So can someone tell me what happened to the OctoCore promise? Boards like the Asus L1N64 SLI WS were supposed to enable two AMD Quad cores to run together (4X4) without Opterons or server software. Like this review; http://www.goodgearguide.com.au/in [...] ;3740;pt;1
But these are not socket F processors. So what is going on? If Intel (via ASUS) puts two Quad cores (not Xeon) on one Mobo soon then AMD will have lost that race as well. What the heck?
Good point! I remember all this "clear upgrade path" BS. Quad FX was nothing but a joke. Not that many people have use for 8 cores. If you need 8 cores, you probably are doing server stuff, so suck it up and get server level hardware with the ECC RAM.
Good point! I remember all this "clear upgrade path" BS. Quad FX was nothing but a joke. Not that many people have use for 8 cores. If you need 8 cores, you probably are doing server stuff, so suck it up and get server level hardware with the ECC RAM.
Have you actually come across anyone with a Quad FX setup? Seriously. I have honestly not seen ANYONE claim to have one running, which is rather odd since there are a lot of AMD fanboys out there. Baron pulling out after making all those promises about getting Quad FX tells you something, doesn't it...
It really depends on how AMD written their driver. If both cores need to access the same information, then yes, they would go to the same place. But for example, in Cinebench, where four cores render a different part of the image, the cache then needs to be divided to accommodate it.
Actually that's wht the L2 still is there for. The L3 is for shared information like TechnologyCoordinator pointed out. In addition the K10 can bypass L2 and L3 and directly get data from the memory.
As the errata with the broken L3 has shown, disabling the L3 causes a performance hit of up to 10% (that's what i read). So L3 seems to be working to a degree.
I don't think there really is a performance problem with the hardware. It performs as it was ment to - clock for clock it's faster than K8. AMD will, without doubt, improve performance a little by tweaking how the cache is used etc. But the separate cores on the K10 seem to be the "problem" if anything is.
I'm doing a lot of floating point computation using only CPU core. I'm missing in the benchmarks how one CORE performs (there are still lots of application which are not multi-threaded). Have you seen any one CORE benchmarks anywhere?
I think yomamafor1 might be on the right track ... but cache size alone is unlikely to be the major issue.
The latency going from 939 to AM2 went up and I imagine with K10 not being arch that different from K8 then this might be impacting even more on a quad.
Intel's cache arch is clearly superior ... and I agree a simple spin won't fix that much.
The 9500 is listed 30 bucks more than the q6600 here in Aust ... that isn't a good start when I can't even figure if my AM2 mobos will take the upgrade ... ASUS mentions nothing on their site about compatibility for the various AM2 boards in production.
I'd like to see some detailed tech article comparing Phenom / Barcelona / Core2 on this area like this older one, which seems to identify some of the issues ... look at the performance hit when the mem controller / cache frequency is lower:
Looking at the facts, I see a few things a note that have not been mentioned. First and foremost, AM2's processors had performance penalties for running the memory controller at less than 2.4ghz. Agena at this point is locked at 2.0ghz. Second, AMD originally stated they were going to have more power control features for each core on the die and they have been removed, I imagine, in an attempt to get the product out sooner. The idle power of Agena SHOULD be way lower than it is. I believe the power numbers under load look about where predicted, but, idle power was supposed to be both Barcelona and Agena's highpoint. It is curious what might be occuring with the errata issue the 2.4ghz and higher bin chips are having. IMHO, the chip needs to have it's NB/Memory controller running at the speed of HT3 to perform properly, and at this time, that is not possible.
Looking at the facts, I see a few things a note that have not been mentioned. First and foremost, AM2's processors had performance penalties for running the memory controller at less than 2.4ghz. Agena at this point is locked at 2.0ghz. Second, AMD originally stated they were going to have more power control features for each core on the die and they have been removed, I imagine, in an attempt to get the product out sooner. The idle power of Agena SHOULD be way lower than it is. I believe the power numbers under load look about where predicted, but, idle power was supposed to be both Barcelona and Agena's highpoint. It is curious what might be occuring with the errata issue the 2.4ghz and higher bin chips are having. IMHO, the chip needs to have it's NB/Memory controller running at the speed of HT3 to perform properly, and at this time, that is not possible.
Well on thing is that HT3 is available if you buy a new mobo with the 790xx chipset. There goes one thing. ALso, why would AMD drop their power control features if they had 1.5 years more than C2D/C2Q to be developed?
I for one do not think they dropped the power control features. I think this chip is just not as good as they expected it to be.
According to AMD all AM2 mobos can support Phenom but you will need to get a BIOS update for full support. Also on older AM2 chipsets you will be missing out on some of the power control features, no HT3 and some of the other things for Phenom.
Good point! I remember all this "clear upgrade path" BS. Quad FX was nothing but a joke. Not that many people have use for 8 cores. If you need 8 cores, you probably are doing server stuff, so suck it up and get server level hardware with the ECC RAM.
I run heavy processing software (CAD rendering and raytracing) where one render can take 20+ hours. More cores, even if they are slow and cheap, means less time to process without resorting to the really fast high-end cores. The software is expensive and usually one license is good for one computer, regardless of how many cores.
I am not 100% sure but I have been told that dual socket workstation computers (Xeon and Opteron) act like two computers and I would need two licenses. For me that is an extra $1000 per program.
That is why I was looking forward to the dual quad core setup. Perhaps I am the only person who wants a really powerful computer for something other than games? I don't know.
Well on thing is that HT3 is available if you buy a new mobo with the 790xx chipset. There goes one thing. ALso, why would AMD drop their power control features if they had 1.5 years more than C2D/C2Q to be developed?
I for one do not think they dropped the power control features. I think this chip is just not as good as they expected it to be.
