AMD CPU speculation... and expert conjecture

Page 327 - 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.
The CPU component of the SB series (i5) is significantly stronger then the CPU component of the BD (FX/APU) series chips. The reason your seeing those type of benchmark is that they are using the iGPU to do the calculations vs just the CPU, AMD's iGPU's pretty much destroy Intel's. Intel is catching up though, but they are still years away not only in the hardware but in the software driver stack too. If you were running pure CPU benchmarks then the Intel ones tend to jump ahead.

Also everyone, cease referencing hafijur or anything he says in PMs. Don't call people names, don't ridicule, don't give me anymore work then I already have to do.
 
for fresh new speculations!!
AMD hints at faster, more power-efficient Xbox, PS4 chips
http://www.techhive.com/article/2056220/amd-hints-at-faster-more-powerefficient-xbox-ps4-chips.html

AMD to Tape Out First 20nm, 14nm FinFET Chips Within Next Two Quarters.
http://www.xbitlabs.com/news/other/display/20131017231002_AMD_to_Tape_Out_First_20nm_14nm_FinFET_Chips_Within_Next_Two_Quarters.html
AMD Excavator Core May Dramatic Performance Increases.
AMD Excavator’s AVX2 Support Hits at New FPU
http://www.xbitlabs.com/news/cpu/display/20131018224745_AMD_Excavator_Core_May_Dramatic_Performance_Increases.html

 
For all my life I was an Intel customer, starting from Pentium 200 MMX, Pentium 2 350 Mhz, Pentium 3 700, Core2Duo E6600, Core2Duo E7500 and now Core I7-4770.

I want AMD to succeed and kick Intel's butt, but right now AMD simply does not have "it" in the high end range, which is my point of interest.

I am a simple man, I go with what I have now and right now Intel's I7 is a king, while AMD's Steamroller is pretty much a vaporware, they can claim that it can power the damn NASA supercomputers in their papers and slides for all I care, cause they are not here yet for desktop high end and it seems they won't be here even 1 year from now.

That is sad, because this makes Intel too comfortable for my taste. I am very disappointed with Intel lately, because it seems like they did little for us, enthusiast-lite desktop guys in the last 2 years.
 


Nope - I think all mobile i7s have HT. Some however are just dual-core parts (just to make things more confusing). I haven't seen a mobile i7 without HT - could be wrong though.

The general rule seems to be:

mobile i3 = 2C, 4T, no Turbo Boost
mobile i5: = 2C, 4T, with Turbo Boost
mobile i7 = 4C, 8T, with Turbo Boost OR high-clocked 2C, 4T, TB

Here is what Intel says:

http://ark.intel.com/compare/64893,65712,65713,65714,71255,72054,64898,64900,65709,65710,65711,67356,71258,71458,71460,71670,72015,71459,64901,64899

 


I don't know what you consider "high end" but a cheap 8350 can compete with a 3770k if the software is up to date.

Steamroller vaporware? LOL You seem to have missed that AMD is already _shipping_ Kaveri to its channel. Or do you believe that partners have ready the Kaveri enabled mobos using waporware?

With HSA enabled software Kaveri will be miles ahead of your 4770.
 


what is hsa ?

Do I need a fm2 motherboard for kaveri/steamroller ?

when are they being released ?



 

HSA is a combined CPU+iGPU architecture, basically = IRMOARSPEEDZ
For HSA, you need an APU, ala FM2+. Steamroller will likely come out on AM3+ with moar cores, but no IGPU.
@palladin, "CPU" being per core?
 


DID EVERYONE JUST COMPLETELY OVERLOOK THIS POST?!?

Lol, this is a HUGE bit of info! I can't wait!!! I can hold off on buying a new CPU.
 

