AMD CPUs, SoC Rumors and Speculations Temp. thread 2

[8350rocks]

http://www.hardocp.com/news/2015/10/13/phil_rodgers_amd_fellow_jumps_ship_to_nvidia#.Vh0WE_lVhBc

All hands, abandon ship? A lot of names starting to leave over a short timespan is worrying.

This is an interesting piece of the link:

"Rogers will play a key role in software development for AMD's "Fusion" technology initiative, where CPUs and GPUs are combined and integrated to improve energy efficiency and performance capability."

That's what AMD said when they promoted him and I can see why he left the company. Fusion is not moving forward after the GPU unit streamline. I mean, it might be moving, but I bet they made his position irrelevant with the streamline.

And Jimmy, I don't know what brings NVMe to the table that you can't get already, but sATA is hardly a bottleneck for any system that is not dependent on it's "hard-drive" stuff. I know we all love faster loading times and all, but I don't think the cost benefit of it will make me switch or think of switching any time soon.

Cheers!

EDIT: Typos.

http://www.pcworld.com/article/2899351/everything-you-need-to-know-about-nvme.html

SATA is a bottleneck for SSDs. There are a lot of programs that could benefit from a faster SSD or a SSD that can fully utilize the speeds NAND is capable of.

But this is off topic. My point is that AMD needs to update their chipset, they are sorely lacking, and a lot of people look at these features as part of the platform, not just myself, as reasons to upgrade. You cannot discount a feature just because you don't see a benefit in it. That is like saying that PCIe 3.0 is pointless because no GPU currently saturates a x8 PCIe 3 connection. Yet if you built a system today and only did GPU upgrades for 5 years you could run into a bottleneck with PCIe 2.0 since GPUs will be getting major upgrades in the next few years. Someone with PCIe 3.0 will not.

Then with AMD if you want these features you are now forced to go to their lower end platform, A88X, which currently only has 16 PCIe 3.0 lanes and 4 PCIe 2.0 lanes. This presents a possible CPU bottelneck since the APUs are focused on the GPU more than the CPU. So now if you want a higher performing CPU instead from AMD you are going to get a 970/990FX platform which limits you to PCIe 2.0. AMD could have kept updating the chipset but they only have for new APU sockets. The 990FX chipset has been around since Bulldozers launch (slightly before it actually).

How many SSDs are you running that 3.0 GB/sec is too little? sATA3 is pretty crazy fast. TBH...the NVMe stuff is essentially a new way to package a widget to a consumer to drive them to upgrade. You show me a SSD currently available for less than commercial workstation class money that approaches saturation on sATA3 and I will show you an experimental piece of hardware not fully ramped for production.

Most current consumer SSDs top out at 550MB/s which is the max that SATA 3 gives (BTW it is 6Gbps not 3). Every review shows that a M.2 NVMe based SSD or a PCIe based NVMe SSD performs vastly faster.

http://www.newegg.com/Product/Product.aspx?Item=N82E16820147425&cm_re=samsung_951-_-20-147-425-_-Product

While it is not as cheap as SATA SSDs it is still not as expensive as workstation class SSDs. A lot of people said SATA was also the same thing, just something to drive people to upgrade and that their PATA 100 was fast enough and not bottlenecked. I highly doubt you are anyone would use a PATA based HDD today.

Sure...those drives top out at a maxed out SATA bandwidth performance.

However, aside from workstation rendering reading from one set of SSDs and writing to another set of SSDs...when are you going to use it? When you boot up Battlefield or Call of Duty?

I can see a power user/work station user/developer maybe having use for more than what SATA allows, but those types will already be doing something to accommodate those purposes that suits their needs.

The average consumer has zero use for M.2 or NVMe. As for PATA, there are probably a great deal of consumers out there that would be fine with it aside from the incredibly slow boot/load times.

Honestly...considering how many machines were running Win XP even a year ago...how many consumers/businesses do you think have a desktop PC that is even running SATA3 at this point? Sub 50% for sure...but how far below that threshold?

