop using that example that was on GF horrible first try at their 32nm design, not to mention Amd did not improve a lot of things that they could have on Llano. Look at the article i posted on 32nm BD did worse then Phenom in most of their test's when the phenom was on a 45nm design. For the 100th time until Amd drops the module design they will never have a great CPU they will have a meh CPU with extra features. PD is easily a better processor compared to the X6 by around 20% on average however that is with a newer memory controller-RCM, newer instruction sets and yes a more mature 32nm design. Llano did not have any of this AND their was a IGPU attached.
 


Every design has a natural limit to what you can do with it. Eventually it needs to be reworked to fit into a different manufacturing technology. That reworking is expensive and costs grow astronomically as transistor count grows. The 980BE was only clocked @3.7 though you get it to about 4.0 before energy usage and heating got stupid. The 1100 was 3.3Ghz. BD was AMD's answer to the speed wall they had hit, they couldn't afford to pour the amount of R&D required to redesign it to go much faster so they instead opted for a modular approach. This got them higher clocks but at the expensive of a less efficient design. They are now in the process of reworking and improving the design one piece at a time. That's why you will see 10~15% efficiency increase in each iteration, they are fixing small things and tightening up the design. SR APU's #1 "CPU" problem is the manufacturing technology they went with. Vector processors require high density to pack in the amount of transistors they need while scalar processors require high clock speeds to use the same transistors more often. AMD wanted to make the graphics performance of the APU stand out and so went with a manufacturing technology better designed for that purpose, at the expense of lower clocks rates. It's the lower rates that masked the CPU efficiency increase's.

Honestly I see this the same way I see Haswell. While Haswell wasn't much faster then IB, it used less power and had a significantly better iGPU. In that same concept the A8-7600 has about the same performance as the 6800K while using much less power. The A10-7850K was just for press release as it's far too expensive to be taken seriously, the A10-7700K is in a much better position @ $160 though it should drop to $140 soon. The A8-7600 should be amazing when it's out, assuming it's price is $120 USD as advertised. APU's make the most sense when their hovering south of $150 USD as a decent dGPU costs $100 USD and your not getting a better CPU for under $50 USD. The closer to $200 USD you get the more realistic a dGPU + dCPU becomes unless your power / space limited.
 
Mantle patch for ‘Battlefield 4′ reportedly delayed again
http://vr-zone.com/articles/mantle-patch-battlefield-4-reportedly-delayed/70627.html
just a little longer...


did amd really go with that process or glofo forced amd by declining to optimize soi? it seemed to me(could be due to my lack of knowledge) that glofo asked for too much money from amd to develop 28nm hpp like tsmc did for their 32nm node. glofo did manage to fabricate apus on 32nm soi process, and amd seemed to achieve higher ipc with sr.
 

So when defeated in a debate your answer is to demand the opponent stop using material that defeats your statements? Yeah that ain't gonna happen.

The CPU component of Trinity was tested thoroughly with the iGPU disabled, which isn't hard to do btw. Right now you can buy a A8-3870K for ~$100 USD, put a dPGU with it and disable the iGPU. That will net you a clock increase to about 3.7 Ghz on air, water can get you a little higher but you slam right into that 3.8~4.0Ghz wall that the previous generation Phenom II's ran into. Only way past that is some pretty heavy cooling design's and lots of juice. BD, Llano and PD all are produced on the same manufacturing technology, the increases in PD are the results of tuning the uArch. Phenom II has already been highly tuned as it's derived from Phenom and Athlon with each successive iteration being a tuned version of the former. Eventually you get to a point where you can't tune any further and your only option to to move to a smaller manufacturing node and crank up clock speeds, which has it's own problems with thermal density. Intel proved this quite well with Haswell. Could go lower power and higher transistor density but the thermal load is too high for cheap consumer cooling. The chip is simply too small, transistors packed too tightly and thus very sensitive to heat.
 


Honestly without deeper information I can only speculate. We know there were two SR design's and my gut feeling is that each was targeted at a different manufacturing technology. AMD GPU's are already made with a 28nm high density process from TSMC so there was already the option of AMD going that route. The chip's wouldn't of been clocked very high though. Going with a different manufacturing process would of allowed high chip speeds but the GCN GPU would of had to been leaner. Look at the chip density and imagine how much of that is taken up by the iGPU then compare that to the previous iGPU's. Everyone can agree that the iGPU portion of these APU's had a large performance increase yet nothing's free so something had to give to get that.
 

speculation is just fine with me. that's what this thread is for, after all. :)
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom