Soldato
One of the few guys that posts with factual accuracy, is constructive and doesn't throw his toys out when posting - top man the DM!
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD Launches Three Kaveri APU SKUs in February 2014 – Feature Set For A10 and A8 APUs Detailed
- Thread starter Boomstick777
- Start date
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Nice to be appreciated, a rare thing on these forums.
One of the interesting things to consider is in a few more years when GloFo has been up to speed with TSMC for a couple of years, a trusted alternative and making seemingly as good chips as TSMC can, then we might start to see a bit of competition in the foundry space which is going to be good for pricing for everyone. Processes are getting pretty ridiculous at the moment, what would cost low 100's of millions to low billions a decade ago for companies to make the switch from one process to the next, that cost has sky rocketed since maybe 65nm towards closer to $10billion for a process shrink. I think it was only 40nm that wafer costs were around the $3k mark at TSMC. These are 300mm circular wafers so it works out as roughly say 200 chips per wafer at Tahiti size(350mm2 or so). SO you're looking at $3k/200 = $15 a chip. Reality is failed chips/yields push that to closer to $30 a chip, then to actually cover R&D costs, production and some actual potential profit you're starting to look at more like $100 a chip minimum for AMD to see any profit from it.
With 28nm those wafer costs went up hugely already towards more like $6k + probably pushing that cost per chip to more like $150-200, and the predicted cost of a 10nm wafer is 4x that of a 28nm wafer which is not going to be good. One of the reasons AMD/Nvidia for the first time in 15 years aren't going to be pretty much the first full production chips out on a new process, is it will simply cost too much. You end up with lower yields early in a process life and with lets say maybe $10k a wafer on 20nm... poor yields will mean per chip costs that AMD/Nvidia can't produce £300-400 graphics cards at. It's fine to have a Titan or 290x, but neither company sell a huge amount and neither company could only sell something that costs that high.
Production of ARM chips is pushing demand through the roof which also increases costs. Apple taking the bulk of the first 6 months of TSMC production for damn Iphones screws everyone else.
So one of the key things we actually need over the next 5 years and through the drop towards 7-12nm processes, is competition between the foundries themselves. GloFo getting more capacity is pretty crucial to increasing supply, reducing demand, reducing costs and letting big chip makers like GPU companies not get squeezed out of the running for a competitive process.
One of the most interesting possibilities is AMD switching to GloFo eventually for GPU's, it could potentially happen for 20nm. Even Intel are having trouble both with processes themselves, and demand and using up the capacity available in their own fabs so they've been taking on more and more customers and are becoming a small scale foundry of their own. But being Intel they are somewhat trying to help people who could hurt/take market share off their competitors but in markets Intel aren't themselves competing.
One of the interesting things to consider is in a few more years when GloFo has been up to speed with TSMC for a couple of years, a trusted alternative and making seemingly as good chips as TSMC can, then we might start to see a bit of competition in the foundry space which is going to be good for pricing for everyone. Processes are getting pretty ridiculous at the moment, what would cost low 100's of millions to low billions a decade ago for companies to make the switch from one process to the next, that cost has sky rocketed since maybe 65nm towards closer to $10billion for a process shrink. I think it was only 40nm that wafer costs were around the $3k mark at TSMC. These are 300mm circular wafers so it works out as roughly say 200 chips per wafer at Tahiti size(350mm2 or so). SO you're looking at $3k/200 = $15 a chip. Reality is failed chips/yields push that to closer to $30 a chip, then to actually cover R&D costs, production and some actual potential profit you're starting to look at more like $100 a chip minimum for AMD to see any profit from it.
With 28nm those wafer costs went up hugely already towards more like $6k + probably pushing that cost per chip to more like $150-200, and the predicted cost of a 10nm wafer is 4x that of a 28nm wafer which is not going to be good. One of the reasons AMD/Nvidia for the first time in 15 years aren't going to be pretty much the first full production chips out on a new process, is it will simply cost too much. You end up with lower yields early in a process life and with lets say maybe $10k a wafer on 20nm... poor yields will mean per chip costs that AMD/Nvidia can't produce £300-400 graphics cards at. It's fine to have a Titan or 290x, but neither company sell a huge amount and neither company could only sell something that costs that high.
Production of ARM chips is pushing demand through the roof which also increases costs. Apple taking the bulk of the first 6 months of TSMC production for damn Iphones screws everyone else.
So one of the key things we actually need over the next 5 years and through the drop towards 7-12nm processes, is competition between the foundries themselves. GloFo getting more capacity is pretty crucial to increasing supply, reducing demand, reducing costs and letting big chip makers like GPU companies not get squeezed out of the running for a competitive process.
One of the most interesting possibilities is AMD switching to GloFo eventually for GPU's, it could potentially happen for 20nm. Even Intel are having trouble both with processes themselves, and demand and using up the capacity available in their own fabs so they've been taking on more and more customers and are becoming a small scale foundry of their own. But being Intel they are somewhat trying to help people who could hurt/take market share off their competitors but in markets Intel aren't themselves competing.
Last edited:
Soldato
Suppose that is what is intel's logical yet selfish business model. 
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Suppose that is what is intel's logical yet selfish business model.
Oh there isn't anything wrong with it, it's smart, but the fact that Intel is taking on customers more each process backs up that both their volume is down and the very widespread rumours of fairly low utilisation of their fabs which is bad. Not great to spend a few billion on equipment to get a fab to a new process just for half of it to go unused while you have people working there and the equipment being used but only for small batches.
Volume is a difficult one though, I don't actually know how many chips Intel shipped say 5 years ago vs today, desktop has tanked, laptop has gone up(compared to Intel, I don't know if it's up or down vs laptop 5 years ago), mobile stuff like atom is selling more. Their real issue is 5 years ago they were making quad cores that were significantly bigger than they are making their quad cores today, so effectively they get 50-100% more chips per wafer, which means, even the same actual number of chips on a smaller process leads to a lower number of wafers being needed. they are still profitable but they want to eek every last cent out of it, who wouldn't.
What TSMC do being a foundry with pretty much 1000's of customers all wanting different things is, they don't upgrade every fab for every process. They still have 180nm wafers ticking along in some small fab somewhere, so do GloFo. What they'll do is say for 28nm, they built a new fab, that is dedicated purely to 28nm, and they'll switch over 1 or 2 fabs to 28nm also, but not all of them. Then when they go 20nm, they'll pick 2 other older fabs and upgrade those instead, so the 28nm gets used by customers that want less high end process, so it gets used for say 5+ years at least, maybe 10.
Intel and used to be AMD were the bleeding edge, old processes were all but worthless to them(they'd do chipsets in some small corner on older processes on the cheap), so Intel have 5 fabs, each one gets kitted out to the tune of billions with every process upgrade. I think it's pretty likely we'll not see that in the future from Intel as they have alternative customers. Leave a couple fabs 22nm, upgrade a few for 16nm, keep that mostly to themselves. In 2-3 more years upgrade the 22nm to 10nm. It could drastically reduce their costs and leave them happier to take on customers.
You want a process advantage, keep the best for themselves and take on customers on "older" tech to give them little to no advantage, win win for Intel.
Glofo as they increase number of fabs will do the same.
Here’s a look at a Kaveri APU engineering sample
http://vr-zone.com/articles/heres-look-kaveri-apu-engineering-sample/62565.html
http://vr-zone.com/articles/heres-look-kaveri-apu-engineering-sample/62565.html
Soldato
According to some speculation over on the SA forums,the CPU code indicates a 3.5GHZ base clockspeed.
A healthy clock speed to start with, looking forward to see how well it performs.
AMD's next-gen Kaveri chip due in January
More here: AMD Kaveri APU Launch Details: Desktop, January 14th
More here: AMD Kaveri APU Launch Details: Desktop, January 14th
Last edited:
Soldato
This is a review of the HD7750 GDDR3:
http://ht4u.net/reviews/2012/msi_radeon_hd_7750_2gb_ddr3_test/index36.php
The core and RAM runs at 800MHZ.