I mean, you can make the comparison between chipsets...but practicality demands we look at real world scenarios as well. In most applications Intel is appealing to the "has to be the biggest/newest" crowd. If you are as practical as you make it sound on occasion, you can surely see the point I am making. Even a hardcore gamer will seldom, if ever, make use of anything beyond SATA3 at this point.

 

[Vogner16]

maybe sub 50% for sata rev. 3 but sata (any rev.) would be 90% of computers I would guess. even windows xp computers 2004 on had sata (a massive improvement over Pata

 

[gamerk316]

For those of you watching, AMD releases it's financial statement after close today. Considering Intel beat expectations, a miss would be a REALLY bad sign for AMD.

 

[Cazalan]

For those of you watching, AMD releases it's financial statement after close today. Considering Intel beat expectations, a miss would be a REALLY bad sign for AMD.

Up 5% ahead of earnings. The short sellers are concerned it might meet or exceed and are covering their puts.

 

[Vogner16]

http://wccftech.com/amd-8-core-arm-cpu/

new arm cpu from amd. I honestly doubt these benchmarks as it seems early, but something to discuss

 

[Cazalan]

http://wccftech.com/amd-8-core-arm-cpu/

new arm cpu from amd. I honestly doubt these benchmarks as it seems early, but something to discuss

The source is reputable (www.osadl.org). Of course there will be some variations in Linux kernel performance.

Distribution
openSUSE Factory "Tumbleweed"

Kernel
4.0.5-rt4 #42 SMP PREEMPT RT Tue Sep 8 17:34:17 CEST 2015 aarch64

 

[Vogner16]

performance is better than puma. I mean that's what its competing against right? but also twice the tdp so im not sure if its much better.

 

[jimmysmitty]

performance is better than puma. I mean that's what its competing against right? but also twice the tdp so im not sure if its much better.

No this would not be competing against Puma. Puma is a consumer oriented SoC. This is a server oriented SoC that is designed to compete with other low power server solutions such as an Intel Atom or other ARM based low power server chips.

 

[Cazalan]

Revenue was higher than expected but still a big loss due to inventory write down and restructuring costs. Investors seem to be excited about the joint venture that will give AMD an extra 370mil revenue for 1H2016. Meaning they could break positive by middle of 2016.

 

[Cazalan]

"AMD will contribute its Penang and Suzhou assembly and test manufacturing facilities, and will include roughly 1,700 employees and its local management team. Nantong Fujitsu Microelectronics is shown to be taking an 85% ownership stake and will serve as the controlling shareholder of this venture. "


That should really cut operating expense moving 1700 people to the joint venture.

 

[-Fran-]

That 65MM write off was very unexpected. I wonder if that is Richland and Trinity leftovers? That really took a dent in the already red numbers.

And that ATMP (assembly, test, mark and pack?) spin off... That has me a bit buffled. So the QA and distribution (?) will be at the hands of the Chinese Company with AMD's supervision? Not that it changes the way things are currently are by much, but when inventory does not move, that Chinese company will piggyback AMD for it, I'm sure. I might not get a bigger picture of that yet, but smells fishy. In any case, at least it will bring money in and they'll get rid of what hit them the most today. It will cut their profits when things go right as well.

Cheers!

 

[Cazalan]

That 65MM write off was very unexpected. I wonder if that is Richland and Trinity leftovers? That really took a dent in the already red numbers.

And that ATMP (assembly, test, mark and pack?) spin off... That has me a bit buffled. So the QA and distribution (?) will be at the hands of the Chinese Company with AMD's supervision? Not that it changes the way things are currently are by much, but when inventory does not move, that Chinese company will piggyback AMD for it, I'm sure. I might not get a bigger picture of that yet, but smells fishy. In any case, at least it will bring money in and they'll get rid of what hit them the most today. It will cut their profits when things go right as well.

Cheers!

They're being coy about the savings but 1700 employees is another 18% reduction in staff they won't be paying for. They're not in the most expensive area but it should bring down their operating expenses further. And they get 370M cash for it sometime in 1H2016.