Just get a 1055T or 970BE and OC it to hold you off 😛. Perhaps we could get 20nm KAYVERHEH and HEFF HEX products in 2014.
http://www.ebay.com/itm/AMD-Phenom-II-X6-1055T-2-8-GHz-Six-Core-HDT55TFBGRBOX-Processor-/221301031221?pt=CPUs&hash=item338691c135
or
http://www.ebay.com/itm/AMD-Phenom-II-X4-970-BE-DeskTop-CPU-AM3-938-HDZ970FBK4DGR-HDZ970FBK4DGM-3-5Ghz-/140750182871?pt=CPUs&hash=item20c55d59d7
or
JUST GET THE 8320 BEFORE YOU CHANGE YOUR MIND GOSHDARN IT 😀
 


Lol, I'm waiting for the Christmas Bonus! So I've got plenty of time. And even if SR doesn't meed expectations, I should be able to get a deal on a FX 8320.
 


HSA stands for Heterogeneous System Architecture. It is a new computer architecture developed by the HSA foundation (which AMD is founder member) for heterogeneous computing

Heterogenenous computing refers to computing by mixing different kind of processors each one optimized for specific tasks, but that work together in harmony as a whole

http://developer.amd.com/resources/heterogeneous-computing/what-is-heterogeneous-computing/

http://hsafoundation.com

You need a FM2+ motherboard for Kaveri. Kaveri will be released next month probably. FM2+ motherboards are already available.


HSA is about CPUs, GPUs, DSPs... You don't need a GPU for a HSA complaint system. And nowhere the HSA specification says that GPUs have to be internal.


HSA will be available on dGPUs and CPUs. AMD will release HSA dGPUs. Seattle CPU is also a fully HSA compliant chip according to AMD.



Will likely doesn't.
 


I'm not sure how I missed that, lol. I blame the forum sliders.

http://www.globalfoundries.com/technology/14XM.aspx
Reading GloFo 14nm goals, they are mentioning it can scale to "advanced servers". I'm hoping that translates to a big Opteron chip which would then trickle down to FX chips. It would be awfully nice to go from 32nm FX to 14nm FX. 8m/16c would be very doable with a decent die size.

A few possible scenarios that I'm playing with in my head:

1. There are two versions of Streamroller, HEDT will get version 2, which is based on 14nm GloFo, the 14nm GloFo will be for FX/Opteron. The 20nm TSMC bulk will be for mobile parts like Temash replacement and GPUs.

2. SR for HEDT/server has been completely cancelled and AMD will just skip to 14nm excavator on a new platform, probably released with HSA with dGPU as well as DDR4

3. (less optimistic) Fabrication processes only allow for mobile chips in the future so 32nm PD is the end of the line.

4. If SR for HEDT and server is cancelled, we might get lucky and see excavator released earlier than anticipated.

Just curious, but if GloFo will have 14nm, and it can clock better than TSMC bulk, why doesn't AMD make GPUs @ GloFo?

The foundary news is interest. Once again I've ran back to look at old rumors, and it looks like 2013/2014 was supposed to be 28nm, then 20nm, and now rumors of 14nm.

I am starting to think that AMD can not find a place to make big SR/EV and that's where the problem with it is, and AMD is going "OMG APUS!" because they only have fabs that can make those types of chips. I almost want to think that AMD is focusing on APU first because they can't make big, high clock chips anywhere besides GloFo @ 32nm and there's only so much you can do with a big chip like Orochi at 32nm.
 


These initial 14nm nodes are "Extreme Mobility". I wouldn't expect to see more than quad-Jaguar or quad-ARM on these nodes.

Steamroller/Excavator is likely out of the question.



Not likely. Even Intel is sticking to 22nm for their high end desktop platforms through 2014.



Exactly why they call it "Extreme Mobility".



If 2015 is earlier than expected then probably.



Track record. GF has no history of making 5B-6B transistor parts on these newer nodes. Where TSMC has made up to 7B transistor parts on their 28nm node.



Xbit is being extremely liberal in their headlines. AMD did not mention taping out at all, they said transitioning. Big difference. Meaning their design tool sets will be upgraded for 20/14nm cell libraries.