Trinity with 2133MHZ RAM is quite close to an HD6670 GDDR3 on average,and with 1600MHZ GDDR3 RAM is closer to an HD6570 GDDR3.
IIRC,the HD6670 and the HD6570 GDDR3 cards HT4U used,have slightly faster 900MHZ GDDR3 IIRC. Hence with 1600MHZ GDDR3 you are looking at least a 30% speedup over Trinity by my estimates,perhaps higher with faster RAM and in newer games. With the suspected improvements in the memory controller and memory handling,it probably would be more overall IMHO.
Edit!!
They might be using a newer version of GCN which the HD7750 does not.
Soldato
Seems like a great budget box base to run things like SteamOS on or a youngsters do-it-all PC!
Was this from the developer summit?
Was this from the developer summit?
Soldato
http://hothardware.com/News/AMD-Ann...s-at-APU13-Confirms-Kaveri-is-A-512Core-Part/
It seems AMD demoed the A10 7850K running the BF4 SP campaign at 1920X1080 on medium settings. It was running at 28FPS to 40FPS whereas the Core i7 4770K using an HD4600 IGP could only run the same sequence at 12FPS to 14FPS.
The A10 7850K runs at 3.7GHZ for the CPU and 720MHZ for the GPU.
Caporegime
28 to 40 FPS? Talk about a variance.
Interesting to see that 1080p gaming can be a reality on Kaveri however.
I know on my HD4600 (When I was waiting on my R9 290) I could game pretty well at 720p at low
Interesting to see that 1080p gaming can be a reality on Kaveri however.
I know on my HD4600 (When I was waiting on my R9 290) I could game pretty well at 720p at low
Soldato
Kaveri has a smaller die size than previous APUs:
http://img.new.livestream.com/events/000000000026c2ed/d7106387-20d6-40e2-94e4-4046bed391ce.png
Kaveri against a Core i7 4770K with a GT630:
https://www.youtube.com/watch?v=HjAM2zYNqko
It seems it was not the HD4600 being used but a GT630.
I did not properly read the article.
http://img.new.livestream.com/events/000000000026c2ed/d7106387-20d6-40e2-94e4-4046bed391ce.png
Kaveri against a Core i7 4770K with a GT630:
https://www.youtube.com/watch?v=HjAM2zYNqko
28 to 40 FPS? Talk about a variance.
Interesting to see that 1080p gaming can be a reality on Kaveri however.
I know on my HD4600 (When I was waiting on my R9 290) I could game pretty well at 720p at low
It seems it was not the HD4600 being used but a GT630.
I did not properly read the article.
Last edited:
Caporegime
Agh fair enough.
Not sure why they couldn't have just used one of their chips with the GT630.
Oh wait, of course we know why.
AMD's APU's are rendering entry level GPU's pretty moot to be honest (As to a lesser extent are Intel)
I wonder when we'll get to a point of having AMD's big chips with IGP's, I think that was the plan?
When I was waiting on my R9 290, I can't believe how useful even a HD4600 was, imagine a 7660D?
Not sure why they couldn't have just used one of their chips with the GT630.
Oh wait, of course we know why.
AMD's APU's are rendering entry level GPU's pretty moot to be honest (As to a lesser extent are Intel)
I wonder when we'll get to a point of having AMD's big chips with IGP's, I think that was the plan?
When I was waiting on my R9 290, I can't believe how useful even a HD4600 was, imagine a 7660D?
Caporegime
APU alone running BF4 at 1080P on medium settings is pretty impressive.
Whats the CPU performance?
Whats the CPU performance?
Soldato
It'll be interesting to see what kind of overclocking potential this thing has.
Looking forward to release date.
Looking forward to release date.
Caporegime
Trinity's IGP overclocked pretty well to be fair, hopefully this is the same.
Yeah, on a 6800K you'd not be able to play at 1080p on low settings, so it's a big leap. I think that performance looks to be around a 7770.
Caporegime
There's some youtube vid's of people playing BF3 1080p with the 6800K.