AMD CPU speculation... and expert conjecture

[noob2222]

Interesting reading, seems to confirm everything that I have said in this thread:

http://www.pcper.com/reviews/Editorial/AMDs-Processor-Shift-Future-Really-Fusion/Future-Still-Fusion

No Steamroller for AM3+, only legacy support for 8000/9000 series.

AMD will continue selling 8000/9000 series before releasing top high-performance APUs. No 4M APU for 2014, 3M APU still possible in late 2014 (but unlikely).

Warsaw will be substituted by Excavator version of Berlin.

HyperTransport obsolete.

2/4 cores for mainstream desktop/laptop. 6/8 cores for enthusiast desktops. 4/8/12/16 cores for servers.

Full transition to APUs by about 2015.

Some more predictions/thoughts that the article doesn't consider:

ARM APU for 2016

Move to ARM (from x86) for 2018-2020.

your looking for any excuse to cling to your assumption that AMD is going to be a low-end only company from now on. You should be praying that your wrong instead of praising the move to VIA status.

Then again, maybe AMD is trying to make a gpu-only computer with a core clock of 1.1 ghz.

 

[gamerk316]

Intel Dual without HT is barely good for single player, useless for multiplayer.

Intel Dual with HT is kind of playable in multiplayer and works fine in single.

Intel Dual without HT is barely better then the FX-6350 in singleplayer. Works both ways. Mutli, obviously, favors more cores, especially on 64-player maps. But for singleplayer, the i3 competes against FX. That's AMD's problem in a nutshell: The i3 is good enough, even without HT, to make the FX-4xxx lineup essentially DOA, and those are supposed to be the mainstream chips. You can't live only on high end parts.

Hence the pivot: You can't make low-end FX chips that compete against low-end Intel, and low-end is where the bulk of the revenue is.

 

[esrever]

Intel Dual without HT is barely good for single player, useless for multiplayer.

Intel Dual with HT is kind of playable in multiplayer and works fine in single.

Intel Dual without HT is barely better then the FX-6350 in singleplayer. Works both ways. Mutli, obviously, favors more cores, especially on 64-player maps. But for singleplayer, the i3 competes against FX. That's AMD's problem in a nutshell: The i3 is good enough, even without HT, to make the FX-4xxx lineup essentially DOA, and those are supposed to be the mainstream chips. You can't live only on high end parts.

Hence the pivot: You can't make low-end FX chips that compete against low-end Intel, and low-end is where the bulk of the revenue is.

the pentium makes the i7 worthless too if you are using that kind of reasoning.

 

[noob2222]

Intel Dual without HT is barely good for single player, useless for multiplayer.

Intel Dual with HT is kind of playable in multiplayer and works fine in single.

Intel Dual without HT is barely better then the FX-6350 in singleplayer. Works both ways. Mutli, obviously, favors more cores, especially on 64-player maps. But for singleplayer, the i3 competes against FX. That's AMD's problem in a nutshell: The i3 is good enough, even without HT, to make the FX-4xxx lineup essentially DOA, and those are supposed to be the mainstream chips. You can't live only on high end parts.

Hence the pivot: You can't make low-end FX chips that compete against low-end Intel, and low-end is where the bulk of the revenue is.

I can tell by the way you talk about this, you have never played BF3. Single player is nothing, only put in the game to get new players used to some of the controls. You don't get to fly, (although you do get to be the co-pilot in a jet once), you get to drive a tank once, but 99% of single player is nothing but walking and shooting NPCs with various weapons. Why does single player not take any resources? its all scripted, no AI to speak of and its just you with no destruction.

Who buys BF3 for single player? NO ONE. why is bf3 single player performance so important? ITS NOT EVEN RELEVANT, all its good for is a GPU guideline for what your MP FPS goal is. IT HAS ALMOST NOTHING TO DO WITH THE CPU or even the actual game for that matter.

Might as well have a single player world of warcraft benchmark, would be just as relevant.

 

[8350rocks]

Intel Dual without HT is barely good for single player, useless for multiplayer.

Intel Dual with HT is kind of playable in multiplayer and works fine in single.

Intel Dual without HT is barely better then the FX-6350 in singleplayer. Works both ways. Mutli, obviously, favors more cores, especially on 64-player maps. But for singleplayer, the i3 competes against FX. That's AMD's problem in a nutshell: The i3 is good enough, even without HT, to make the FX-4xxx lineup essentially DOA, and those are supposed to be the mainstream chips. You can't live only on high end parts.

Hence the pivot: You can't make low-end FX chips that compete against low-end Intel, and low-end is where the bulk of the revenue is.

I can tell by the way you talk about this, you have never played BF3. Single player is nothing, only put in the game to get new players used to some of the controls. You don't get to fly, (although you do get to be the co-pilot in a jet once), you get to drive a tank once, but 99% of single player is nothing but walking and shooting NPCs with various weapons. Why does single player not take any resources? its all scripted, no AI to speak of and its just you with no destruction.

Who buys BF3 for single player? NO ONE. why is bf3 single player performance so important? ITS NOT EVEN RELEVANT, all its good for is a GPU guideline for what your MP FPS goal is. IT HAS ALMOST NOTHING TO DO WITH THE CPU or even the actual game for that matter.

Might as well have a single player world of warcraft benchmark, would be just as relevant.

+1 BF3 SP campaign can be completed, in it's entirety, in under 15 hours.

No one buys BF3/BF4 for single player...it isn't enough content to justify the price. Now if we were talking about Ghost Recon or Rainbow Six titles...that's another matter. However, CoD and BF series are all about MP and that's the reason people buy it now.

 

[griptwister]

Lol, here's a intersting fact for you guys. In BF3, Linus was running a bench mark with a dual core CPU and a Quad Core CPU. And it turns out that in one of the scenes, a guy was missing from the Dual Core benchmark... But again, go back to quad core, and he's back. Lol.

 

[noob2222]

@8350, if you skip the cutscenes, I went through it a 2nd time just to see how fast ... I think it was ~1 1/2 hrs maybe.

 

[blackkstar]

Intel Dual without HT is barely good for single player, useless for multiplayer.

Intel Dual with HT is kind of playable in multiplayer and works fine in single.

Intel Dual without HT is barely better then the FX-6350 in singleplayer. Works both ways. Mutli, obviously, favors more cores, especially on 64-player maps. But for singleplayer, the i3 competes against FX. That's AMD's problem in a nutshell: The i3 is good enough, even without HT, to make the FX-4xxx lineup essentially DOA, and those are supposed to be the mainstream chips. You can't live only on high end parts.

Hence the pivot: You can't make low-end FX chips that compete against low-end Intel, and low-end is where the bulk of the revenue is.

The FX 4000 line will be discontinued, I can promise you that. It has no reason to exist. For like $10 more you can get 50% more cores. And it actually hurts AMD's plans for HSA adoption, as people who buy FX 4000 series are buying into a platform which doesn't support HSA as opposed to an APU that has HSA. FX 4000 series is directly competing with APUs while FX 4000 series harms AMD's HSA push.

What the hell are you trying to say? Are you implying that there's also no difference between Intel Dual, Intel Quad and Intel Hex as an i3 is just fine in a single core game?

You remind me of the people I used to talk to before dual core CPUs existed. I said back then "it'd be so cool to have a dual CPU system" and they all laughed, going "nothing will use both CPUs it's a total waste and it always will be."

Now, if you build a single core CPU system now, you're a joke. Coincidentally, at that very same time, Intel was talking about 10ghz Nehalem Pentium chips (the old Nehalem that got cancelled, not LGA1366) and there was a massive wall of FUD on the internet about how multiple cores was a complete waste and we'd have 10ghz Pentiums soon.

And if you remember correctly, AMD made multiple core CPUs back in the day, and now everyone has a CPU with more than one core. Christ, you have to dig pretty deep to buy a CELLULAR PHONE with a single core.

I've lived through this so many times already. "TWO CORES IS STUPID", "FOUR CORES IS STUPID", "8 CORES IS STUPID". How many times are we going to go through this before you realize that this is the only way to scale x86 CPUs now and that single thread performance gains are DEAD. You're basically advocating that the past is going to be different this time because of reasons that everyone else said in the past and were wrong about. And you're probably all frothing at this ready to go "BUT IM A GAME DEVELOPER!!!! ZOMG LOOK IM AN AUTHORITY ON THE SUBJECT!!!!"

You're a game developer, for (*&#$'s sake. Go make some toys while the computer scientists develop algorithms to solve the problems you can't comprehend solutions to.

AMD isn't the only company going wide with MOAR COARS! Samsung is. Intel is. IBM is. It's coming whether you like it or not and complaining that "wah wah wah muh games don't scale" or "wah wah Crytek moved computation off of the performance bottleneck and moved it to something not utilized THATS SO STUPID!!!!!!" isn't going to change a thing.

I mean really, you whine all day long about how it's IMPOSSIBLE for all this sharing of memory to happen, and look: http://en.wikipedia.org/wiki/Transactional_Synchronization_Extensions

It's already been solved.

SO please, just go. You're the type of person that would defend Intel single core Pentium hitting 10ghz to the death and nothing would have changed your mind.

Also, just curious there buddy, but what CPU you running? An i3 I take it, right? Because all those extra cores are a waste. Right? That's the point you're trying to so ardently make.

 

[The Q6660 Inside]

Intel Dual without HT is barely good for single player, useless for multiplayer.

Intel Dual with HT is kind of playable in multiplayer and works fine in single.

Intel Dual without HT is barely better then the FX-6350 in singleplayer. Works both ways. Mutli, obviously, favors more cores, especially on 64-player maps. But for singleplayer, the i3 competes against FX. That's AMD's problem in a nutshell: The i3 is good enough, even without HT, to make the FX-4xxx lineup essentially DOA, and those are supposed to be the mainstream chips. You can't live only on high end parts.

Hence the pivot: You can't make low-end FX chips that compete against low-end Intel, and low-end is where the bulk of the revenue is.

The FX 4000 line will be discontinued, I can promise you that. It has no reason to exist. For like $10 more you can get 50% more cores. And it actually hurts AMD's plans for HSA adoption, as people who buy FX 4000 series are buying into a platform which doesn't support HSA as opposed to an APU that has HSA. FX 4000 series is directly competing with APUs while FX 4000 series harms AMD's HSA push.

What the hell are you trying to say? Are you implying that there's also no difference between Intel Dual, Intel Quad and Intel Hex as an i3 is just fine in a single core game?

You remind me of the people I used to talk to before dual core CPUs existed. I said back then "it'd be so cool to have a dual CPU system" and they all laughed, going "nothing will use both CPUs it's a total waste and it always will be."

Now, if you build a single core CPU system now, you're a joke. Coincidentally, at that very same time, Intel was talking about 10ghz Nehalem Pentium chips (the old Nehalem that got cancelled, not LGA1366) and there was a massive wall of FUD on the internet about how multiple cores was a complete waste and we'd have 10ghz Pentiums soon.

And if you remember correctly, AMD made multiple core CPUs back in the day, and now everyone has a CPU with more than one core. Christ, you have to dig pretty deep to buy a CELLULAR PHONE with a single core.

I've lived through this so many times already. "TWO CORES IS STUPID", "FOUR CORES IS STUPID", "8 CORES IS STUPID". How many times are we going to go through this before you realize that this is the only way to scale x86 CPUs now and that single thread performance gains are DEAD. You're basically advocating that the past is going to be different this time because of reasons that everyone else said in the past and were wrong about. And you're probably all frothing at this ready to go "BUT IM A GAME DEVELOPER!!!! ZOMG LOOK IM AN AUTHORITY ON THE SUBJECT!!!!"

You're a game developer, for (*&#$'s sake. Go make some toys while the computer scientists develop algorithms to solve the problems you can't comprehend solutions to.

AMD isn't the only company going wide with MOAR COARS! Samsung is. Intel is. IBM is. It's coming whether you like it or not and complaining that "wah wah wah muh games don't scale" or "wah wah Crytek moved computation off of the performance bottleneck and moved it to something not utilized THATS SO STUPID!!!!!!" isn't going to change a thing.

I mean really, you whine all day long about how it's IMPOSSIBLE for all this sharing of memory to happen, and look: http://en.wikipedia.org/wiki/Transactional_Synchronization_Extensions

It's already been solved.

SO please, just go. You're the type of person that would defend Intel single core Pentium hitting 10ghz to the death and nothing would have changed your mind.

Also, just curious there buddy, but what CPU you running? An i3 I take it, right? Because all those extra cores are a waste. Right? That's the point you're trying to so ardently make.

I have always thought that the FX4000 has no reason to exist, considering the FX6300 was in the same price range and had much better performance. *Thinks of excuse gamer is mashing up to justify his 2600K*

 

[noob2222]

I doubt they will "discontinue" it as its not a separate cpu die. Its a binned 8350 with 2 bad modules, cost to produce - $0 for the silicon (if you consider the alternative is scrap value at $0) and ~$1 to speed test it.

The reason it exists is because its still a useable piece of silicon that didn't make the cut to 6xxx or 8xxx.

In business there is 2 options. 1: sell it, or 2: junk it.

If you examine it per part, say an 8350 silicon die costs $50. you can sell the 8350 for ~200 for a profit of $150
you still paid for all the parts that didn't pass 8350 or 6300 tests

profit for selling 4xxx @100 ~50
profit for scrapping it, net loss of $50

your already going to be taking losses on every piece of silicon that doesn't pass the 4xxx tests, why increase your losses?

 

[juanrga]

I can guarantee ever CoD game this generation uses 6 cores on the consoles, likely at 100%. On the PC, 2. Same for every other title out there. I can not WAIT for Novemeber, because I will be doing nothing but posted "I TOLD YOU SO" when people find that graphical quality didn't jump, and we're still stuck on two main threads and maybe a worker thread or two.

To me, people are fooling themselves just like they did before the BD launch.

BF4, not still next gen, but already recommends to use a quad or a hexa core CPU

Just like every game released in the past 4-5 years? BF4 is going to perform about the same as BF3. Which is to say: GPU bound in singleplayer, with somewhat better scalability in multiplayer.

In fact, BF3 recommended a quad-core AMD CPU. BF4 recommends a six-core AMD CPU.

Interesting reading, seems to confirm everything that I have said in this thread:

http://www.pcper.com/reviews/Editorial/AMDs-Processor-Shift-Future-Really-Fusion/Future-Still-Fusion

No Steamroller for AM3+, only legacy support for 8000/9000 series.

AMD will continue selling 8000/9000 series before releasing top high-performance APUs. No 4M APU for 2014, 3M APU still possible in late 2014 (but unlikely).

Warsaw will be substituted by Excavator version of Berlin.

HyperTransport obsolete.

2/4 cores for mainstream desktop/laptop. 6/8 cores for enthusiast desktops. 4/8/12/16 cores for servers.

Full transition to APUs by about 2015.

Some more predictions/thoughts that the article doesn't consider:

ARM APU for 2016

Move to ARM (from x86) for 2018-2020.

No one is going to move exclusively to ARM, simply due to that thing called Windows.

First, where did you read the word "exclusively"? I cannot find it in anything that I wrote.

Second, Windows also runs on ARM. Nvidia expects a RT+ARM future.

Third, Windows is a minority OS and Microsoft has shown that no longer drives the PC market. AMD announced that abandon Windows exclusivity, and Intel has announced the same at IDC. I gave the link a pair of posts ago. Intel is joining with Google.

The future is Linux and ARM.

 

[esrever]

They will likely just rebrand the FX4XXX parts as athlons in the future. But I doubt even that any time soon. They seem to be working to unify their CPU socket to the APU side. Which means nothing will happen until AMD3+ is dead and probably FM3 comes out.

 

[juanrga]

i have a couple of questions about the jaguar and radeon cores in the new console socs:
when new nodes become ready, will sony and microsoft make the socs on new nodes e.g. 20nm/16nm/14nm?
with node shrinks, will they be able to clock the cores higher or the performance improvements will be more subtle or none at all?

i ask this after reading microsoft's publicity on squeezing out minor clock boosts on their final xbone(R) soc.

No. This would generate two consoles, one with slow clocks and other with faster clocks, which makes no sense for customers neither for developers, who would develop the game for the people with the slower clock.

- PC development ecosystem is based in continuous upgrade of specs, without optimization.
- Console development ecosystem is based in continuous optimization for fixed specs.

 

[juanrga]

Interesting reading, seems to confirm everything that I have said in this thread:

http://www.pcper.com/reviews/Editorial/AMDs-Processor-Shift-Future-Really-Fusion/Future-Still-Fusion

No Steamroller for AM3+, only legacy support for 8000/9000 series.

AMD will continue selling 8000/9000 series before releasing top high-performance APUs. No 4M APU for 2014, 3M APU still possible in late 2014 (but unlikely).

