sergionographybut then amds fake cores have a scaling of 80% on average, intels fake core barely top out 30% on their best day

I'm not sure if many people saw this, but I was discussing HT scaling in 7-Zip on the benchmark thread for it on the forums here, and HT scaling isn't that good. HT can actually slow you down if you're running 4 threads on a 4c/8t setup with 7-Zip. Here is some data, interpret it how you will, but it clearly shows that HT does not scale well and when you start scheduling things onto the HT cores when they're not needed, you lose performance.



This is on the same exact hardware with cores and ht being disabled and enabled in different ways. If I had a Vishera CPU, I would test with that too. Trust me, AMD has the better idea for multi-threading because at least the modules are more closely symmetric, unlike HT vs a real core.

Edit: Also don't misunderstand what I'm saying. Intel makes a very fast processor, AMD's core just scale better compared to Intel's CPUs with HT.

Prima.VeraI am not trolling, but obvious you are an AMD fanboy.
I am looking to this die pic and I can see 4 cores, no 8. Forgive me.

I'm obvious an AMD fanboy because I have a skt2011 rig and I actually did tests that resulted in numbers? Give me a fricken break.

At least I produced data to prove my point, you're just blabbering useless non-sense.

Also, the only thing I see four of from that picture is the L3 caches in the center of the die.

You obvious don't know what you're talking about, so before you make a fool of yourself maybe you should actually find out what you're looking at.

I've marked the CORES in red (including L2 caches, which are technically shared I think but changes nothing), and the L3 cache in purple...

Edit: Yes, I know. I'm over-simplifying it, but you get the basic idea. The FPU is mixed somewhere in there, but my vision is only 20/20 and the resolution of the picture is kind of small. I can't see the 32nm wires to actually describe what is what, so you shouldn't claim that you can.

Let's stop calling people names like fan boy. Okay?
That's not a suggestion by the way. ;)

this looks good, the power usage i mean

Prima.VeraI am not trolling, but obvious you are an AMD fanboy.
I am looking to this die pic and I can see 4 cores, no 8. Forgive me.

hothardware.com/articleimages/Item1923/small_vishera-die.jpg

4 Modules 2 Cores.

We don't need low performance with many cores. We need few cores with high perfomance per core ratio

KreijLet's stop calling people names like fan boy. Okay?
That's not a suggestion by the way. ;)

fool and troll are allowed then?

AquinusI've marked the CORES in red (including L2 caches, which are technically shared I think but changes nothing)(?), and the L3 cache in purple...

www.techpowerup.com/forums/attachment.php?attachmentid=49523&stc=1&d=1356619955

Please explain then what happens when you have L2 cache shared and also L3 shared, compared to an individual core, individual cache.

Prima.Verafool and troll are allowed then?

Prima.VeraPlease explain then what happens when you have L2 cache shared and also L3 shared, compared to an individual core, individual cache.

1: You know better than to double post. You know how to use multi-quote and the edit button.
2: Intel uses a shared L3 cache as well.
3: AMD shared the L2 cache to save space. The benefits out-weight the costs because you won't see a performance loss on a single thread because nothing else is touching the cache.
4: You're being called a fool and troll because you don't know what you're talking about. Not that I condone it and you started by calling me a fan boy, so I wouldn't go opening that can of worms.
5: The main reason is because the IPC is low due to increased chance of branch mis-predictions due to the length of the pipeline as well as how many instructions each "core" (NOT MODULE) can execute per cycle (which Vishera improved to some extent with the improved branch predictor and added x86 decoders.)

Yeah, AMD hasn't gotten it perfect for single-threaded applications, but it's a multi-threaded beast and as soon as AMD improves the IPC a bit more, it will be a lot more powerful than it already is not even considering the benefits to be had once they start producing it on a smaller process.

Also, if you knew anything about how memory heirarchies work, you would know that shared L2 cache would be the bottleneck when both cores in a module are running full power, not on single-threaded applications. So maybe you need to do a bit more research before you start claiming things that have no relevance considering AMD's multi-core performance is pretty good, despite executing fewer IPC.

So all in all, I don't agree with the labeling and I'm guilty of doing it from time to time (I'm trying not to, I really am. :p) but there is a reason why people get frustrated and start resorting to such tactics.

All in all, your comments have no factual backing and you're making a boatload of false assumptions... and you know what they say about making assumptions. :)

LTUGamerWe don't need low performance with many cores. We need few cores with high perfomance per core ratio

Or we need to wait for AMD to continue improve how many instructions per clock so all 8 cores run fast. I also don't find many people with Vishera chips complaining about them which usually is a sign that they're not that bad.

eidairaman14 Modules 2 Cores each you fool

Now now, you don't need to call Prima.Vera a fool, only the pitty is required. :)

Prima.VeraYou enjoy accusing people over and over and playing smart on forums? You work on AMD that you know so much? That would explain the performance of this CPU... :)

No, I like correcting people when people don't know how to provide factual information. Now are you going to continue to insult me because I'm trying setting the record straight or are you going to actually do some research so I have someone knowledgeable to debate this with? If you're going to continue to attack me and turn this into an Intel/AMD war, then you need to stop. All you're doing is degrading the quality of this thread for people who actually want to talk about it.

ok.

Prima.VeraPlease explain then what happens when you have L2 cache shared and also L3 shared, compared to an individual core, individual cache.

