Thank you sxs112.
http://www.c627627.com/AMD/Athlon64/
(Refresh)
Printable View
Thank you sxs112.
http://www.c627627.com/AMD/Athlon64/
(Refresh)
Don't think he can OC because of the mobo..Quote:
Originally Posted by shuRe
my badQuote:
Originally Posted by furyfax
sxs112, why don't you give GCPUID a shot, see how it works with your new chip?
Download GCPUID here
the cache is shared between the cores, physically its one block of cache from what i have seen, and the cores use HT links to access it
No its not, they are two seperate cores.
How hard is it to believe? they are two seperate K8 Venice/San Diego cores connected at the memory controller.
sorry. but there is no way that's right. why would they take up die space making an slow HT link between the cache and core now when they never did before?Quote:
Originally Posted by reject
it's two seperate caches. i think all the cores will be made with 1mb+1mb, but any of them with a defect in the cache (likely, due to size) get cut down to 512kb+512kb
hmm so according to the newer roadmap only the 4800+ has 2mb of L2 and both the 4600+ and the 4800+ are 2.4 ghz.
Look at the picture in the post above yours. You're right, these are 2 seperate processing cores, with their own dedicated L1 and L2 caches, and the only place they meet is the memory controller.Quote:
Originally Posted by ozzimark
I have to disagree with you on your statement that all cores will be 1mb+1mb, when clearly, already its not the case.
20F30 is a CPUID reading of a dual core 1MB+1MB CPU, the new DC Opteron have a 20F10 CPUID reading. The chip we see in this thread is 20FB1, which is not similar to either of these CPUIDs.
The 20F30 and 20F10 are like 2 San Diegos stuck together, this 20FB1 is like 2 Venices stuck together, there is no disabled L2 cache on this chip. I don't think I've ever seen an ES (and this is a semi-ES as it doesn't have a model number programmed into it, that's why the BIOS reads it as 2200+) that had disabled cache, or been disabled by any other means.
No, they meet at the crossbar, the System Request Queue. This in turn is connected to the HT links and the memory controller. So CPU-CPU cache snooping traffic never goes out to the HT links, like in a dual CPU system. This improves performance slightly.Quote:
Originally Posted by The Coolest
Code:MC HT
\ /
SRQ
/ \
core1 core2
Not exactly. An AMD dual core CPU will be able to *appear* as a single core with HT support. This is because some morons develop software for HyperThreading, but fail to spawn threads for multi-CPU. By appearing as a HT CPU an AMD dual core would be able to take advantage for HT-aware Dual-UNaware software.Quote:
Originally Posted by LowRun
Pjotr>> Yes you're right on the fact that there's System Request Queue between the cores and the memory controller. But what I wanted to say basically is that the CPUs are still two completely seperate cores.
HyperThreading is relatively old tech, dual core is relatively new, not all developers that were writing their programs for multi CPU support, never knew about Intel and AMD releasing multi-core CPUs in the future, so this feature allows older software to still have a noticable performance increase over running on a single core.
So I wouldn't call developers morons just because they didn't know about this.
my question is, how can we overclocked it?
independent clock speed for each core, or just one for both? ummm I think 1, since it only have 1 mem controller :) and independent clock speed adjustment for each core is somewhat not logic
Quote:
Originally Posted by c627627
To sell a part at 3.2, it needs to run at 3.4, at least, with no problems. This is known as a "guard band". I'm not sure 90nm will deliver that, especially by Q106.
It's also pure speculation, not really a "roadmap". Perhaps you should cite the evidence you have for each prediction on your website with a link to an explanation somewhere else on your site. That way, folks can distinguish between supposedly leaked roadmaps and speculation.
Oh, and I wouldn't rely on the Ediot for anything. :)
Well this is strange.Quote:
Originally Posted by sxs112
If 4400 = 2 x (2.2GHz, 512K L2)
and (speculating) 4800 = 2 x (2.4GHz, 512K L2)
What the heck is 4600? Are they really going to release 2.3GHz cores? I guess that's possible, and perhaps makes some sense, even.
The alternative of: 4600 = 2 x (2.4, 512K L2), 4800 = 2 x (2.4, 1MB L2) is possible, but it doesn't make as much sense from the model number viewpoint.
I think 4600+ is 2.4Ghzs 512Kb cache and 4800+ 2.4Ghzs 1Mb Cache.Quote:
Originally Posted by terrace215
Its very possible that this is what's going to happen:Quote:
Originally Posted by terrace215
4200+ = 2.0GHz 2x512KB <-- Might not happen at all.
4400+ = 2.2GHz 2x512KB
4600+ = 2.4GHz 2x512KB
4800+ = 2.4GHz 2x1MB
I think that at least for now, both of the dual core CPUs are going to be called Toledo, as we don't have any other info on this.
or, it could be that
4400+ = 2.2 GHz 2x512KB
4600+ = 2.2 Ghz 2x1MB
4800+ = 2.4 Ghz 2x512mb
and maybe a higher version later on with 2x1MB
This is analagous to the new Venice and San Diego cores w/
3500+ = 2.2 Ghz 512k
3700+ = 2.2 Ghz 1MB
3800+ = 2.4 Ghz 512k
Just MHO.
Not if you accept the italian site CPUZ shot, which featured:Quote:
Originally Posted by Orthogonal
2.4GHz 2x1MB
It would be odd to be sampling a part this far in advance that wasn't going to be part of the launch (and thus probably the 2400+/4800+ A64 X2)
Of course I could be way off my rocker. If AMD is going to have the 4800+ as their new Flagship, it would most certainly have 2x1MB, if any at all have that much.
Yes. But HyperThreading is rather new, while dual core is over ten years old. It's not a new phenomenon.Quote:
Originally Posted by The Coolest
I'm a developer and I would. Anyone programming threads based on number of CPUs *should* ask the system how many CPUs there are. They should *not* check if there is HT and forget to check for multiple CPUs. That's just stupid and I would slap my developers if they did that. :slapass:Quote:
So I wouldn't call developers morons just because they didn't know about this.
Have you actually take the time to read how dual core works?
on the Intel side, which has HT, makes it a bit more complex:
First find out how many processors windows detects.
Then find out how many cores a single chip has.
Then find out how many virtual CPUs a single chip has.
Then find out which CPU is a virtual and which one is physical.
Build a list of which virtual CPU belongs to which physical, then make softare make the correct decision on what specific CPU to start a new thread....
Except programmers are stereotypically lazy and will ask if the machine uses HT and not do any other checks. Which is obviously stupid/bad programming.
if you program multithreaded, this shouldn't be a problem ?Quote:
Originally Posted by Adamantine
compilers don't have to be adjusted either, do they ?or am i mistaking here ?