The HBM supply issues have been worked out. Newegg fully stocked as I checked today.

In the earnings call Su confirmed multiple finfet tapeouts for both 16nm (TSMC) and 14nm (GF). Possibly the reason for higher expenses this quarter. She said they would see revenue from at least one of the new semi-custom wins in the back half of 2016. Nintendos new console maybe for holiday season??

Su confirmed Zen is still on target for 40% IPC gain.

 

[gamerk316]

http://ir.amd.com/phoenix.zhtml?c=74093&p=irol-newsArticle&ID=2097673

ATMP Manufacturing Facilities Joint Venture
As a part of AMD's ongoing strategic plan to sharpen its focus on designing high-performance technologies and products that drive profitable growth, AMD today announced the signing of a definitive agreement to create a joint venture with Nantong Fujitsu Microelectronics (NFME) that combines AMD's high-volume ATMP facilities and experienced workforce in Penang, Malaysia and Suzhou, China with NFME's established outsourced semiconductor assembly and test (OSAT) expertise to offer differentiated capabilities and scale to service a broad range of customers. The value of the transaction is $436 million and NFME will take an 85 percent ownership of the joint venture. AMD will receive $371 million in cash and expects net proceeds of approximately $320 million, net of taxes and other charges at close. This transaction is expected to close in the first half of 2016, pending successful completion of regulatory approvals.

Gets them to Zen. Still bleeding cash everywhere NOT their semi-custom unit.

 

[vmN]

Guess we no longer have to worry that AMD might not make it to zen.
Good news for AMD.

 

[juanrga]

http://wccftech.com/amd-8-core-arm-cpu/

new arm cpu from amd. I honestly doubt these benchmarks as it seems early, but something to discuss

The source is reputable (www.osadl.org). Of course there will be some variations in Linux kernel performance.

Distribution
openSUSE Factory "Tumbleweed"

Kernel
4.0.5-rt4 #42 SMP PREEMPT RT Tue Sep 8 17:34:17 CEST 2015 aarch64

Those benchmarks are pretty useless and outdated

Su confirmed Zen is still on target for 40% IPC gain.

Her words were:

As we stated in the Financial Analyst Day we had a target of 40% IPC performance of Zen over our previous generation. We believe we’re on track for that.

We believe...

 

[cdrkf]

http://wccftech.com/amd-8-core-arm-cpu/

new arm cpu from amd. I honestly doubt these benchmarks as it seems early, but something to discuss

The source is reputable (www.osadl.org). Of course there will be some variations in Linux kernel performance.

Distribution
openSUSE Factory "Tumbleweed"

Kernel
4.0.5-rt4 #42 SMP PREEMPT RT Tue Sep 8 17:34:17 CEST 2015 aarch64

Those benchmarks are pretty useless and outdated

Su confirmed Zen is still on target for 40% IPC gain.

Her words were:

As we stated in the Financial Analyst Day we had a target of 40% IPC performance of Zen over our previous generation. We believe we’re on track for that.

We believe...

I wouldn't read too much into that- that is simply management speak. Please show me an official statement from a CEO (other than nVidia 😛) that will ever say anything more certainly than that.

The main issue is 40% ipc increase in relation to what exactly? Integer code, FP code, AES encryption? If they don't get a broad 'across the board' 40% ipc increase, they'll find *something* that runs 40% faster on Zen and use that as 'proof' so at this point the statement is pretty meaningless whichever way you slice it.

That said, if those leaked layouts are roughly accurate then Zen looks to have a lot more muscle than any dozer' derived core so it may not be so far off.

 

[Vogner16]

Revenue was higher than expected but still a big loss due to inventory write down and restructuring costs. Investors seem to be excited about the joint venture that will give AMD an extra 370mil revenue for 1H2016. Meaning they could break positive by middle of 2016.

end of day its not good looking amd expects to see $100mil losses for the next year every quarter until zen.

 

[esrever]

I hope Zen comes sooner rather than later. I was expecting AMD to break even this quarter due to the seasonal revenue increases. To see a loss of this size is worrying.

