![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
amd 3rd Generation FX Questions
- Thread starter maxius
- Start date
that's a fair question. one i had to think about.
i think it really depends on whether they integrate pci-e (and other functions) on the server/desktop chips - this would require a new socket.
we don't really know what steamroller looks like yet so it's impossible to speculate about what an APU might contain. don't forget that 8 cores currently gives marginal gains for desktop use.
i think it really depends on whether they integrate pci-e (and other functions) on the server/desktop chips - this would require a new socket.
we don't really know what steamroller looks like yet so it's impossible to speculate about what an APU might contain. don't forget that 8 cores currently gives marginal gains for desktop use.
octoberasian
2[H]4U
- Joined
- Oct 13, 2007
- Messages
- 4,082
Someone linked this in the other thread:
http://www.pcper.com/reviews/Editor...nd-New-Design-Philosophy-Dictates-Faster-Pace
I highly recommend you read it. It should answer some of your questions.
From everything I've read up to now, it is pointing towards a new socket. Some sites say G34 or C32 sockets from their server processors or a new AM4 socket. I don't know about an FM3 socket ever being mentioned.
If going by PCPer's article, AMD is moving towards getting their APU from mobile to entry/mid-range desktop to eventually high-end/enthusiast and server use. The next APU, Kaveri, after Trinity will have a GCN-based GPU on the same die as the CPU.
Imagine an AMD FX processor in 2013 or 2014 with improved Piledriver modules paired with the next generation GCN cores from their GPU products. And, imagine an AMD Opteron CPU based around that as well. That's where AMD is headed.
And, with all intents and purposes, it looks like they will use an entirely new socket.
ManofGod
[H]F Junkie
- Joined
- Oct 4, 2007
- Messages
- 12,864
topcaliber
n00b
- Joined
- Jun 26, 2012
- Messages
- 50
Haven't used an AMD in such a long time. but that might change soon.
It would not make sense to drop AM3+. Though I do feel FM1 was a mistake and that a unified socket will be on the cards down the road.
Personally I find the integration of GPU's useful for budget builds, quite unimportant for mid to higher end builds.
I avoided FM1 for builds sticking to AM2+/AM3 boards which were cheaper and more cost effecitive overall. FM1 was a time limited dead end socket personally I think AMD should have made one unified socket and just wrung a bit more beef out of AM3 rather than 2 sockets.
Personally I find the integration of GPU's useful for budget builds, quite unimportant for mid to higher end builds.
I avoided FM1 for builds sticking to AM2+/AM3 boards which were cheaper and more cost effecitive overall. FM1 was a time limited dead end socket personally I think AMD should have made one unified socket and just wrung a bit more beef out of AM3 rather than 2 sockets.
Pieter3dnow
Supreme [H]ardness
- Joined
- Jul 29, 2009
- Messages
- 6,784
Read that somewhere too but are AMD far enough ahead to make their main desktop APU into something which works well enough for the current market (cpu).
Or does it mean a larger design, which would mean worse yields...
If Steamroller is such a product what would be it's target audience?
ITSTHINKING
[H]ard|Gawd
- Joined
- Mar 1, 2006
- Messages
- 1,260
I am just going to wait, I have a feeling that this next round of processors are going to impress. Maybe it's because it seems like every time AMD has there back against the wall and everyone is all but calling them dead, they hit a home run (in the gpu department i'd say the hd3000 and 4000 series came out of frickin nowhere after the VERY disapointing 2000 series) and we all know about the glorious amd64 days (still have a pc rocking a 1st gen fx and a 9600pro i built for hl2).
