Yes, teksyndicate is very open minded. Very clear by the opening of the video /sarcasm
I wish people were smart enough to question why they're the only ones with the results they're getting instead of being swayed by their fanboyism.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Time to Leave AMD for Intel?
- Thread starter UMASS
- Start date
Yes, teksyndicate is very open minded. Very clear by the opening of the video /sarcasm
I wish people were smart enough to question why they're the only ones with the results they're getting instead of being swayed by their fanboyism.
cageymaru
Fully [H]
- Joined
- Apr 10, 2003
- Messages
- 22,090
I got similar results when my buddy who has a 2500K compared his fps to my FX-8120 fps when playing BF3, BF4, and some MMOs like The Secret World and FFXIV. Standing side by side and shooting at the same enemy targets, we were pretty much tied as far as performance goes. Running dungeons together we had the same experience. I usually end up giving him my graphics settings as they work perfectly fine for him. He just bought an i5 4670K I believe and a GTX 780. I'll ask him for his exact system build tonight if he has time to chat. If I can find a 290(x) soon we might try it again to see if there's a real world difference in our rigs since I also upgraded to a FX-9370.
Now when we play single core games his rig dusts mine by a decent percentage. If AMD doesn't announce new desktop chips in their next lineup, I'll probably look into a 6 or 8 core Intel. I don't need to upgrade at all truth be told. I just like shiny things from time to time.
Now when we play single core games his rig dusts mine by a decent percentage. If AMD doesn't announce new desktop chips in their next lineup, I'll probably look into a 6 or 8 core Intel. I don't need to upgrade at all truth be told. I just like shiny things from time to time.
Last edited:
I fail to see how something being $100 cheaper isn't considered "good value". I've seen a lot of budget builders go with a cheap FM2(+) mobo and pair them up with those $80~90 Athlon x4's, which are unlocked. Compare to Intel's cheapest unlocked quads which cost well over $100 more. For gamers or whatever, the Athlon provides more than enough grunt to drive a single GPU config -- usually paired up with something like a GTX 650 Ti/Boost or R7-260x/R9-270(x) and I've never heard anyone complain.
Might not sound like much, but $30 or even $100 is a lot for someone on a tight budget looking to build a rig and get it over with.
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,941
Why are you making this comparison? Compare them to i3s or Pentium chips which better match performance to AMD quad core processors.
Last edited:
Agreed. That mere $100 in savings can mean the difference between a reference GTX 760 or a GTX 770, or even most of the difference between a reference GTX 770 and a 780. In some cases by shopping around, and with a bit of luck, you can find a CPU+Board combination for around the same price as an i5 alone much less an i7. For anyone doing heavy threaded work or running multiple VM's that is a huge potential cost savings. Core for core Intel is stronger but lets face it, outside of the enthusiast and/or professional niche, most modern CPU from both sides are going to be overkill even in gaming. Thus your average user isn't likely to notice much less really care about the difference.
That said I am disappointed by the apparent lack of Steamroller FX chips or even official news about why they aren't coming out, regardless of the socket they'd have ended up using. If I knew something was at least coming out I'd likely be all over it like white on rice but sadly that's not the case. Thus short of an amazing deal I'm going to end up with Intel this time around unless I decide to just limp my current guts along a little bit longer. Which I might just do anyway as they work well enough for the moment plus the worry that something /else/ will come up to potentially screw things up again despite having cash in hand. Like what happened during the past two months.
Why am I making that comparison? Why do people constantly mention i7's or Iris Pro when they try to shit talk AMD APU's? We can ask such questions all day, but we won't get anywhere.
And yeah, we'll see about those Celerons and Pentiums. Let's go load up a BF3 or BF4 MP match with those CPU's and see how their superior single-threaded chokes to death under the rigors of a 32+ man match, whereas the quad-core Athlon would feed the GPU just fine. It's been tested and proven. So that "performance match" thing is totally bogus when it comes to real-world gaming that uses more than a single strong thread.