 

[gamerk316]

Here's the thing: A margin of 23% is outright low, and yet, despite practically giving away their product at cost, they needed a $65 million writeoff on unsold inventory. Not good. No one want's their products, even at only 20% above cost.

 

[salgado18]

Here's the thing: A margin of 23% is outright low, and yet, despite practically giving away their product at cost, they needed a $65 million writeoff on unsold inventory. Not good. No one want's their products, even at only 20% above cost.

Which is why they need a new product fast, and it better be good.

I see AMD now as a small, streamlined company, trying to make that one product that comes out with features that the big guys don't have. Unfortunatelly, I don't believe much that Zen and it's platform will break new ground and be special enough to wow the market, but they still need to try.

That said, I think Su's strategy is slowly working: cut the fat, work the muscles and live on a diet.

 

[Vogner16]

Here's the thing: A margin of 23% is outright low, and yet, despite practically giving away their product at cost, they needed a $65 million writeoff on unsold inventory. Not good. No one want's their products, even at only 20% above cost.

Which is why they need a new product fast, and it better be good.

I see AMD now as a small, streamlined company, trying to make that one product that comes out with features that the big guys don't have. Unfortunatelly, I don't believe much that Zen and it's platform will break new ground and be special enough to wow the market, but they still need to try.

That said, I think Su's strategy is slowly working: cut the fat, work the muscles and live on a diet.

you say that and its what they should be doing, but look at operating costs today vs last year. amd spent a TON more than last year this time. is that all due to zen costs of development? I don't think so. where is this spending coming from.

 

[Guest]

Here's the thing: A margin of 23% is outright low, and yet, despite practically giving away their product at cost, they needed a $65 million writeoff on unsold inventory. Not good. No one want's their products, even at only 20% above cost.

Which is why they need a new product fast, and it better be good.

I see AMD now as a small, streamlined company, trying to make that one product that comes out with features that the big guys don't have. Unfortunatelly, I don't believe much that Zen and it's platform will break new ground and be special enough to wow the market, but they still need to try.

That said, I think Su's strategy is slowly working: cut the fat, work the muscles and live on a diet.

you say that and its what they should be doing, but look at operating costs today vs last year. amd spent a TON more than last year this time. is that all due to zen costs of development? I don't think so. where is this spending coming from.

i don't think it's just zen fx 8 core. there will be 4 core, 6 core ones too. there will be zen apus and gpus. and all of them on finfet. there will be server specific chips too. there is no end to demand of money within amd, only supply is dried.

 

[esrever]

Here's the thing: A margin of 23% is outright low, and yet, despite practically giving away their product at cost, they needed a $65 million writeoff on unsold inventory. Not good. No one want's their products, even at only 20% above cost.

Which is why they need a new product fast, and it better be good.

I see AMD now as a small, streamlined company, trying to make that one product that comes out with features that the big guys don't have. Unfortunatelly, I don't believe much that Zen and it's platform will break new ground and be special enough to wow the market, but they still need to try.

That said, I think Su's strategy is slowly working: cut the fat, work the muscles and live on a diet.

you say that and its what they should be doing, but look at operating costs today vs last year. amd spent a TON more than last year this time. is that all due to zen costs of development? I don't think so. where is this spending coming from.

They are in the final design stages of a full range of finfet products. I assume that is a decent bit of spending. That and the continued work on HSA that is only now starting to expand.

 

[Guest]

i'm quite late to notice this -
http://www.kitguru.net/components/anton-shilov/globalfoundries-we-started-to-tape-out-products-using-second-gen-14nm-process-technology/
amd reported about 2 finfet tapeouts, this means the other is from tsmc i.e. greenland gpu. so by q2/q3 2016 we should see 2 finfets from amd in market.

 

[Cazalan]

I hope Zen comes sooner rather than later. I was expecting AMD to break even this quarter due to the seasonal revenue increases. To see a loss of this size is worrying.

Higher than I expected as well. The joint venture really saved their bacon. AMD knows they need a new product so they seem to be doing every thing they can to get the finfet products made.