Warsaw will be substituted by Excavator version of Berlin.

HyperTransport obsolete.

2/4 cores for mainstream desktop/laptop. 6/8 cores for enthusiast desktops. 4/8/12/16 cores for servers.

Full transition to APUs by about 2015.

Some more predictions/thoughts that the article doesn't consider:

ARM APU for 2016

Move to ARM (from x86) for 2018-2020.

Umm ... you realize he's just stating what everyone else has been stating the last few months. It's like guy A talks to guy B who talks to guy C who then *informs* guy A about the "news" and guy A goes off and writes an article about it.

We all know that FM2+ will be the first SR chips in a 2M configuration, that isn't being debated. The question is about the mainstream performance sector @180~200 USD. A four core APU isn't going to cut it, neither is the current 8350 or the golder chip 9 series. They will need to release a SR module based desktop chip to satisfy this significant market segment, that module doesn't need a GCN based GPU on it at all. So the fact that they can't fit a GCN GPU on the same die as 3~4M is about as relevant as them not being able to fit a Southern Islands GPU on a die with 3~4M. Of course I'm expecting AM3+ to be phased out in favor of FM2+ as there is no need to maintain two different socket types.

Basically and AMD version of the below linked Intel 4+1 die configuration
http://wccftech.com/haswell-die-configurations-intel-ivy-bridge-revealed/

Just enough of a "iGPU / whatever" unit to function for HSA but no where near what's necessary for full blown graphics rendering. And with AMD's modular design they can do this much easier then Intel can.

In this same thread I can find people who still believes that AMD will release an octo-core FX steamroller for the AM3+ platform, when none is coming.

Steamroller comes as APU/CPU. We know that. I already mentioned that Berlin comes as APU and also as CPU. For the desktop I expect an Athlon CPU derived from Kaveri, but this will come as 2/4 core. There is no 6/8 core Steamroller CPUs for FM2+.

I also explained why "need" doesn't mean anything. Look to my discussion about FirePro. AMD did need PCIe3 and faster Opterons and none come to us. As a consequence, people used Firepro with the competence: Intel Xeons. Is this optimal? No, but provides a clever precedent to what some of you claim cannot happen.

The current 8000/9000 series provide enough performance for the >99% of users. Pretending that AMD has to release a 8-core FX Steamroller for the remaining <1% looks rather unrealistic. Moreover, thanks to the next gen of games being well-multithreaded the FX 6000/8000/9000 will perform much better than now. Users of those chips will see a huge boost in games, without changing the hardware, and new users will prefer those to the competence (Intel). I already provided a link to the poll where all game developers chose the FX-8350 as better gaming CPU instead of the i5-3570k.

All this makes irrelevant an update of the entire FX series.

I think you misunderstand the goal of an HSA APU. The iGPU is not for "full blown graphics rendering" but its main goal is for compute. In fact, a APU+dGPU configuration, where the iGPU is used for compute and the dGPU for graphics could be AMD secret master plan.

For some reason, people still believes that APU is about gaming in a budget, and that is very funny.

Interesting reading, seems to confirm everything that I have said in this thread:

http://www.pcper.com/reviews/Editorial/AMDs-Processor-Shift-Future-Really-Fusion/Future-Still-Fusion

No Steamroller for AM3+, only legacy support for 8000/9000 series.

AMD will continue selling 8000/9000 series before releasing top high-performance APUs. No 4M APU for 2014, 3M APU still possible in late 2014 (but unlikely).

Warsaw will be substituted by Excavator version of Berlin.

HyperTransport obsolete.

2/4 cores for mainstream desktop/laptop. 6/8 cores for enthusiast desktops. 4/8/12/16 cores for servers.

Full transition to APUs by about 2015.

Some more predictions/thoughts that the article doesn't consider:

ARM APU for 2016

Move to ARM (from x86) for 2018-2020.

Very old news with the same old (probably wrong) assumptions. "ITS WHAT EVERYBODY IS SAYING"

* NO more new uarch AM3+ upgrades is pretty clear, since even Warsaw will have a **NEW COMPATIBLE** socket (old G34 is EOL to)... and Warsaw is still PD.

