AMD CPU speculation... and expert conjecture

[blackkstar]

Spoiler

^^ referring back to your BSN(or wherever that was) article from the guy that got fired over a year ago, just before GF announced FD-SOI?

GF SOI plans were announced at 2011 (28nm was SOI)

http://www.nordichardware.com/Science-Technology/globalfoundries-discuss-future-nodes-committed-to-soi.html

AMD decision to ditch SOI by bulk was announced a year latter by Mark Papermaster

http://www.fudzilla.com/home/item/27572-amd%E2%80%99s-2013-to-rely-on-bulk-28nm-process

This same Mark Papermaster

http://www.amd.com/us/aboutamd/corporate-information/executives/Pages/mark-papermaster.aspx

The same Mark Papermaster who gave us the famous HotChip talk about Steamroller.

Glofo has now shown its unability to get ready the 28nm SOI process. A pair of months ago an enthusiast SOI poster here tried to convince all us that GF 28nm SOI was ready and that Kaveri delay was for the transition to SOI. It was all fake. AMD decision to migrate to bulk was a good one.

your first article NEVER ONCE MENTIONED 28NM SOI. I talks about 20nm SHP wich GF canned.

Your second article if you believe everything you read ...

Papermaster said that the company's cooperation with ARM is mainly to satisfy client demand for comprehensive functions and to allow quick product development, and is not founded to target any specific competitor.

only time will tell wether or not bulk was a good decision, a slow-down of 11% on the cpu and 15% on the GPU. It was a cheap decision as it allowed for fast implementation of GCN into all of AMD's APU products.

GF had no 28nm SOI until ST-Ericsson contracted them for their new phone chip that they wanted to produce with 28nm FD-SOI, not 20nm.

http://www.advancedsubstratenews.com/2013/01/st-ericssons-28nm-fd-soi-smartphonetablet-chip-at-vegas-a-great-start-to-2013/

For the folks designing smartphones and tablets (and ultimately for the end-user), that port to FD-SOI gets the NovaThor L8580:
•CPUs running 35% faster and GPU and multimedia accelerators running 20% faster
...
Wondering what’s next? The 14nm FD-SOI node is already in development, the ARM Cortex-A15‘s on the radar, and the FD-SOI roadmap is already defined up the 10nm node.

What actually came next was ST-E breaking up and GF saying fk it to SOI. This was after your papermaker was told there is no soi at 28nm, but would be implemented at 20nm by GF. Now it looks like GF isn't doing any SOI past 32nm.

. The ST-E chip was fully functional in February and operating at 2.8-3.0 ghz

http://translate.google.com/translate?sl=auto&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&u=http%3A%2F%2Fpc.watch.impress.co.jp%2Fdocs%2Fnews%2F20131205_626454.html&act=url

Where are you getting no more SOI?

This is 13 hours old and it's claiming that the CTO of Global Foundaries is declaring FD-SOI necessary for smaller nodes. There is even a slide from this presentation given 13 hours ago with ST right next to Samsung and IBM pointing directly to Global Foundaries.

EDIT: (added spoilers) and have you not picked up on it? NY fab = FD-SOI, Dresden 28nm = bulk. AMD not waiting for 28nm SOI at NY fab because they don't want to push back Kaveri even further and be stuck depending on GloFo to release a new node. Remember when Read said he was going to focus on more proven process nodes?

 

[noob2222]

Spoiler

^^ referring back to your BSN(or wherever that was) article from the guy that got fired over a year ago, just before GF announced FD-SOI?

GF SOI plans were announced at 2011 (28nm was SOI)

http://www.nordichardware.com/Science-Technology/globalfoundries-discuss-future-nodes-committed-to-soi.html

AMD decision to ditch SOI by bulk was announced a year latter by Mark Papermaster

http://www.fudzilla.com/home/item/27572-amd%E2%80%99s-2013-to-rely-on-bulk-28nm-process

This same Mark Papermaster

http://www.amd.com/us/aboutamd/corporate-information/executives/Pages/mark-papermaster.aspx

The same Mark Papermaster who gave us the famous HotChip talk about Steamroller.

Glofo has now shown its unability to get ready the 28nm SOI process. A pair of months ago an enthusiast SOI poster here tried to convince all us that GF 28nm SOI was ready and that Kaveri delay was for the transition to SOI. It was all fake. AMD decision to migrate to bulk was a good one.

your first article NEVER ONCE MENTIONED 28NM SOI. I talks about 20nm SHP wich GF canned.

Your second article if you believe everything you read ...

Papermaster said that the company's cooperation with ARM is mainly to satisfy client demand for comprehensive functions and to allow quick product development, and is not founded to target any specific competitor.

only time will tell wether or not bulk was a good decision, a slow-down of 11% on the cpu and 15% on the GPU. It was a cheap decision as it allowed for fast implementation of GCN into all of AMD's APU products.

GF had no 28nm SOI until ST-Ericsson contracted them for their new phone chip that they wanted to produce with 28nm FD-SOI, not 20nm.

http://www.advancedsubstratenews.com/2013/01/st-ericssons-28nm-fd-soi-smartphonetablet-chip-at-vegas-a-great-start-to-2013/

For the folks designing smartphones and tablets (and ultimately for the end-user), that port to FD-SOI gets the NovaThor L8580:
•CPUs running 35% faster and GPU and multimedia accelerators running 20% faster
...
Wondering what’s next? The 14nm FD-SOI node is already in development, the ARM Cortex-A15‘s on the radar, and the FD-SOI roadmap is already defined up the 10nm node.

What actually came next was ST-E breaking up and GF saying fk it to SOI. This was after your papermaker was told there is no soi at 28nm, but would be implemented at 20nm by GF. Now it looks like GF isn't doing any SOI past 32nm.

. The ST-E chip was fully functional in February and operating at 2.8-3.0 ghz

http://translate.google.com/translate?sl=auto&tl=en&js=n&prev=_t&hl=en&ie=UTF-8&u=http%3A%2F%2Fpc.watch.impress.co.jp%2Fdocs%2Fnews%2F20131205_626454.html&act=url

Where are you getting no more SOI?

This is 13 hours old and it's claiming that the CTO of Global Foundaries is declaring FD-SOI necessary for smaller nodes. There is even a slide from this presentation given 13 hours ago with ST right next to Samsung and IBM pointing directly to Global Foundaries.

http://www.extremetech.com/gaming/172015-leaked-amd-roadmap-shows-excavator-arriving-in-2015-and-possibly-the-end-of-amds-big-core-x86-business

.

GF isn’t rolling out multiple types of 20nm this time around, but is pursuing a unified strategy of offering one type of silicon that can stretch to cover multiple targets.

GF. SOI or no SOI ... that is the question.

EDIT: (added spoilers) and have you not picked up on it? NY fab = FD-SOI, Dresden 28nm = bulk. AMD not waiting for 28nm SOI at NY fab because they don't want to push back Kaveri even further and be stuck depending on GloFo to release a new node. Remember when Read said he was going to focus on more proven process nodes?

proven process nodes ... Is that why he jumped to bulk to see how fast the cpu can function?

Found some other interesting notes.

http://www.bizjournals.com/albany/blog/2013/12/globalfoundries-cuts-jobs-in-new-york.html

30 researchers getting the axe.

The reason AMD is using GF: http://blogs.barrons.com/techtraderdaily/2013/11/18/amd-take-or-pay-globalfoundries-deal-may-spell-inventory-build-says-bernstein/

AMD has to buy 400M in wafers this quarter (likely all kaveri) ... wonder why AMD is pushing kaveri's selling points so hard ... thats why. I would expect some red numbers for Q4 earnings as there won't be any sales from that purchase yet.

 

[Ags1]

Palladin, I put up a new version of the benchmark with the requested changes at:

http://www.headline-benchmark.com/download.jsp

(Posting here because other on this thread seem to have downloaded it).

 

[Ags1]

I made this graph showing cache performance of an 8350 (grey) and i7-2700K (green), both heavily overclocked:

http://www.headline-benchmark.com/results/8234095e-f69e-41a0-ad23-e7f077c33ded/9a411141-526d-4452-8e78-1c9891c2efa0
For small block sizes, the caching performance predominates, and for larger block sizes the system memory predominates the results.

Does Steamroller offer any improvements in cache performance, or is it largely irrelevant as AMD are only shipping APUs without L3 cache?

 

[noob2222]

http://www.headline-benchmark.com/results/1cbd380e-ff9f-4fed-846f-639876c9bdee

ill play with some settings later and see how that affects things. Don't remember right now where I have everything set at as far as nb, htt, fsb.

waiting for my 290x water block still, might put it back in just to test it again if i can do it without unhooking the 6970 hoses.

^^ also looking at Gamerk's bench, his memory testing carried all the way down to 2mb and yuka's dropped off at 128k. Did GK test with a xeon to have that much caching?

 

[Ags1]

Thanks!

Palladin's 780 doesn't score as high as it should. No idea why - it would be interesting to see where your 290X comes in - if it also comes in at the same level as a GTX770 then there is definitely something wrong with my graphics tests!

also looking at Gamerk's bench, his memory testing carried all the way down to 2mb and yuka's dropped off at 128k. Did GK test with a xeon to have that much caching?

...random statistical noise? Also, the memory curve is sensitive to anything else going on on the computer.

 

[noob2222]

^^ could be, his cpu and memory aren't detected either.

Checking some others, looks more like a haswell system.

http://www.headline-benchmark.com/results/0f004cfc-1942-4647-97cd-1a3970ade933

or ... now im confused, yuka's seems to be the one thats off with the drop after 128mb.

http://www.headline-benchmark.com/results/93a871e9-4a76-42aa-8f91-ef08fedef7dc

 

[Ags1]

I use wmic to get the CPU name - but it is returning 0 on GKs 'puter and only his 'puter. I'm going to try pull the data from the registry as a fallback.

^^^^If there are other things going on during the memory test, they can suppress part of the memory curve. Because it needs lots of data points, it also means I do less sampling of each point, so the results can vary a bit. From what I've seen, the curve goes off a cliff around the size of the L3 cache - which is what you would expect and 128mb looks too low.

 

[de5_Roy]

The reason AMD is using GF: http://blogs.barrons.com/techtraderdaily/2013/11/18/amd-take-or-pay-globalfoundries-deal-may-spell-inventory-build-says-bernstein/

AMD has to buy 400M in wafers this quarter (likely all kaveri) ... wonder why AMD is pushing kaveri's selling points so hard ... thats why. I would expect some red numbers for Q4 earnings as there won't be any sales from that purchase yet.

woa.. i didn't know about the figures. that's a lot of wafers. i thought amd changed the wafer agreement with glofo when they spun them off of amd. the deal looks exploitative towards amd. who made that? ruiz?
i remember an old roadmap partially revealed by semiaccurate where it showed glofo was supposed to manufacture some of the lower end gcn gpus... such as bonaire, oland, curacao.. if i remember correctly. if discreet gpus were mixed with... i think the wafer agreement would make a bit more sense to me. now amd has to sell apus only (or may be ramp up fx and centurions... is that why centurion prices keep dropping?)...loads of them. could have a "llano" redux if amd doesn't execute carefully.


who needs intel when amd's got glofo? i hope people will stop bringing up intel's illegal shenanigans without mentioning glofo's exploitation and incompetence when they discuss amd's poor state.

 

[noob2222]

^^ Im sure amd wanted to use gf for some radeons. Problem is that apparently 28nm is barely ready, kinda hard to purchase wafers that arent ready. GF should have to eat the remainder since their the ones that delayed soo much.

Either way, things don't look good for either company.

 

[jdwii]

I'd hate to say it but GF day's are probably numbered i bet some company will buy these guy's out one day their just completly crap. Delay this underperform here that's all they do, and talk about future products before they can even get their current ones right

 

[de5_Roy]

I'd hate to say it but GF day's are probably numbered i bet some company will buy these guy's out one day their just completly crap. Delay this underperform here that's all they do, and talk about future products before they can even get their current ones right

glofo was the 3rd biggest contract semiconductor foundry, iirc.
i'd like to know if mediatek and rockchip are exclusive to glofo or they source from other foundries. i know mediatek and rockchip have massive arm marketshare in some markets - that would not be possible without high yield, high volume socs. if mediatek's and other ulp arm socs are made on 28nm bulk version, then glofo might not have issues with the mobile nodes, just with the high performance plus node.

 

[synce]

Kaveri sounds perfect for my future mini-ITX build. Can anyone reliably say what kind of GPU performance to expect, relative to current cards? I'm hoping at the very least it's more powerful than my laptop's GTX 560m

 

[juanrga]

etayorius Richland was not planned originally.

It was released because Kaveri was delayed to 2014. AMD will release a Kaveri update a la Richland, in 2015, only if Excavator gets delayed, which hopefully I don't expect.

noob2222 Globalfoundries official 2011 roadmap

Mark Papermaster announcement of selecting 28nm bulk was made one year latter (2012)

http://www.fudzilla.com/home/item/27572-amd%E2%80%99s-2013-to-rely-on-bulk-28nm-process

Besides your continuous post of misinformation and lies like your "the guy who announced bulk is not more working at AMD"... I note your hypocrisy again. You complain against me when ARM is discussed in this thread, but now you use an article about SOI vs bulk to quote a part of it about ARM. You don't want ARM being discussed here but you are the first to bring up ARM to the discussion... Guess what? I am not going to bite.

 

[gamerk316]

http://www.headline-benchmark.com/results/1cbd380e-ff9f-4fed-846f-639876c9bdee

ill play with some settings later and see how that affects things. Don't remember right now where I have everything set at as far as nb, htt, fsb.

waiting for my 290x water block still, might put it back in just to test it again if i can do it without unhooking the 6970 hoses.

^^ also looking at Gamerk's bench, his memory testing carried all the way down to 2mb and yuka's dropped off at 128k. Did GK test with a xeon to have that much caching?

^^ could be, his cpu and memory aren't detected either.

Checking some others, looks more like a haswell system.

http://www.headline-benchmark.com/results/0f004cfc-1942-4647-97cd-1a3970ade933

or ... now im confused, yuka's seems to be the one thats off with the drop after 128mb.

http://www.headline-benchmark.com/results/93a871e9-4a76-42aa-8f91-ef08fedef7dc

Standard 2600k, at stock. 8GB DDR3 (forget the brand offhand). Might be the fact I've highly tuned my Windows installation for performance, and something is getting messed up settings wise...I'll toy around over the weekend and see if i can figure out why the CPU/RAM isn't getting detected right, but the GPU is...

 

[juanrga]

Spoiler

I made this graph showing cache performance of an 8350 (grey) and i7-2700K (green), both heavily overclocked:

http://www.headline-benchmark.com/results/8234095e-f69e-41a0-ad23-e7f077c33ded/9a411141-526d-4452-8e78-1c9891c2efa0
For small block sizes, the caching performance predominates, and for larger block sizes the system memory predominates the results.

Does Steamroller offer any improvements in cache performance, or is it largely irrelevant as AMD are only shipping APUs without L3 cache?

Steamroller includes a bigger L1 cache that improves cache misses with efficiency back to 99% and includes a faster L2 cache of the same size. I am being said that the SR L2 cache is 20% faster than in BD/PD. The remaining unknown is the performance of the new IMC.

 

[juanrga]

Kaveri sounds perfect for my future mini-ITX build. Can anyone reliably say what kind of GPU performance to expect, relative to current cards? I'm hoping at the very least it's more powerful than my laptop's GTX 560m

Kaveri iGPU graphics performance will be between a HD7750 DDR3 and a HD7750 GDDR5.

Kaveri iGPU compute performance will be better thanks to hUMA.

 

[etayorius]

Kaveri sounds perfect for my future mini-ITX build. Can anyone reliably say what kind of GPU performance to expect, relative to current cards? I'm hoping at the very least it's more powerful than my laptop's GTX 560m

Kaveri iGPU graphics performance will be between a HD7750 DDR3 and a HD7750 GDDR5.