And even the i3's are quite a bit more expensive than the Athlon's are, priced quite ludicrously for being dual-cores with HT.
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
Honestly don't know what you're on about here.
I've got a system here with a Core i3 3225 in it (that's a 3.3GHz dual core with hyperthreading), and it's faster than my old Phenom II X4 920 at everything.
I've had both boxes running side-by-side before. Even in heavily-threaded operations like h.264 encoding, the i3 wins by a bit.
That's because the i3's are generally quite a bit faster than the Athlons...
Did you run BF3/BF4 MP with them? If not, then what you just said wasn't a valid counter-point. I never said the Athlon's were faster at everything, but any game that truly uses multiple threads can allow the Athlon to outpace the Intel dual-cores. I'm well aware of how far behind AMD is in ST perf compared to Intel.
No, that's because that's precisely what Intel wants to price the processors at, has nothing to do with speed. Even back in the day when A64 was spanking Intel left and right, some of the Intel chips were priced higher than the AMD while still being outperformed.
No, that's because that's precisely what Intel wants to price the processors at, has nothing to do with speed. Even back in the day when A64 was spanking Intel left and right, some of the Intel chips were priced higher than the AMD while still being outperformed.
Dangman
Ninja Editor SuperMod
- Joined
- Dec 15, 2005
- Messages
- 46,056
Tested and proven by a decent review site or tested and proven by a bunch of forum goers?
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
...Did you read my previous post? It's more than valid (in fact, I stacked the deck in AMD's favor). To reiterate: I said h.264 encoding, which is FAR more CPU intensive than any videogame.
h.264 encoding also has nearly perfect scaling (100% linear), in that going from a dual core to a quad core (of the same architecture and clockspeed) precisely doubles the speed of your encode.
What this means is that h.264 encoding is absolutely THE BEST CASE SCENARIO for the Phenom II X4 920, and it's still a hair slower than the i3 3225...
Well, the best-scaling threaded application around isn't faster... so I don't really see how my Phenom II X4 920 could beat my Core i3 3225 at anything.
The i3 has the advantage at single-threaded AND multi-threaded, from my tests... (Also, keep in mind, the i3 does have Hyperthreading, so there are 4 'logical' cores available for highly threaded workloads)
Er, no. If you turn a blind-eye to wattage (and single-threaded performance) and go by price/performance alone, Intel's i3's are priced fairly close to their AMD counterparts. Slightly higher, but not bad.
However, once you get into i5's things start going loopy, because AMD has no comparable offering. At that point intel does whatever the hell they want with prices.
Edit:
As an example, The i3 3240 ($119) is fairly comparable to the A10-6790K ($119).
In the interest of full-disclosure, here are the trade-offs with the above two:
- Intel chip has 25% faster single-threaded performance
- Intel chip is half the wattage
- Intel chip supports hyperthreading (4 logical cores)
- Fairly identical multi-threaded performance (in spite of the i3 being a dual-core and the A10 being a quad-core)
- AMD chip is multi-unlocked
- AMD chip has arguably better integrated graphics
A lot of this comes down to how AMD defines "cores" these days. The A10-6790K only has two bulldozer units, yet they still label it as a "Quad Core" because each Bulldozer unit has two execution units... performance doesn't lie, though. It still behaves more like a dual core + hyperthreading than a true quad core.
I still prefer AMD's older STARS architecture (Phenom II-derived parts like Llano) for this very reason. Their core layout makes a lot more sense than anything based on Bulldozer.
Last edited:
Tested by forumers over at OCN. It was a massive thread and of course there was a benchmark war in said thread about it. When it came to BF3/BF4 MULTIPLAYER (emphasis on this, because you can run the single-player on a toaster's CPU), it was found that any dual-core, regardless of being AMD or Intel, lagged behind everything else when it came to MP performance. The i3's caught up with the AMD quad's, but which one was faster depended on whether or not the AMD's were OC'd or not.