You're not listening. HT3 is available between the CPU and chipset. On older AMD setups, the same speed was used between the CPU and memory. They are now clocking them independently (which is good), but they are locking the memery controller's speed at 2.0ghz (which is bad). Furthermore, they intended to make each core's power state completely separate as well as its frequency. When things turned for the worse, they decided to only make each core's frequency adjustable. From what I've seen of phenom's design, the issue it is having is a failure to execute. The design's issues have not been worked out properly, causing a chip that even a year late, still isn't what should have been released. The performance is abnormally dismal. If they had made a K8 quadcore chip with no design updates, they would have better performance. That tells you that things are not as planned.
i'll buy one when they reach $100 as a final upgrade to my final amd system. they screwed up having a price war cause i will never spend more than $200 on any current cpu again. they'll play hell convincing others too as well. except the fx and ee boy's, but they're crazy/stupid anyway. i dont think they'll survive to be competitive again. if they do, i'll need another system 10 years from now anyway.
i might consider them then, but not for a long time to come. i'm done waiting amd. too little, way to late.
Totally agreed there. The fact that even the fastest, recalled at launch, processor they have is still slower than the Q6600 speaks volumes.
Not many people on today. lol. The question I have is, if I stick an Agena in my old AM2 board, will the memory controller run at the CPU's speed? If so, will clocking it above 2.4ghz cause the same "errata" we have heard about?
The bottom line to me is that, after years of hype, AMD just didn't get it right. I suspect that they are having fabrication issues, but that is just an inference. I use both AMD and Intel processors, but this is a disappointment. I especially like processors with a lot of overclocking headroom, because I like to hot rod everything.
The bottom line to me is that, after years of hype, AMD just didn't get it right. I suspect that they are having fabrication issues, but that is just an inference. I use both AMD and Intel processors, but this is a disappointment. I especially like processors with a lot of overclocking headroom, because I like to hot rod everything.
I agree, but I was especially interested in being able to overclock the hell out of one core for the best single-threaded performance I could get, and just keep the other cores around for the background stuff that doesn't really require much. Maybe this will still be possible in the future, but not yet.
You're not listening. HT3 is available between the CPU and chipset. On older AMD setups, the same speed was used between the CPU and memory. They are now clocking them independently (which is good), but they are locking the memery controller's speed at 2.0ghz (which is bad). Furthermore, they intended to make each core's power state completely separate as well as its frequency. When things turned for the worse, they decided to only make each core's frequency adjustable. From what I've seen of phenom's design, the issue it is having is a failure to execute. The design's issues have not been worked out properly, causing a chip that even a year late, still isn't what should have been released. The performance is abnormally dismal. If they had made a K8 quadcore chip with no design updates, they would have better performance. That tells you that things are not as planned.
To be honest, K10 core is just almost exactly like K8, with slight improvements. To call it a brand new architecture is really a little overrated.
K10 is what you get by badging four K8 cores together, doubling some bandwidths, and connecting them with a shared L3. Given that K8 is about 20~30% less IPC than Core 2, I'm really not surprised that K10 falls short of Core 2.
Hey, What you just few peoples are talking about? AMD will never get success in making a true Processor that will attract the world. These 95% peoples are absolutely correct that “Intel is and will remain the performance king. What AMD has done, Forced Intel to get competitive technologically, OR Forced it to keep the prices low for its Dual and Quad Core CPU’s . What AMD possesses as compared to Intel: (i) Large number of fabrication sites? (ii) A good brand image through out the world? (iii) Large number of OEM purchasers? But Sorry, Will you Intel people remember “What was the price of a Pentium 4 2.4 GHz CPU 3-4 years before? Just same what we are giving Intel for its latest Quad-Core CPU now. How can this be? Is Intel thinking about those 30% middle-class fellows, so that they can benefit with the advantage of latest processors. Surely not, As like any other business Intel is there to earn profit and not to benefit the poor. Well, I am not a tech-champ But I would surely like to make a point as a general computer user, “Yes, Intel will win this performance battle and AMD once again will have to go in the past and make RAM’s and other things for PC’s running on Intel Processors. Then I will congratulate all of you who praised Intel and made it the one and only CPU supplier for the whole world. But after that do you know what we at this forum will be talking about, Nothing but the Increasing Prices of CPU’s Day after day. Any of you are familiar with the history of India. There were a large number Indian kings like you here who praised Britishers and made them stronger and stronger, They handed their kingdom’s to them and gained a title as “Sir”, “Jagirdar” etc. But what happened next is known to the whole world. Well, I don’t say AMD or Intel is best, But, can we as a single PC user can’t purchase an AMD processor and help that small company to survive and compete with a big Giant. How fast is today’s Intel’s Pernyn then AMD’s Phenom 15-20%, And How much big is Intel as compared to AMD 4-5 times. Well that’s a shameful to all of us what we are comparing. Is it good for all of us to stick on a single company and help making a monopolistic market. Just be true and ask your-self what will be the CPU market without AMD. Well, I will praise AMD for the time until it is next-equivalent to Intel and than I will say “Intel Is Best”.
Alpesh Shroff
DIGITECH SYSTEMS
To be honest, K10 core is just almost exactly like K8, with slight improvements. To call it a brand new architecture is really a little overrated.
K10 is what you get by badging four K8 cores together, doubling some bandwidths, and connecting them with a shared L3. Given that K8 is about 20~30% less IPC than Core 2, I'm really not surprised that K10 falls short of Core 2.
From what AMD originally stated they were doing with the way SSE instructions were performed, either, they did not follow through with their original intentions, or something is very wrong. Just looking at the Core 2 and Original Barcelona Designs specs, the Barcelona design SHOULD be able to pull significantly ahead (theoretical difference ouf 100%) of the Core 2.