The same thing that happens when L2 is shared between two cores in Core 2 Duo. Squat. With your way of counting cores looking at cache hierarchy, Core 2 Quad must be a dual-core processor.

btarunrThe same thing that happens when L2 is shared between two cores in Core 2 Duo. Squat. With your way of counting cores looking at cache hierarchy, Core 2 Quad must be a dual-core processor.

each core must have an individual cache to be called a core. Everyone knows this.

3870x2each core must have an individual cache to be called a core. Everyone knows this.

Ever heard of L1 instruction and data cache? Pretty sure that every core has both. So what does that mean? *cough* 8 cores. :p

btarunrThe same thing that happens when L2 is shared between two cores in Core 2 Duo. Squat. With your way of counting cores looking at cache hierarchy, Core 2 Quad must be a dual-core processor.

+1: You just became my best friend for the day. :o

Prima.VeraWell, Intel also have 4 fake cores and it goes to 77W...;)

Hyperthreading is a technique. A technique doesn't exist in the physical, thus it virtually doesn't produce addtional heat.

AMD's additional cores isn't a technique, it's cores exist in the physical, thus produce heat.

So you want 4 physical cores to produce the same heat as a technique which doesnt exist in the physical?

Dent1Hyperthreading is a technique. A technique doesn't exist in the physical, thus doesn't produce heat.

Wrong, but the heat produced is minimal. A core with HT will run a bit warmer than one without (assuming both threads are fully utilized,) because it's using the used CPU resources to execute that second thread (hence why gains aren't all the great.)

AMD however has added shared resources in order to run two almost symmetric cores in tandem while saving die space.

Neither are bad, they're just different technologies with different goals in mind. HT adds efficiency to multi-threading sacrificing multi-threaded performance where AMD has something that scales better. Even more so when you improve how many instructions you can execute, because even small gains on an 8-core will scale almost equally across all 8 cores, unlike HT.

Keep in mind, though, I can't emphasize this enough. Neither technology is bad.

Dent1Hyperthreading is a technique. A technique doesn't exist in the physical, thus doesn't produce heat.

AMD's additional cores isn't a technique, it's cores exist in the physical, thus produce heat.

So you want 4 physical cores to produce the same heat as a technique which doesnt exist in the physical?

+1 I'm crunching with an FX-8 and an i7 Hex core (among others) and I just noticed that CPUID HW Monitor reads all 8 cores on the FX but only the 6 physical cores on the i7

OT- very interested in a 95w FX-8 Vishera but the price needs some work..... :shadedshu

AquinusWrong, but the heat produced is minimal. A core with HT will run a bit warmer than one without (assuming both threads are fully utilized,) because it's using the used CPU resources to execute that second thread (hence why gains aren't all the great.)
[/U][/B]

Yes I know this. But I'm comparing hyperthreading to multiple physical cores, not non hyperthreaded processor vs a hyperthreaded of the same physical core count.

I guess I should have said it virtually doesn't produce additional heat, because the additional heat produced is negligible compared to having more cores, to the point where I didn't think it had to be mentioned.

I think that 95w 8 cores is very good, since I have 4 cores for 160w

Given Intel's practice of charging ridiculous cost premiums for slightly-lower-wattage CPUs, I don't see why AMD can't do the same. In fact I'm wondering why AMD isn't making more noise about this - they've now got a CPU that is (on paper, at least) competitive with the i5-2500K in all aspects.

All they need to do now is get the 22nm transition right and improve the IPC and they'll have a winner on their hands. Let's see if Steamroller can pull it off.

sergionographynot really true, the 8320 has the same tdp as 8350 but with lower clocks, that means amd will prioritize the better bins for the 8350, and with the 8300 being a 95watt tdp it will probably be the same good bins as the 8350
so if you get a 8300 you are probably more likely to get a good clocker than a 8320

A 95W TDP doesn't guarantee a good bin, all it guarantees is that the silicon performs within that power envelope, at the specified frequencies within tolerable limits. It doesn't guarantee that it will scale in performance and efficiency at higher voltages and clocks just as the 8350s do. Some might perform as well and as efficiently as an 8350, but I would bet it's not many of them. Unless, of course, the fab is producing super-duper good silicon, more than they need for 8350 skus... I doubt that.

sergionographynot really true, the 8320 has the same tdp as 8350 but with lower clocks, that means amd will prioritize the better bins for the 8350, and with the 8300 being a 95watt tdp it will probably be the same good bins as the 8350
so if you get a 8300 you are probably more likely to get a good clocker than a 8320

TDP doesn't work that way. The 8320 is labeled as 125w because it is higher than 95w and lower than 126w, it could be anywhere between 95w and 125w. The 8320 could be 100w, the 8350 could be 120w. Just because they are both labeled 125w doesn't mean they both actually use the same amount of power.

TDP = cooling needed, not power draw. ;)

IF AMD is charging more, it's a better chip. I'll believe that, no problem. There is ZERO reason for a higher price, except that the silicon is better. 8350 is 4.0 GHz, 4.2 GHz Turbo.

cadavecaTDP = cooling needed, not power draw. ;)

IF AMD is charging more, it's a better chip. I'll believe that, no problem. There is ZERO reason for a higher price, except that the silicon is better. 8350 is 4.0 GHz, 4.2 GHz Turbo.

So you don't believe that heat comes from power drawn? A processor that produces 125w of heat is going to pull more power than a processor that produces 95w of heat. You know very well that the power draw of two processors in the same family can be compared using TDP, the processor with the higher TDP will pull more power.

Heat comes from power wasted not power drawn.

I'm not into this shit because i understand nothing about threading and stuff but for me you all sounds like school kids fighting over the number of rebound a ball could do only based on it's colour.