Yes I read your post. The reason why I said what you said was invalid is because it had no relevance to BF3/BF4/other games which are properly-threaded, which is specifically what I was talking about. I never said anything about encoding, and anyone who knows anything about those is well-aware that they will utilize as much CPU grunt as possible and thus are far more intensive than games are.
Yes, and like I said, this isn't what I was talking about. I was originally talking about "value", then talking about gaming. I agree with everything you said here, but it wasn't my original point.
They're called quad-cores because that's precisely what they are. The reason they "perform like [intel] dual-cores" is because Intel's current CPU's are literally that much faster on a per-core basis that they can out-run their AMD counter-parts. This is why processors like the AMD FX 8350 were compared to higher-end I5's rather than i7's.
When it comes to core count under the K15 family (Bulldozer), each module has two full integer units in them. The downside to performance doesn't have much to do with the FPU (besides the FPU being vastly inferior to Intel's current stuff), but the multi-threaded performance that gets gimped when both int units in a module are fully-loaded. One core will give out 100% performance whilst the other gives out around 80%. This was a problem inherent to Bulldozer and Piledriver, but nixed in Steamroller.
AMD's implementation of CMT scales much higher than Intel's SMT does. You'll get about 30~35% the performance of a real core with Intel's SMT, and that's in a BEST CASE scenario.
The thing is that Bulldozer wasn't ever meant for super-high ST throughput, which was a terrible misjudgement by the old (now purged) AMD management. They decided at the time that by now, everything would be massively-parallel and thus all those hexa and octocores would be beasts. As we know, that didn't exactly pan out. I remember when Bulldozer was on it's way, the engineers actually clearly stated that x86 ST grunt would remain neutral or drop in some cases, and they would be focusing on MT throughput. Somehow everyone ignored this, and then all the tech enthusiasts hyped it up to infinity and we know what happened after that...
Now AMD are stuck with this uarch until post-Carrizo/Basilisk at least.
As for Llano, Trinity outperformed Llano by a good margin in both x86 and graphics, whilst using either the same amount of power or even less than Llano did. Though I remember Llano got screwed due to GloFo's terrible initial 32nm yields. From what I recall reading, Llano was supposed to have clocks much higher than the final silicon did, with them reaching or exceeding 4ghz. Oh well, that didn't pan out either... AMD can never catch a break these days, lol.
/rambling
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
By that logic, Phenom II x4s aren't true quad-cores, because they can only match a hyperthreaded Haswell dual-core. You showed it right in your very own tests. Phenom II x6s aren't true hexacores because they can only match hyperthreaded Nehalem quad-cores. Yet you claim Phenom II and Llano are true quad-cores. Hypocrisy is amazing, isn't it?
Performance does not indicate whether or not something is a dual core or a quad core. It only indicates what it indicates: performance.
Also, who are you to decide what is and is not logical in a CPU?
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
What are you talking about? I said exactly the OPPOSITE in my post.
I take issue with the definition of "cores" in the context of Bulldozer-based products because it's blatantly wrong. A "quad core" A10 only has two Bulldozer units, and while a Bulldozer unit contains two execution units, it does NOT contain two complete cores. It's fairly well-known how Bulldozer is laid out internally, and the odd configuration actually required a kernel patch for Windows and Linux to avoid putting two threads on one bulldozer unit whenever possible.
Older chips based on the STARS architecture (Like Phenom II) actually have fully defined cores. A quad-core Phenom II is 4 full cores. A hexa-core Phenom II is 6 full cores.
No, No I didn't.
You've somehow gotten it into your head that performance is the issue. It's not. I clearly stated that I take issue with how "cores" are defined on Bulldozer and that I PREFER chips based on Phenom II specifically because they use a more sensical core layout.
There's no hypocrisy, you simply have no idea what you're talking about.
Llano uses the Stars architecture, not Bulldozer. Ergo it is undeniably a true quad core.