* Hypertransport is not obsolete, even if all adapters will be PCIe, every CPU and APU since this far is a Hypertransport switch for the "internal" Xbar, and HSA IOMMU standard is based on HTT internals. Matter of fact Piledriver was supposed to debut this Xbar/HTT at 4Ghz version with better power management (AND HTT HAS MUCH BETTER POWER MANAGEMENT POSSIBILITIES THAN CRAP PCIe), which seems didn't happen. Perhaps new version of PD Warsaw, and PD FX can have it, even for AM3+, only this will make an upgrade worthed.

Hypertransport perhaps makes the same or more sense in GPGPUS than in CPUs... a MCM of GPUs is what is needed to crush Nvidia definitely without redemption and their monster chip approaches, it would be better in all aspects, and it will be single "chip", single socket SKUs. In this light i'm still hoping for a HTX+PCIe combo slots, though that probably end of 2014 or 2015.

* Everything APU, as in a chip with a GPU is simply NOT the best option for server... server needs GFLOPS not graphics... so a "middle" ground must be found. Else is the same of AMD abandoning server altogether... a very strange and penalizing move, considering that there is an Open Platform definition for server.

Hypertransport is obsolete and one of the reasons that Cray abandoned AMD. Cray is now using PCIe Gen 3 in its new HPC clusters. I have checked and there is no mention to Hypertransport in the HSA spec. Moreover, the substitute of the Opteron (the new Berlin and Seattle) come without Hypertransport. I don't know if Warsaw comes with Hypertransport, but if it comes, this is for legacy users.

Even AMD is using PCIe 3 in its FirePro series for the professional market.

Everything APU is the future for servers, precisely by the same reason that you mention: GFLOP. AMD has already showed some typical server workload where an APU is about 8x faster than a CPU. No strange that Nvidia is preparing its own APU for servers as well.

AMD is not abandoning anything in servers.

 

[juanrga]

Interesting reading, seems to confirm everything that I have said in this thread:

http://www.pcper.com/reviews/Editorial/AMDs-Processor-Shift-Future-Really-Fusion/Future-Still-Fusion

No Steamroller for AM3+, only legacy support for 8000/9000 series.

AMD will continue selling 8000/9000 series before releasing top high-performance APUs. No 4M APU for 2014, 3M APU still possible in late 2014 (but unlikely).

Warsaw will be substituted by Excavator version of Berlin.

HyperTransport obsolete.

2/4 cores for mainstream desktop/laptop. 6/8 cores for enthusiast desktops. 4/8/12/16 cores for servers.

Full transition to APUs by about 2015.

Some more predictions/thoughts that the article doesn't consider:

ARM APU for 2016

Move to ARM (from x86) for 2018-2020.

your looking for any excuse to cling to your assumption that AMD is going to be a low-end only company from now on. You should be praying that your wrong instead of praising the move to VIA status.

Then again, maybe AMD is trying to make a gpu-only computer with a core clock of 1.1 ghz.

That is only in your imagination.

 

[8350rocks]

Hypertransport is obsolete and one of the reasons that Cray abandoned AMD. Cray is now using PCIe Gen 3 in its new HPC clusters. I have checked and there is no mention to Hypertransport in the HSA spec. Moreover, the substitute of the Opteron (the new Berlin and Seattle) come without Hypertransport. I don't know if Warsaw comes with Hypertransport, but if it comes, this is for legacy users.

Even AMD is using PCIe 3 in its FirePro series for the professional market.

Everything APU is the future for servers, precisely by the same reason that you mention: GFLOP. AMD has already showed some typical server workload where an APU is about 8x faster than a CPU. No strange that Nvidia is preparing its own APU for servers as well.

AMD is not abandoning anything in servers, despite your insistence...

HT is currently an 12 year old technology...the man *who wrote the standard for it* also happens to be the man in charge of AMD CPUs.

That would be one Jim Keller.

The issue I have with what you're saying is basically that HTX is actually *still* faster than PCIe 3.0. You can even integrate the 2 so you can speed up PCIe 3.0. See here:

http://www.hypertransport.org/

It's also a MUCH better system wide interconnect system. Consider...HTX 3.1 allows 6400 MT/s...which is actually faster than SATA3. That's why IBM uses it, among many others. You have a one way throughput of 25.6 GB/s and dual ways of 51.2 GB/s. Nothing else out there gives you that capability.

You can also make it an expansion slot and that standard is called HTX which allows you to use an expansion card to communicate directly with the CPU.