Kaveri iGPU compute performance will be better thanks to hUMA.

We know Kaveri will have greater IPC compared to Phenom, but how about actual Multithreaded performance? how much faster will Kaveri be compared to Phenom in this regard?

If Kaveri can Match or get very close to the i5 2500k performance without HSA/MANTLE, it will be a win... if it´s 10% slower or more to the i5 2500k then it will be a fail.

I really want to get Kaveri and pair it with a R9 270X or 280X, if it can get very very close to the i5 performance, with a 10-15% OC should be quite good.

 

[tracker45]

We never know what AMD will release, I don't think they know themselves.

I mean, who saw the fx-9590 coming.

 

[gamerk316]

Kaveri sounds perfect for my future mini-ITX build. Can anyone reliably say what kind of GPU performance to expect, relative to current cards? I'm hoping at the very least it's more powerful than my laptop's GTX 560m

Kaveri iGPU graphics performance will be between a HD7750 DDR3 and a HD7750 GDDR5.

Kaveri iGPU compute performance will be better thanks to hUMA.

We know Kaveri will have greater IPC compared to Phenom, but how about actual Multithreaded performance? how much faster will Kaveri be compared to Phenom in this regard?

If Kaveri can Match or get very close to the i5 2500k performance without HSA/MANTLE, it will be a win... if it´s 10% slower or more to the i5 2500k then it will be a fail.

I really want to get Kaveri and pair it with a R9 270X or 280X, if it can get very very close to the i5 performance, with a 10-15% OC should be quite good.

If Kaveri is 15% faster then PII in single-core IPC, then you can calculate multithreaded performance by simply scaling that up to four cores.

 

[Ags1]

Does the the thread die at 10,000 posts?

 

[gamerk316]

We broke the BD thread; why not this one. Go for 10k!

 

[tracker45]

Kaveri sounds perfect for my future mini-ITX build. Can anyone reliably say what kind of GPU performance to expect, relative to current cards? I'm hoping at the very least it's more powerful than my laptop's GTX 560m

Kaveri iGPU graphics performance will be between a HD7750 DDR3 and a HD7750 GDDR5.

Kaveri iGPU compute performance will be better thanks to hUMA.

What's the nvidia equiavlent of that ?

What's the chances they release something faster in the future ?

 

[szatkus]

What's the chances they release something faster in the future ?

100%. You know, process shrink etc.

 

[Ags1]

http://www.headline-benchmark.com/results/1cbd380e-ff9f-4fed-846f-639876c9bdee

ill play with some settings later and see how that affects things. Don't remember right now where I have everything set at as far as nb, htt, fsb.

waiting for my 290x water block still, might put it back in just to test it again if i can do it without unhooking the 6970 hoses.

^^ also looking at Gamerk's bench, his memory testing carried all the way down to 2mb and yuka's dropped off at 128k. Did GK test with a xeon to have that much caching?

^^ could be, his cpu and memory aren't detected either.

Checking some others, looks more like a haswell system.

http://www.headline-benchmark.com/results/0f004cfc-1942-4647-97cd-1a3970ade933

or ... now im confused, yuka's seems to be the one thats off with the drop after 128mb.

http://www.headline-benchmark.com/results/93a871e9-4a76-42aa-8f91-ef08fedef7dc

Standard 2600k, at stock. 8GB DDR3 (forget the brand offhand). Might be the fact I've highly tuned my Windows installation for performance, and something is getting messed up settings wise...I'll toy around over the weekend and see if i can figure out why the CPU/RAM isn't getting detected right, but the GPU is...

I altered the maths on the site slightly to resolve the issues of variance you could see in some scores. The system scores were giving a lot of weight to the performance found at around 4 threads (ostensibly to simulate gaming loads) but I have now dropped that as the results were a bit unreliable and varied a lot from run to run. Also, a user informed me that the 4-thread results could be massively influenced by playing with the process priority, but now I am using data points that are largely immune to that.