I never said it did... you went off on that tangent all on your own.
Last edited:
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
Umm... try again?
The "odd" configuration is to prevent a decoder bottleneck. The same thing happened with hyperthreading. Windows had to be patched to avoid putting two threads on the same core if the other cores weren't occupied.
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
Not sure what you're pointing out here. It does exactly what I said (behaves like a dual core with hyperthreading).
And it behaves that way for exactly the reason I said it does (it's not a true quad core with 4 fully discreet cores, each execution unit shares resources with one other execution unit). The only reason I mentioned performance AT ALL is because it's in-line with what I would expect from such a core layout. It's supporting data, not the main point, and I'm not sure how you've confused it so badly.
What exactly are you taking issue with? You even QUOTED the bit where I stated each Bulldozer-unit has two execution units jammed in it, which WAS the main point.
Right, I used an example that's PERFECTLY threaded. I gave the Phenom II the best possible chance when comparing the two chips I happen to have on-hand, and it still couldn't keep up.
Seems totally valid to me.
Yes and no, they still have STARS...
Llano was basically just die-shrunk Phenom II, and I absolutely loved those chips. Low power, overclocked like crazy (motherboard willing), expected performance, and decently high IPC.
They could always pull an Intel. Drop the entire Bulldozer architecture (and all derivative), then go ahead and have another crack at advancing STARS.
Llano was handicapped by the silicon and some seriously poor chipsets (and no unlocked-multi chips to get around the poorly-clocking chipsets).
It looked like a great chip, but it was like they were actively trying to nerf it.
Last edited:
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
What?
Are you talking in terms of performance, or in terms of scaling? In terms of performance, a Core2Quad and a Phenom II behaves like a hyperthreaded dual-core Haswell. In terms of scaling, Bulldozer achieves near 80% scaling, Steamroller achieves near 100% scaling, while Hyperthreading is 0-20% scaling. If you want to justify your position by performance, then previous generations no longer become valid quad-cores. If you want to justify it by scaling, well, it flat out doesn't work because they're not even remotely close.
So what exactly are you trying to prove? That you take issue with Bulldozer's design because of relatively poor performance, or that because you have your own personal definition of a core that doesn't apply to others, or both? If it's because of the poor performance, whatever, AMD screwed up, but that's not because it's not true cores. If it's because of your personal definition, well, no one cares about your opinion, especially AMD.
Llano overclocked like crazy from what, 3.0 to 4.0, when previous generation Phenom IIs were achieving 4.2+? IPC was a mere 6% increase on average over Phenom II. It did NOT have decently high IPC. It was mediocre. Bulldozer was abysmal, but had clock speed that somewhat compensated. And why would AMD actively try to nerf it? There was absolutely no reason for them to do so, and every reason to do the opposite.
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
For the third time: in terms of how the chip itself is internally laid out. Not sure how you're still not getting this...
That I don't agree with calling a single bulldozer unit two cores... said it multiple times now.
And it's for most of the same reason I don't call a "quad-core i7 with hypertherthreading" an "octo-core"... because it's not! There are only 4 fully defined cores.
You tell me. They released Llano along side chipsets that COULD NOT lock the PCI/PCIe bus clock.
Result? Trying to overclock also increases the speed of the PCI and PCIe busses, which causes things like SATA controllers and video cards to flake out. This horribly nerfed any overclocking potential Llano had.
A Llano with an unlocked multi could have gotten around this issue, but they didn't bother releasing one of those either.
The information you just gave outlines that it DID have decent IPC... (not compared to today's options, but when it was originally launched)
At the time Llano launched, it had 6% higher IPC than Phenom II and MUCH higher IPC than the first-gen Bulldozer options... that made it AMD's highest-IPC part for a while.
This is EXTREMELY reminiscent of Intel's early struggles with Netburst, where overclocked late-model Pentium 3's were keeping up with (and surpassing) the performance of higher-clocked Pentium 4's... took Intel about 5 years to realize their mistake and go back to their older architecture. I'm curious if AMD will eventually do the same.
Last edited:
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
So what exactly defines a core? Separate L1, and L2 cache, along with separate decoders, prefetch, etc, for each core? Basically, what you're saying is that it's your opinion, and that you will not budge on your opinion no matter what anyone else says.
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
What's opinion here?
AMD's own documentation and press-releases clearly spell out that a "Bulldozer Unit" is, effectively, a single core with two execution units. Running more than one thread on a single bulldozer unit degrades the performance of both threads, which is not how discreet cores behave.
But then they turn around and label them all physical cores in their marketing material, when they should clearly be defined as logical cores. Oh well, that's how marketing works "Higher numbers are better! Reality be damned!"
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
Link to said documentations please.
And Steamroller doesn't share the same degradation (that I've seen), would you still classify that as a single module with two execution units?
And Steamroller doesn't share the same degradation (that I've seen), would you still classify that as a single module with two execution units?
The point being that your comparison didn't have FUCK ALL to do with what I was talking about, THAT's why it wasn't valid. What's so hard to understand about this? Just because Athlon got beat in another situation doesn't invalidate the fact that it'll outperform the others in the specific situation I was talking about.
They don't still "have" STARS, because no new chips are being made with that uarch. Llano overclocked like crap, the average max overclock was 3.6ghz, not that amazing when you consider the highest-end Llano -- the 3870k -- had a stock clock on 3.0ghz. The IPC of Llano was only 6~7% higher than Phenom II, so that's not "decently high". Trinity came out with 3.8ghz base/4.2ghz Turbo right out the gate, while using the same or less power than Llano to do so, which basically made Llano completely irrelevant.
And no, they couldn't just "go back" to Stars. Phenom and Phenom II were still part of the K8 family, and they had reached the end of any decent optimizations they were gonna get from it. It would be incredibly stupid for them to go back to K8.
What? The chipsets didn't have anything to do with it. It is well-documented that GloFo was having yield issues with Llano and their 32nm SOI node at the time. Even when the later revisions came with unlocked multis, the average max OC was STILL the same as it was with the locked multi chips, which was 3.5~3.6ghz. It was a flop when it came to that.
As for Bulldozer modules, I've never seen ANY AMD documentation that labeled a module as a "single core". In fact, they've always called them "dual core modules". Not to mention you can clearly see two physical cores on any die shots from K15 processors.
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
Yes it did. It was perfectly valid. You wanted to talk multi-core scaling, so I tested using the best-case scenario for scaling available. I stacked the deck in the AMD processor's favor as best I could.
My Phenom II X4 had the best advantage it could possibly have (an application that scales PERFECTLY with additional cores), and it still couldn't keep up with my box with a Core i3 in it.
Sorry, but that's pretty clear-cut. The i3 is faster. Games don't scale as well as h.264 encoding, so there's no way the Phenom II could be any faster in games if it's slower in h.264 encoding.
Nothing stopping them from going back to it. Intel dropped Netburst and went back to their Mobile Pentium 3's to create the Core architecture.
Already explained what caused that, poorly-clocking chipsets.
No chipset for Llano-based CPU's supports locking the PCI / PCIe bus. When you increase FSB speed, the speed of the PCI and PCIe bus increases as well. When these buses are clocked too high (rather than locked), one of the first things to flake out are SATA controllers, which made it impossible to boot the system at high FSB speeds.
Again, this is a chipset limitation, not a CPU limitation.
Already covered this as well. When Llano launched it had better IPC than Phenom II and better IPC than 1st-gen Bulldozer.
That would make it decently fast by AMD standards (at time-of-release). Best IPC of anything they had on-offer at the time.
Just like intel couldn't go back to the Pentium 3 architecture and optimize it further?
Oh wait, they did that, and the current Core architecture is a direct descendant of the Pentium 3 Coppermine core...
Yes it did. No chipset available for Llano supports locking the PCI / PCIe bus.
Ever tried overclocking, using FBS alone, on a motherboard that can't lock its buses? It doesn't work very well. Things like your sound card, SATA controller, and video card start to flake out because you're overclocking THEM along with your CPU.
This was HIGHLY limiting to Llano overclocking.
I've seen the die shots. They show what appears to be a single-core with two execution units and a crossbar. Looks nothing like a traditional dual-core...
Marketing can call them full cores all they want, but they sure don't look or act like it...
Last edited:
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
Did you intentionally skip over the part where unlocked Llano chips didn't overclock any better? I had completely forgotten about the 3870ks.
Edit: On HWbot:
Max on water: 4.368 ghz, failed CPU-Z validation (unstable overclock)
Max on air: 4.1 ghz, failed CPU-Z validation (unstable overclock)
Highest CPU-Z validated overclock (doesn't guarantee stability): 3.73 ghz. Only 2 out of 4 submissions at around 3.7 ghz passed CPU-Z validation.
Edit: On HWbot:
Max on water: 4.368 ghz, failed CPU-Z validation (unstable overclock)
Max on air: 4.1 ghz, failed CPU-Z validation (unstable overclock)
Highest CPU-Z validated overclock (doesn't guarantee stability): 3.73 ghz. Only 2 out of 4 submissions at around 3.7 ghz passed CPU-Z validation.
Last edited:
Unknown-One
[H]F Junkie
- Joined
- Mar 5, 2005
- Messages
- 8,909
Sorry, lots of quotes to cover there.
Unlocked Llano has a much better shot at overclocking (even with the gimpy motherboards available). Chips with good lithography have reached 4.8GHz.
He's not wrong about Global Foundries having issues at 32nm, but that doesn't mean there's a problem with the architecture, or that it can't scale-up (given a good pressing). It just means Global Foundries couldn't produce decent chips at 32nm, on-average.
You guys seem to be of the opinion that AMD designed themselves into a corner with Phenom II, and couldn't progress without throwing the old architecture in the garbage and starting over. I don't share that opinion, and believe Bulldozer (and derivatives) are an ill-advised direction to continue pursuing.
Last edited:
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
I'm more of the opinion:
Bulldozer, given enough time, can and has surpassed Phenom II performance. Bulldozer was designed for HSA, Phenom II was not. AMD had already sunk a lot of money and time into Bulldozer (at least 4-5 years), it would be a waste to just throw it away. It would waste even more money to skip Bulldozer and go straight to Piledriver and make a Phenom III in between, which probably wouldn't outperform Phenom II to any significant degree.
Also, proof for the 4.8 ghz stable 4-core Llano. Still waiting on links for documentations as well.
Bulldozer, given enough time, can and has surpassed Phenom II performance. Bulldozer was designed for HSA, Phenom II was not. AMD had already sunk a lot of money and time into Bulldozer (at least 4-5 years), it would be a waste to just throw it away. It would waste even more money to skip Bulldozer and go straight to Piledriver and make a Phenom III in between, which probably wouldn't outperform Phenom II to any significant degree.
Also, proof for the 4.8 ghz stable 4-core Llano. Still waiting on links for documentations as well.
K8 was done, it reached its wall. That's the main reason AMD tried to get Zambezi out as quickly as they could to "replace" the previous gen.
And I dunno where you got this 4.8ghz Llano thing from, that's pure nonsense. I've never seen ANYONE clock it that high and actually have it stable, let alone run it 24/7. Every review site topped out at either 3.5, 3.6, or even rarer -- 3.7ghz tops.
I wasn't talking about the random Phenom you pulled out of nowhere, I was talking about the Athlon x4 FM2 chips, which are $80~90, and comparing them to Celerons/Pentiums and i3's since they were in that bracket. Whether or not the Intel chips are faster in encoding has no bearing on the gaming performance in those games, and if actual people tested them and found that the Athlons were faster, then that's it.
I don't see how you're gonna try to argue a low-clocked Phenom II being slower than the i3's as "proof" that the (stock and overclocked) Athlon x4's keeping up with/being faster than the Intel's isn't true. It's completely nonsensical logic. I NEVER said the Athlons would be faster universally, so I dunno why you're trying to act like that's what's happening here.
Nothing stopping them except common sense. If Jim Keller himself says that they're on track to catch up on high-performance cores, I can logically discern that Stars is dead in the water, and for good reason. They squeezed as much out of K8 as they could, and that's the end of it.
The FSB technically hasn't existed on these chipsets for ages now. You're thinking about the reference clock, and the ref clock could easily be taken to 133mhz and above on many different MOBO's. No review site or professional or otherwise ever blamed it on the chipset, there is a hard wall at 3.6ghz regardless of how much voltage or ref clock you throw at it. And if the unlocked chips couldn't get passed 3.6ghz with any semblance of consistency, how is that a limitation of the chipset?
Which in the end, didn't matter at all since both the Zambezi and pre-existing Phenom II chips could out-clock the Llanos, totally invalidating the per-core IPC boost those chips had.
What sort of logic is this? You would've had a point if the K8 uarch was the same as the P3 uarch...except it isn't. Different uarchs will have different scaling capabilities. If AMD's engineers could've gotten more out of K8, they would've done so rather than rushing Zambezi out. In fact, Bulldozer was meant to launch four years prior to it's 2011 launch, but they couldn't work anything out due to the 65/45nm process nodes being too inefficient for any octocore chips like that. Had Bulldozer released way back then, the IPC would've been considered decent, and we would've been past Excavator by now. You can see how so many horrible executive decisions led AMD to where they are today.
Dude, I fucking HAVE a Llano chip, TWO of them actually. I know about OC'ing them, I know from actual experience, and I've already told you why this doesn't matter. I won't repeat it because I've already explained why above and in previous posts. If the unlocked chips could get past 3.6ghz, you'd have a case, but they couldn't, and thus you don't.
Oh really? Since you clearly have no idea what you're looking at, here is a Piledriver module with everything labeled for you:
You're clearly just looking at the entire module and calling it "one core", which fundamentally makes no sense whatsoever. You can try to arbitrarily redefine what a "core" is all you want, but your opinion won't replace reality.
Pieter3dnow
Supreme [H]ardness
- Joined
- Jul 29, 2009
- Messages
- 6,784
This just shows that you and many other Intel users have no idea what a cpu does you can only rate it if it goes faster in benchmark X or Y that defines a cpu for you.
When you don't know what you are talking about you come to the conclusion that they will just go back to older architecture just because you and others think it was better.
The idea what started with Bulldozer is more/better parallel computing. This what AMD does on the FM2+ platform as well. That benchmark X or Y still sucks on AMD hardware is not always AMD fault.
Software will prove a bigger problem but once it does get rolling AMD has far better cards then many on this forum think is possible.
First signs of this is the starswarm demo where the FX 8350 does really well also due to parallel computing capabilities and the developer even said that AMD core solution was one of the reasons it did so well.
Some games (Total War: Rome II for one) still don't multi-thread/task and require the fastest CPU available to run well.
The current trend is to move the multi-threading to the GPUs. So this is going to further discourage developers from multi-threading their games.
AMD has almost completely abandoned the desktop market (No 22nm based CPUs to compete with intel) All AMDs focus is on APUs with multiple cores, but notice the cores are distributed 2 GPU cores for ever CPU. Since the PS4 and Xbox One are both using a modified AMD APU, the trend to move processing to the GPU is only going to get worse
Last edited:
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
How is the trend towards iGPU computing a bad thing if it boosts performance?
I think there's a fear that since iGPU is really just a push towards mainstreaming hardware (smaller form factor), performance will eventually be forgotten.