Interesting enough Broadcom just announced possibly the first 16nm FinFet part saying it would run up to 3Ghz. Most likely made at TSMC but they did not give a release date at all.

http://www.xbitlabs.com/news/cpu/display/20131017233012_Broadcom_Announces_Its_First_ARMv8_A_64_Bit_Server_Microprocessors.html
 
*Sigh* If only we had this in 2013....
board.jpg
 


I cannot give details, but the article could appear published in a famous tech site, because it is being considered a good estimation of Kaveri/Steamroller performance. Meanwhile, AMD refuses to comment on the same slide that they presented during the "programming kaveri" talk and doesn't confirm/deny if Steamroller module is a 3 ALU per core or not. :pfff:
 

Its always fun to have surprises 😛
 


http://www.globalfoundries.com/technology/14XM.aspx

Yeah, it is mobile oriented, but GloFo is making a lot of claims that it can scale to "advanced servers" and posting pictures of full racks.

The point I'm getting at is that it may end up like Intel's process. Intel is most definitely mobile-oriented in their fabrication technology. Notice how they keep improving power consumption while losing maximum overclocks? And they still release HEDT chips?

I'm not saying it's just because of the fabrication process, but I think you would be quite foolish to claim that Intel's 22nm is designed around performance first and power consumption later.

Perhaps AMD had two options:

1. Release SR at 32nm since GloFo 28nm wasn't going to come and have it be a modest increase without HSA features

2. Make drastic changes to the existing SR design that would take advantage of the fact that the chip would be going from 32nm to 14/20nm hybrid design. Meaning that the design choices made would be oriented towards increasing IPC and leaving the original Bulldozer design of low IPC high clocks behind. Of course, this would cause delays and it would take time to redesign the chip AND to wait for GloFo to get the process online

http://www.globalfoundries.com/technology/14XM-FAQ.aspx

GloFo is already claiming that tape-outs should be happening in 2013.

Also, yes, the xbit article is high on speculation. But they are also speculating that 14nm will only be used for mobile parts while GloFo is claiming it can scale to multi-core and advanced servers.

There is a lot of contradicting information and whatever side you pick probably depends on if you're more optimistic or pessimistic about AMD. I'm personally more optimistic but that's just me.
 


I have no doubt AMD is playing the best possible hand they can with the cards they were dealt.

That GF page says they were expecting deliveries of 20nm in 2H2012. We're a year beyond that and I don't think any 20nm part from GF has hit the market yet. Kaveri will likely be the first 28nm part.

AMD has sizable wafer commitments to GF and they're currently all 32nm. Something just doesn't add up there. Now GF is talking about 14nm. That's 3 deep in the pipeline of vaporware (28nm/20nm/14nm). Maybe I missed some press releases but has anyone found any chips GF made with these nodes?
 


http://www.tomshardware.com/news/TSV-LPM-3D-Global-Foundries,21815.html

They currently have 20nm wafer production up, it's just ramping slowly. The yields aren't mature yet, and so far as I know, only IBM is planning to use 20nm imminently, and it's all UTBB FD-SOI.

EDIT: Unless there's an ARM player using the 20nm ULP nodes for something under the radar that I missed.
 


Looks like one of the customers in that video are in a similar boat. Cavium was quoted talking about the 28nm process. The product was recently announced but doesn't look available yet (package TBD).

http://www.cavium.com/OCTEON-III_CN7XXX.html

It's a network processor. With up to 48 cores, of MIPS 64-bit variety, and quad DDR. Says it runs up to 2.5Ghz. That's quite a bit of processing power.


Another possibility is Qualcomm who mentioned they were working with 4 foundries for 28nm as TSMC couldn't handle their 28nm demand.
 


Yes and no, depends on how you define APU and CPU.

AMD is investing heavily into mixed processor's, otherwise known as HSA. They're current design's have a huge a$$ iGPU that doubles as a vector coprocessor bolted onto the side of every chip. That's the APU btw. Since all they need is the giant vector coprocessor it's theoretically possible to build a 3~4 module chip with a scaled down iGPU that only functions for HSA work (GPU components being stripped off). In that case it wouldn't be an "APU" anymore as it couldn't do graphics but it would still have the vector processing capabilities.

Also I caution everyone to take all "HSA" benchmark projections with a very large grain of salt. The vast majority of code is scalar, meaning HSA does jack sh!t for it. The workloads that can be done in vector or massively parallel (GCN can do regular integer operations) will get a very large boost from HSA. Any benchmark that only tests the parallel will dramatically inflate the worth of HSA.
 
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom