When is AMD going to release real 8 core chips?

[Idontcare]

nm

 

[Don Karnage]

Can you overclock using the turbo multipliers?

No :'(

 

[lopri]

Bulldozer certainly muddied the way cores are counted. Technically you can call it 8-cores or 4-cores and both would be right. Integer heavy parallel workload, it will behave like 8-cores. Everything else, it's more like 4-cores. Very slow one at that.

Back in the days I was a slow adopter to the increasing number of cores. I did not feel that dual-core was worth the cost (price, lower frequencies, potentially a new board, more power requirement) when it was first introduced. Same thing when quad-cores were out. Sure I bought them to see for myself but for me the benefit was minimal.

Then I realized the clear trend (moar cores!) and decided to make the most out of my hardware, which means that I started doing more than one thing on my systems simultaneously. Somewhere on the line virtualization was introduced to me, and it forever changed my computing habit. The benefits and posibilities of virtual systems, even for single users, are so sweet.

So when I looked at Bulldozer, I wanted to know whether it's going to be useful for me. Turned out, it is not. For my computing need, CMT is just as useless as SMT (Hyperthreading). I don't know if it's something AMD can fix with software/OS makers in the future, but the scheduling issue caused by sharing resources is not something I'd like to deal with on my desktop. My virtual systems have processor affinities assigned to them exclusively.

Sometimes I see reviews like this and facepalm. Admittedly it's a fair comparison, but at the same time a useless one. Nobody is going to play games while encoding video. Better way to use your extra cores is to manually tie their duties (heavy ones like encoding or VMs) to them. Encoding usually takes long time. VMs will always stay up. Your gaming session is likely to be longer than 30 minutes.

Assigning affinities to them takes only a few seconds, and suddenly your multitasking becomes much more pleasant.

I'm digressing but the point is that the Bulldozer is more of a quad-core to me than a octal-core. For example, I wouldn't want to tie core 6/7 to a VM and suffer abysmal performance, nor do I want to tie core 4/5/6/7 just to avoid the performance penalty since that leaves me only half of the resources for everything else. (If I use 4/6, then thread-hopping/scheduling will be inevitable)

 

[Ferzerp]

Core count, IPC, and clockspeed...all are figures of performance merit which on their own are meaningless when discussed in the absence of specifying the other two.

Add in factors of cost and power-consumption and the discussion loses even more relevance unless those two figures of merit are specified as well.

The issue with bulldozer is not core-count, or power consumption, or IPC...it is all of the above when combined into a singular product.

But that is where the great equalizer enters into the equation - price.

I include TCO when computing price though, for family and friends when it comes to recommending gear, and infortunately AMD does not. So while AMD does price their processors to be competitive with Intel on the shelf at Fry's, they don't price them to be competitive once you've factored in your TCO for electricity bills over the course of say 3-4 yrs that an individual will probably have the system.

But that is just me, and thankfully for AMD's sake they can count on most consumers not being like me, so pricing themselves to be competitive with Intel on the shelf is a wise move on their part, same as marketing their chips as "real" 8-cores and so on.

Can you blame them for trying? What kind of lame-ass company doesn't leverage their marketing division to highlight their strong points and minimize their weak points?

The problem is, by insisting that a module is 2 cores, AMD is shooting themselves in the foot. Bulldozer modules, treated as an SMT capable core suddenly don't look nearly as deficient in multithreaded scenarios. A module compares favorably to an Intel HT core if you treat both as 1 core (they trade wins, but only in heavily multi-threaded areas). However, allowing marketing to insist that a module is 2 discrete cores, AMD has committed a pretty large error in the pursuit of "MOAR CORES".

1. It makes the cores look extremely poor. Calling it 8, when seeing an Intel 4 core proc beat it in most (but not all) multithreaded scenarios makes the consumer think that each "core" is crap.

2. It absolutely kills them in the server space where cores can decide licensing costs. (SQL 2012, VMware prior to version 5, etc).

Whoever did the Opteron BD review for AT saw the ridiculousness of it too. See here:

2x AMD Opteron Interlagos 6276 (2.3GHz, 8 cores per CPU, 16 integer clusters)
2x AMD Opteron Interlagos 6220 (3.0GHz, 4 cores per CPU, 8 integer clusters)
2x AMD Opteron Magny-Cours 6174 (2.2GHz, 12 cores per CPU)

http://www.anandtech.com/show/5279/the-opteron-6276-a-closer-look/2


edit: My view is the following. A CPU has as many cores as it can run threads, iff you can take *any* two threads, and place them on an arbitrary pair of cores and always receive the same (within reason) performance regardless of which pair of cores is chosen. This is neither the case with a CPU made up of Intel HT capable cores, nor a CPU made up of AMD modules. In my view, neither have as many cores as they can execute threads at once.

 

[guskline]

Correct. I had a 486/DX2 33

Obsoleet: I remember replacing my Intel 486DX2 33 with a faster chip - an AMD 486 DX 40! Those were the days.

 

[Edrick]

Obsoleet: I remember replacing my Intel 486DX2 33 with a faster chip - an AMD 486 DX 40! Those were the days.

I went from a 486DX 33 to a 486DX2 66. Double the performance with the same socket. Unheard of today.

 

[Ferzerp]

I went from a 486DX 33 to a 486DX2 66. Double the performance with the same socket. Unheard of today.

Are you sure about that?

An i5 2700k uses the same socket as a G440. That's anywhere between 2 and 12-ish times the performance depending upon the workload.

 

[iCyborg]

2. It absolutely kills them in the server space where cores can decide licensing costs. (SQL 2012, VMware prior to version 5, etc).

I doubt it matters much how AMD calls it for licensing costs.

 

[Charles Kozierok]

I went from a 486DX 33 to a 486DX2 66. Double the performance with the same socket. Unheard of today.

That's called "low-hanging fruit".

And it wasn't quite double (though it was a lot closer than you'd get doubling frequencies today).

 

[Ferzerp]

I doubt it matters much how AMD calls it for licensing costs.

You would be incorrect.

edit: Though, to be fair, MS does recognize that AMD "cores" aren't comparable to Intel cores in regards to SQL costs. In that case, a 16 "core" BD Opteron is only licensed as 12 cores. However, if it had been marketted as a 8 core/16 thread proc, it would be 50% less to license. When you're talking $1800 per core retail pricing, any price benefit dries up in no time (Though, the platform price benefit is typically only around 5% for buying AMD instead of Intel, even though the Intel server typically performs better).

 

[iCyborg]

So if Intel releases a new Xeon and calls it single-core with 8 subprocessing units and 16 threads, the software industry would "buy" this BS and count it as a single core for licensing?

SQL 2012 is a good example like you noticed - they simply said the "core factor" is 0.75. If AMD called it 4-core, they would put it as 1.5. Oracle also has "core factor". In many of these server cases, it does perform closer to 8 cores than to 4, so if AMD called it 4-core or dual core they would adjust this factor, just like they would adjust it for the above "single-core" Xeon.

 

[Ferzerp]

SQL 2012 is a good example like you noticed - they simply said the "core factor" is 0.75. If AMD called it 4-core, they would put it as 1.5. Oracle also has "core factor". In many of these server cases, it does perform closer to 8 cores than to 4, so if AMD called it 4-core or dual core they would adjust this factor, just like they would adjust it for the above "single-core" Xeon.

No, they wouldn't. If they'd marketed it as Intel does, it would have been no sort of core factor (much less 1.5x as you claim). If your claims were true, they would break it down by architecture, core count, and clock speed. They don't, but at the same time, they realized if they didn't give some sort of break to the poor saps that bought "16 core" Interlagos chips, they'd never hear the end of it.


It's funny that you mention Oracle, because they didn't give any sort of break at all, and it costs you 2 and 2/3 more to license a 16 "core" Interlagos part as it does to license a 6 core/12 thread Intel part. The core factor would have been 0.5 for the AMD part, I assure you, had they marketed it sanely. (supporting doc here: http://www.oracle.com/us/corporate/contracts/processor-core-factor-table-070634.pdf) Those factors are regardless of the actual performance of the core. I can buy the fastest, or the slowest part, and the licensing is still the same cost. By insisting they have twice the cores that they could have, in Oracle's case, you pay 2x for the same performance that you would if they'd marketted it differently. If you disagree, note that an HT core is the same factor as what AMD is calling a core, and in most cases, the Intel HT core gives you the same or better real world performance than 2 of the AMD "cores".

 

[Phynaz]

IBM also licenses BD as an 8 core. Completely blows any cost savings out of the water.

Feel free to search on JF trying to defend this.

 

[pelov]

IBM also licenses BD as an 8 core. Completely blows any cost savings out of the water.

isn't this a case of OEMs looking to squeeze out more money and not AMD's fault considering Oracle/IBM/MS all have a different number for the "core count"? it seems like the OEMs are able to call them whatever they want.

 

[Ferzerp]

isn't this a case of OEMs looking to squeeze out more money and not AMD's fault considering Oracle/IBM/MS all have a different number for the "core count"? it seems like the OEMs are able to call them whatever they want.

They'll go with what the CPU manufacturer calls it within reason so the guy's made up 1 core 8 thread processor is a little silly.

However, if a CPU vendor does something silly like AMD did, they're going to exploit it.

If AMD called a module a core, they'd see the performance similarity between an AMD module and an Intel HT core (similarity, not parity) and just go with that number. When AMD purposefully inflated core count, they went with that as well. Though MS did provide a small break.

 

[ShintaiDK]

isn't this a case of OEMs looking to squeeze out more money and not AMD's fault considering Oracle/IBM/MS all have a different number for the "core count"? it seems like the OEMs are able to call them whatever they want.

When you market your CPU as a true 8 core etc?

MS/IBM/Oracle PR departments may still be heard laughing.

Specially on the server side its bad. And thats where AMD is already close to annihilated.

If I was AMD I would have said it was a 4 core with CMT/SMT and stay with it. First of all you lower licensing costs in the server space. Secondly on the desktop/mobile you can get away with the lacking performance alot easier. Plus some people would mention that it works better than Intels HT and atleast get alittle star there.

 

[Ferzerp]

If I was AMD I would have said it was a 4 core with CMT/SMT and stay with it. First of all you lower licensing costs in the server space. Secondly on the desktop/mobile you can get away with the lacking performance alot easier. Plus some people would mention that it works better than Intels HT and atleast get alittle star there.

This is my view exactly. If someone plopped BD in my lap and said describe this in the best light, that's the way I'd go.

With most software though, I think it's a poor design choice. I think the HT route is a little better, as in that case, the emphasis is more on single-threaded performance, with a little bonus tacked on. We can argue that "in the future all software will take advantage of multiple cores", but it ignores one major factor. It relies on a magical world where every software developer suddenly learns how to write efficient multi-threaded code. Developers are lazy. Most will write just what it takes to make the program spit out what it is supposed to. That isn't going to magically change beause it's easier to make a 50 core 1Ghz processor than it is to make a 1 core 50Ghz processor.

 

[ShintaiDK]

Maybe it was a hint from Microsoft that kb2592546 turned Bulldozer into 4c/8t. :hmm:

 

[iCyborg]

They'll go with what the CPU manufacturer calls it within reason so the guy's made up 1 core 8 thread processor is a little silly.

However, if a CPU vendor does something silly like AMD did, they're going to exploit it.

If AMD called a module a core, they'd see the performance similarity between an AMD module and an Intel HT core (similarity, not parity) and just go with that number. When AMD purposefully inflated core count, they went with that as well. Though MS did provide a small break.

What's the purpose of core factors then? Given that you're asserting it's identical to what manufacturer's marketing dept decides.
CMT often scales 80+% when enabled, while SMT rarely more than 20%. I don't think they could get away with jacking up the cores and physically duplicating stuff in it so it performs much closer to 2 cores and still insist it's one core, unless licensing depts are run by numbnuts.

I explained Oracle in the same post - if it performs more like an 8-core than a 4-core (meaning it scales well from 4 to 8 cores, your comparison to 4-core Intel is inappropriate), then they can leave the factor the same. It's single-threaded performance that's killing them, not how they decide to market their CPUs.

 

[Ferzerp]

The core factors for Oracle are more family differentiators than anything else.

Power (and Itanium now) is assigned a 1, x86 is assigned a .5, Sparc is a bit all over the place, but for IBM or x86 you pretty much know where it lies.

If performance really mattered, the lowest end quad core Xeon would be around half the core factor of the highest end. To illustrate it even more, a netburst core (for dual core procs) is the same as a SB core as far as Oracle is concerned. It has little to do with actual performance.

 

[iCyborg]

Yes, it's not the absolute performance. But you said yourself that calling Opteron 62xx single-core would be silly, so obviously it's also not just the marketing name.
SQL 2012 already reflects that, Oracle might simply believe that it's closer to 8 than 4, so to keep things simple, leave it at 0.5. Maybe they would do the same if AMD marketed it as 4c, maybe they wouldn't, but I would think they'd be smarter than just buying into the marketing name. Calling CMT module a core would be far more incorrect than calling it 2 cores. And Oracle charges per-core only for their flagship enterprise edition (granted, that one is the most expensive too).

 

[Ferzerp]

It's not "far more incorrect" when the performance difference between an Intel HT core is much, much more than the performance difference between an Intel HT core and an AMD module.

 

[iCyborg]

And it's much, much more than the performance difference of Intel SB HT core and complete dual core netburst CPU, yet the latter is still dual core. You said yourself it's not the actual performance that matters, only to bring it up here yourself?

 

[SickBeast]

To directly answer the OP: Not for a *very* long time. At least 5 more years. If AMD sticks with the BD layout then they will have 16 "cores" at that point, perhaps even more.

 

[blckgrffn]

To directly answer the OP: Not for a *very* long time. At least 5 more years. If AMD sticks with the BD layout then they will have 16 "cores" at that point, perhaps even more.

You can buy an Interlagos AMD CPU (8M, 16CMT or whatever we are calling it) on an ATX board for your PC today. It'd be silly, but the cost is similar to going with an Intel LGA 2011 system. Saying it doesn't exist because it isn't compelling for most users here, well, I don't think that's a compelling argument

http://www.newegg.com/Product/Produc...82E16819113036 <-- Link for the lazy.

If it wasn't so Phenom I'ish (power consumption is high, clock speed is low) I'd probably buy one as I am refreshing my ESXi stuff at home. Since I want it to also run 100% 24/7 and also support the same AVX instructions as SB/IVB, it just doesn't make enough sense.

Frankly, that is what I am looking forward to most from PD. Given that all of the power consumption tricks are much more effective at modest clockspeeds, a ~2.8 ghz 8M/16CMT @ ~125W (ideally, lower) part is going to be pretty interesting, IMHO. It will get a lot of work done if you can feed the beast.

http://www.newegg.com/Product/Product.aspx?Item=N82E16819113040 <-- I expect the resonant clock mesh to get us to 2.8Ghz @ 140W, then the rest of the power consumption magic/silicon maturity to get us the rest of the way there. If they have a 95W 2.5+ Ghz part for $500, I'll probably consider it pretty hard.

I've come to realize that my usage model represents a tiny fraction of the viewpoint on this particular forum, however