It's unlikely to be much larger. The 28nm production will be bulk and probably TSMC/GloFo and they generally have very good transistor density for a given node (or half node in this case). If they do an all-APU lineup then they could potentially decrease the amount of cache thus lowering the size even more. The Bulldozer chips were big partially because of the excess amounts of L3 and L2 cache and this was mainly due to it being a server chip -- that much slow cache is unnecessary on the desktop. Decreasing the cache size (hopefully the L2 per module) and going bulk should help. Bear in mind that they'd still be behind Intel who's already sitting on 22nm
It isn't the first time AMD has gone bulk for a CPU/APU either. Brazos was made at TSMC's 40nm
Pieter3dnow
Supreme [H]ardness
- Joined
- Jul 29, 2009
- Messages
- 6,784
That all sounds nice in theory but look at what happened to Bulldozer, we still haven't heard what exactly went wrong on the production process. If Steamroller runs into the same problems with the die shrink and not enough yields or fast enough for consumers it is another sinking ship.
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,757
AMD did a node and architecture change with Bulldozer at 32nm. So far, it's looking like Steamroller will simply be just a node change with minor architecture tweaks. Most of the major architecture changes are being done with Piledriver. Mostly just speculation on my part though.
AMD did announce they are going on a sort of tick-tock plan, similar to Intel. However, instead of an architecture and node change being the ticks and tocks, the release of the APUs and mainstream CPUs are the ticks and the tocks.
AMD did announce they are going on a sort of tick-tock plan, similar to Intel. However, instead of an architecture and node change being the ticks and tocks, the release of the APUs and mainstream CPUs are the ticks and the tocks.
Nobody knows for sure but hey that's life.
I've got tired of waiting and with s/h prices too high on PhII x6 processors went ahead and ordered an FX6100, AMD have a cashback running right now too. The overall cost is similar to what I paid for the 840 processor.
The biggest problem with FX was not it was rubbish (inconsistent no question), but that the prices were simply not realistic at all. All the models were overpriced at launch FX8150 was over £200..that was not sensible, the 6100 was £150 odd which again was too high, even the 4100 was priced at a point where it made you think twice about it. I'll be happy to share my views with the forum once the CPU arrives. If I think it sucks I'll pass that on..or maybe it's a bargain CPU that is at a nice price point?
Let's consider this this 6100 will cost me less than an Intel i-3, and for heavy multi threaded applications it's not likely to be a poor investment
I've got tired of waiting and with s/h prices too high on PhII x6 processors went ahead and ordered an FX6100, AMD have a cashback running right now too. The overall cost is similar to what I paid for the 840 processor.
The biggest problem with FX was not it was rubbish (inconsistent no question), but that the prices were simply not realistic at all. All the models were overpriced at launch FX8150 was over £200..that was not sensible, the 6100 was £150 odd which again was too high, even the 4100 was priced at a point where it made you think twice about it. I'll be happy to share my views with the forum once the CPU arrives. If I think it sucks I'll pass that on..or maybe it's a bargain CPU that is at a nice price point?
Let's consider this this 6100 will cost me less than an Intel i-3, and for heavy multi threaded applications it's not likely to be a poor investment
Though fab constraints and yield issues will certainly plague everyone on a fresh node at some point (even Intel had problems at 22nm), the difference this time around is that AMD is at least able to use competitors' fabs to produce their products. If GloFo can't produce then they'll go elsewhere, granted the transition can take 6 months to a year it's nevertheless lost business for GloFo so there's certainly incentive for them to step up their game.
Regardless of what AMD is brewing up they'll be at least a half node behind Intel at minimum. Still, their current position is certainly more favorable to last year's much more precarious one.
octoberasian
2[H]4U
- Joined
- Oct 13, 2007
- Messages
- 4,082
If you look at it this way: Many of Intel's processors have an integrated GPU from the low end to the high mid-range, from the mobile to the desktop, except for their Socket 2011 series. However, their integrated GPUs are not powerful enough to do much for both games and server applications. (Ok, maybe some games can run decently if you turn settings down low enough. Intel is getting better at it but they're like at the level of maybe Radeon HD 2000 or 3000 series while AMD is at HD 6000 and 7000 series in terms of graphics technology. I know, it's probably an exaggeration.)
Enter AMD.
AMD has covered the mobile and desktop low end, entry-level to mid-range processors with their APUs such as Brazos, Llano and Trinity. It has been AMD's intention since Fusion to merge both the GPU and CPU onto a single die. This is where HSA, or heterogenous systems architecture, comes into play, which was their Fusion Systems Architecture.
http://blogs.amd.com/fusion/2012/01...re-is-now-heterogeneous-systems-architecture/
http://blogs.amd.com/fusion/2012/06/12/hsafoundation/
Source.
The idea is that you don't need a large graphics card to do your compute and other mathematical calculations or stuff like OpenCL. Why not do it on a low powered solution right on the same die as the CPU? This will make servers efficient in my opinion based on what I've read so far. Trinity is a better indication of that direction where AMD is headed-- a better GPU, memory subpaths for the GPU and CPU and a better processor core than Llano. In the future, AMD will have their GPU products access the same memory space as the CPU in a unified manner.
Source.
AMD's Trinity on the mobile side, and going by articles online, has better battery life and performance-per-watt advantage over Intel's processors. You get a very good mobile GPU and decent performance even if it doesn't match Intel's mobile Ivy Bridge and Sandy Bridge, and is more affordable.
(http://techreport.com/articles.x/22932/13.)
Same sentiment has been mentioned by Semi-Accurate here.
So, when you ask if there is a target audience for this or if there will be worse yields, well, time will tell. Like pelo said above, even Intel had some issues moving to 22nm but resolved them quickly. And, AMD has had their issues as well but since AMD is a fabless company, if GloFo fails them then they have TSMC to fall back on. Intel doesn't have that luxury if any of their fabrication sites fails since Intel produces and manufacturers their own processors. However, Intel has a node advantage and more advanced fabrication techniques. Heck, Intel is going to go use 3D transistors and better materials in their processors.
Who would be a target audience then if AMD does go through with integrating GPUs with their Steamroller modules? Well, first thing, Kaveri, the next APU after Trinity and using Steamroller modules, is already rumored to use Radeon 8000-series GPU technology skipping 7000-series GCN tech. (Note: Trinity may have 7000-series named GPU, but they're still using the older VLIW4 cores from the 6000-series. This is like what happened with Llano-- VLIW5 cores from the 5000-series but were labeled Radeon 6000-series.)
A low powered, high-performance server processor with the ability to do HPA and compute applications on a single processor is going to be very tempting to many companies who would need that kind of flexibility. Combine this with the fact that AMD will allow their GPUs to execute C++ code alongside OpenCL, and a unified memory access on a future APU and you have a very palatable product.
For the consumers, a desktop processor with a GPU that can play games at a better performance than Intel's at a higher resolution than their lower end APUs such as Trinity, while still being very good at everyday computer tasks, and still be more affordable than Intel's without the need of a dedicated GPU and you'll start to see that this kind of high-performance midrange computer is going to go well with consumers.
To give you an idea, I was helping someone trying to find parts for a computer system that can run Final Fantasy XI without being very costly. Now, FFXI can run on an Intel graphics chip, but it is very dependent on what kind. Intel GMA 965 with a driver update can run it. Mobile Intel X3100, Desktop Intel HD 3000 and HD 4000 should run it decently but not great. Many settings will have to be turned down.
This is the lowest (technically) I can get a full Intel-based system with an Intel HD 4000 graphics. (The Intel 3400-series with an HD 4000 graphics is unfortunately not available at Newegg at the time I did this.)
http://ffxi.mmo-notepad.com/img/2012-06-27_22-15-16.png
$649 before shipping and any applicable tax.
$649 before shipping and any applicable tax.
With Intel HD 3000, which I wouldn't recommend because you'll have to lock the FFXI resolution to 1280x800 or 1024x768 resolution for it to be playable. 1440x900 might work and 1280x1024 as well, but I have not seen or read FFXI run on anything higher than that resolution on an Intel HD graphics chip. The CPU performance is still better than the APU but the graphics part sucks.
http://ffxi.mmo-notepad.com/img/2012-06-27_20-55-38.png
$519 before shipping and tax.
$519 before shipping and tax.
Now, let's switch it to a Llano APU.
http://ffxi.mmo-notepad.com/img/2012-06-27_20-04-12.png
$508 before shipping and any applicable tax.
$508 before shipping and any applicable tax.
Now imagine a desktop processor at AMD's estimated "33% faster than Bulldozer" with a graphics core based on the GCN architecture that's faster than Intel's HD graphics, that comes in at price $100 to $150 less than Intel. Yet, the computer is fast enough to play games very good at higher resolutions, maintain high framerates, and can do everyday tasks without the need of a dedicated GPU.
That is where Steamroller comes in if AMD does decide to integrate their GPU onto their higher end desktop processors such as the FX-series. A move to the 28nm process will help in this area as the die should be no larger than AMD's 32nm Trinity while still hopefully being at 125W TDP.
We'll just have to wait and see now because Steamroller-based FX-series is still two years away, while the APU based on those new modules is a year away.
AMD won't be able to use their APUs for HPC purposes until they fix their bus width issues.
This thread is worth a read, if only for 2-3 pages of it and disregarding the AVX2 talk.
My post (pelov) shows the pictures and the architectural issues AMD has to deal with in seeing their HSA agenda becoming free of potential bottleneck issues that they're already facing. Given current implementations, there isn't much to hope for in Trinity for HPC other than for low-grade tasks where you could save money and avoid Teslas+Xeons. Remember that bus width, GDDR5 (ECC version) and perf-per-watt and cost are what govern the sales and success for HPC products. Trinity still relies on DDR3 at 1866 where it's already bottlenecked by an onion and garlic bus width. It isn't wide nor fast enough for HPC tasks.
I guess for HPC it's a bit easier but it would require a new IMC on Trinity otherwise it's a nonstarter. The GFLOPs Trinity is capable of isn't good enough to justify putting together a new IMC and a dedicated PCB + 4 sticks of ECC high speed DDR3. Despite Trinity being cheaper, the RAM cost goes up and the rack space I doubt would change (unless SeaMicro got creative).
AMD really needs to address their reliance on DDR for their stores. It's not fast enough, it's too far away and it takes up too much space.
This thread is worth a read, if only for 2-3 pages of it and disregarding the AVX2 talk.
My post (pelov) shows the pictures and the architectural issues AMD has to deal with in seeing their HSA agenda becoming free of potential bottleneck issues that they're already facing. Given current implementations, there isn't much to hope for in Trinity for HPC other than for low-grade tasks where you could save money and avoid Teslas+Xeons. Remember that bus width, GDDR5 (ECC version) and perf-per-watt and cost are what govern the sales and success for HPC products. Trinity still relies on DDR3 at 1866 where it's already bottlenecked by an onion and garlic bus width. It isn't wide nor fast enough for HPC tasks.
I guess for HPC it's a bit easier but it would require a new IMC on Trinity otherwise it's a nonstarter. The GFLOPs Trinity is capable of isn't good enough to justify putting together a new IMC and a dedicated PCB + 4 sticks of ECC high speed DDR3. Despite Trinity being cheaper, the RAM cost goes up and the rack space I doubt would change (unless SeaMicro got creative).
AMD really needs to address their reliance on DDR for their stores. It's not fast enough, it's too far away and it takes up too much space.
Last edited:
octoberasian
2[H]4U
- Joined
- Oct 13, 2007
- Messages
- 4,082
Yeah, I've gotten a hint of that when reading articles from SemiAccurate and AnandTech regarding Trinity.
My hope is that this gets fixed when AMD moves towards a more unified memory architecture with a better memory bus. Maybe Kaveri APU will fix it.
Maybe. But, we're still a year away from seeing any Kaveri APUs to know if this was improved upon. They've already ditched HyperTransport on the Trinity APU and went with PCI-E as the interconnect bus. They just need to improve the memory part.
Increasing bus width is costly and adding channels is also quite costly and neither of those would purge them of their problem but rather just alleviate the issue momentarily. Whatever Kaveri brings, I'm certain it won't be that creative step they need but rather just an incremental widening to provide for a moderate performance gain (unified memory address space should help quite a bit with GPGPU tasks, though). After all, it's still on the FM2 socket meaning whatever we do get won't increase the pin count nor will be it be a drastic change.