I don't wanna sound inflammatory here, but you're very ignorant to current/recent developments.
Utilizing the compute capabilities of GPU's is the smart thing to do, when it's applicable and makes sense. This will not inherently have any effect whatsoever on devs pushing for more multi-threaded scaling for the CPU's. In fact, the main reason why CPU scaling has been so piss-poor all this time on PC is because DX11 and all of its predecessors mainly use one thread. Yes, some games are "multi-threaded", but DX generally mainly utilizes one thread.
DX12 will change that, as it is essentially Mantle under a different name, and with support for all the major GPU architectures out there (from AMD, Nvidia, Intel and Qualcomm.) Multi-threaded scaling will be legitimate and ideal, so the need for gamers to have incentive to drop more money on beefier processors will diminish, and pre-existing consumers who already have high-core/thread-count chips will benefit from this greatly. Single-threaded perf will still matter, but not nearly as much as it did in the past (as much as that is, since most games have been GPU-bound for many years now anyway.)
As for AMD's desktop presence, they are still very much in the desktop game. AMD doesn't have their own fabs anymore, unlike Intel, nor do they have insane amounts of cash to throw away on R&D for all sorts of different designs. They literally can't afford to be reckless right now. They are at the mercy of GloFo for their CPU process lithographies. Not to mention, some AMD employees already mentioned that they tested some designs at 20/22nm and the results weren't pretty, so it made sense to stick with a more refined 28nm design instead. You can't just look at what process node a technology is on and instantly judge how "competitive" it is against another part.
Maxwell (GTX 750 Ti) is still on 28nm, yet it brought significant perf/watt improvements over its predecessor, which is also on 28nm. Both AMD and Nvidia are very likely to skip TSMC's 20nm since it is not a process tailored nor suitable for high-performance GPU designs at all. They are very likely to do more parts at 28nm, then node-hop to TSMC 16nm FinFET.
This statement in particular though, I just don't even...:
AMD is focusing on APU's because that's what's getting them a lot of money, a lot of revenue. They sell many more of those than they ever did FX processors. Also the amount of GPU CU's has no correlation to the amount of "CPU's" there are, whatever that is supposed to mean. The 7850k has 4 CPU Compute Units (2x modules for a total of 4x cores), yet has 8 GCN CU's on the die. The 7700k has the same number of CPU cores, but has six enabled GCN CU's instead. The Llano and Trinity/Richland parts also didn't follow any rules of how many CU's were linked to each other, it was more about die space than anything else.
The PS4 and X1 aren't using "modified" APU's -- they're using semi-custom APU's. Those APU's were not seen in anything else, nor will they ever be seen in anything else besides the products they were designed to go into: the PS4 and the X1.
The trend for GPGPU Compute is the exact opposite of a bad thing. How is more efficiency a bad thing?
I buy in a lot of cheap so called 'low powered' machines for customers that most here would grin their noses up at.
To be honest when I'm using them they seem just as fast and responsive as the high end FX/i5+ machines I get in for others. I've even slapped in a old 5770 to test their gamin prowess and they get by just fine.
The bottom end of the market has moved on a long way from the 300Mhz Celerons and VIA/Cyrix CPUs of the late 1990's.
What would buy you a 49cc motor scooter 15 years ago buys you a nice 2 litre Focus today.
To be honest when I'm using them they seem just as fast and responsive as the high end FX/i5+ machines I get in for others. I've even slapped in a old 5770 to test their gamin prowess and they get by just fine.
The bottom end of the market has moved on a long way from the 300Mhz Celerons and VIA/Cyrix CPUs of the late 1990's.
What would buy you a 49cc motor scooter 15 years ago buys you a nice 2 litre Focus today.
Tsumi
[H]F Junkie
- Joined
- Mar 18, 2010
- Messages
- 13,758
Greater efficiency means designs can be scaled up for more performance for the same watts or same performance for less watts. Efficiency is a two way street.