HT is not dead...nowhere even close. It's too useful, and too versatile for that.

http://en.wikipedia.org/wiki/HyperTransport

 

[juanrga]

I have just finished the draft of my estimation of the performance of the Kaveri APU.

http://juanrga.com/en/AMD-kaveri-benchmark.html

I include some traditional benchmarks

And some HSA workload

The conclusion is that, CPU alone, the top Kaveri (4 Steamroller cores) is at the level of a Sandy Bridge or Ivy Bride i5. But with HSA enabled software we can see 3x the performance of an i5. Funny some people still pretends this is low-end.

 

[griptwister]

I have just finished the draft of my estimation of the performance of the Kaveri APU.

http://juanrga.com/en/AMD-kaveri-benchmark.html

I include some traditional benchmarks

And some HSA workload

The conclusion is that, CPU alone, the top Kaveri (4 Steamroller cores) is at the level of a Sandy Bridge or Ivy Bride i5. But with HSA enabled software we can see 3x the performance of an i5. Funny some people still pretends this is low-end.

That is only in your imagination.

:lol:

 

[8350rocks]

And some HSA workload

Funny some people still pretends this is low-end.

The issue lies in the fact that many applications for private sector use in media programs are not going to be HSA for quite some time to come likely. Additionally...if you're rendering, the HSA will be essentially useless as the GPU has to be doing something.

Again FX 8350 > Kaveri 2M APU for media and productivity.

Perhaps if you're an amateur using photoshop or something...that's one thing...but for people who are doing heavy workloads that require the raw, brute force, that is a low end solution.

 

[juanrga]

Hypertransport is obsolete and one of the reasons that Cray abandoned AMD. Cray is now using PCIe Gen 3 in its new HPC clusters. I have checked and there is no mention to Hypertransport in the HSA spec. Moreover, the substitute of the Opteron (the new Berlin and Seattle) come without Hypertransport. I don't know if Warsaw comes with Hypertransport, but if it comes, this is for legacy users.

Even AMD is using PCIe 3 in its FirePro series for the professional market.

Everything APU is the future for servers, precisely by the same reason that you mention: GFLOP. AMD has already showed some typical server workload where an APU is about 8x faster than a CPU. No strange that Nvidia is preparing its own APU for servers as well.

AMD is not abandoning anything in servers, despite your insistence...

HT is currently an 12 year old technology...the man *who wrote the standard for it* also happens to be the man in charge of AMD CPUs.

That would be one Jim Keller.

The issue I have with what you're saying is basically that HTX is actually *still* faster than PCIe 3.0. You can even integrate the 2 so you can speed up PCIe 3.0. See here:

http://www.hypertransport.org/

It's also a MUCH better system wide interconnect system. Consider...HTX 3.1 allows 6400 MT/s...which is actually faster than SATA3. That's why IBM uses it, among many others. You have a one way throughput of 25.6 GB/s and dual ways of 51.2 GB/s. Nothing else out there gives you that capability.

You can also make it an expansion slot and that standard is called HTX which allows you to use an expansion card to communicate directly with the CPU.

HT is not dead...nowhere even close. It's too useful, and too versatile for that.

http://en.wikipedia.org/wiki/HyperTransport

I already mentioned how Cray abandoned Hypertransport. I will add some more details:

When we [Cray] built the XT, XC, XK line [of clusters], the best way to talk to a processor was through AMD's proprietary HyperTransport mechanism. That's what we based our network on because there wasn't any better way to talk to a processor, so it started with the network and the Opteron happens to be a great processor. [In the XC30] it starts with the PCI [Express] Gen 3 because you need to have a more scalable network.

Our move to PCI [Express] Gen 3 was critical to having higher bandwidth in and out of compute nodes and currently Intel has the best PCI [Express] Gen 3 support, so it matches our network and the performance is good. Right now Intel's providing us with a PCI [Express] Gen 3 processor that can work with good data rates on our network.

Cray switched to Intel, because the AMD Opteron only provide PCIe Gen 2 support.

Their new XC30 cluster uses the Aries interconnect (which is based in PCIe 3) and substitutes the old Gemini interconnect (based in Hypertransport)

http://www.cray.com/Programs/Cascade.aspx

Note also that I didn't say that Hypertransport is dead. I said it is obsolete. It is obsolete for Cray, it is obsolete for me, it seems is obsolete for AMD (no Hypertransport on new Berlin Seattle). No?

 

[palladin9479]

WTF.....

Juanrga I know your passionate about this, but did you just fudge some numbers by taking speculation guesstimates and multiplying them by another fudge factor? If I ever recommended doing some sh!t like that I'd be laughed out of the room, or fired on the spot depending if they thought I was serious or not.

And you realize *need* is what drives markets and thus revenue? Sitting on a throne and dictating change doesn't work very well in a free market. Companies release products to feed the demand of the market, the demand of the market does not change based on what companies are supplying. AMD not making a mainstream DT chip will not make that market evaporate, they will simply go with an Intel offering instead and thus AMD loses market share and profits. This is how business works, I don't go around pissing on my customers because I think their demands are silly / stupid, instead I find a way to cater to those demands in a realistic fashion.

 

[GOM3RPLY3R]

It seems everything is pretty hectic here. How about a brain teaser and a poll?

Try to guess what's going to happen based on the events that are taking place by each company:

1. AMD will gain some ground.
2. Intel will gain more ground.
3. They will level out.


^_^

 

[juanrga]

And some HSA workload

Funny some people still pretends this is low-end.

The issue lies in the fact that many applications for private sector use in media programs are not going to be HSA for quite some time to come likely. Additionally...if you're rendering, the HSA will be essentially useless as the GPU has to be doing something.

Again FX 8350 > Kaveri 2M APU for media and productivity.

Perhaps if you're an amateur using photoshop or something...that's one thing...but for people who are doing heavy workloads that require the raw, brute force, that is a low end solution.

Maybe you are still unaware, despite being mentioned dozens of times, but AMD is replacing 8-core Opterons CPUs by 4-core Berlin HSA APUs in the servers. And no, those aren't products aimed to "amateurs"

Heterogeneous architectures will dominate the professional "media and productivity" market, the same like they are dominating the HPC market today.

WTF.....

Juanrga I know your passionate about this, but did you just fudge some numbers by taking speculation guesstimates and multiplying them by another fudge factor? If I ever recommended doing some sh!t like that I'd be laughed out of the room, or fired on the spot depending if they thought I was serious or not.

And you realize *need* is what drives markets and thus revenue? Sitting on a throne and dictating change doesn't work very well in a free market. Companies release products to feed the demand of the market, the demand of the market does not change based on what companies are supplying. AMD not making a mainstream DT chip will not make that market evaporate, they will simply go with an Intel offering instead and thus AMD loses market share and profits. This is how business works, I don't go around pissing on my customers because I think their demands are silly / stupid, instead I find a way to cater to those demands in a realistic fashion.

It doesn't matter how I got them, the important point is that there are some actual numbers and can be compared to the final silicon when ready.

At the end, the entire article will be compared to all those claiming that Steamroller has 256-bit FMAC units, SMT cores, 2.4Ghz maximum clocks, and similar nonsensical stuff posted in this thread by the 'experts'...

I already addressed the comment about "needs" before, even providing a real-world example explaining how your 'logic' fails, but it seems was missed or ignored again.

Finally, what I have been saying about the CPU/APU real market is also applicable to GPUs. AMD has stated just some few hours ago that the next-generation of GPUs, which will be presented in brief, "is targeting more of the enthusiast market versus the ultra-enthusiast one".

 

[palladin9479]

It doesn't matter how I got them, the important point is that there are some actual numbers and can be compared to the final silicon when ready.

At the end, the entire article will be compared to all those claiming that Steamroller has 256-bit FMAC units, SMT cores, 2.4Ghz maximum clocks, and similar nonsensical stuff posted in this thread by the 'experts'...

I already addressed the comment about "needs" before, even providing a real-world example explaining how your 'logic' fails, but it seems was missed or ignored again.

Finally, what I have been saying about the CPU/APU real market is also applicable to GPUs. AMD has stated just some few hours ago that the next-generation of GPUs, which will be presented in brief, "is targeting more of the enthusiast market versus the ultra-enthusiast one".

This is just ... wow. I'm gonna step back now for as much as I like debate and all that, it's pointless trying to debate with a fanatic. It reminds me of Baron...

 

[noob2222]

he makes baron look innocent. I don't remember baron posting nda information and pretending to know more about AMD than what AMD themselves know.

Sad thing is he still believes his own credibility.