Link?
BTW
1. K10 isn't cancelled
2. SMT is a subset of SMP (multicore)
3. Huge caches (L3) are being released in mid 2007
4. AMD has never said anything about 10GHz (in fact, that's a fairly silly comment...)
Link?
BTW
1. K10 isn't cancelled
2. SMT is a subset of SMP (multicore)
3. Huge caches (L3) are being released in mid 2007
4. AMD has never said anything about 10GHz (in fact, that's a fairly silly comment...)
1. so where is it? tech prep? that means 2009 tapeout, at best. why bother talking about it.
2. it isn't.
3. so? is it balanced? how much will it cost?
4. marketers are retarded. sneering about marketing bs with 20/20 hindsight is even more retarded.
StrangerGuy
Diamond Member
- May 9, 2004
- 8,443
- 124
- 106
By the time Crysis comes out, it will be all about GPU performance, not the CPU. I bet it will even bring quad-SLI to it's knees.
Cache is cache. If K8L experiences an L2 cache-miss then it'll go to the L3 before hitting main memory. L3, while slow compared to L2, should still be WAY faster than main memory. Games are particularly dependant on cache size, which is why Allendales lose 5-10% performance per clock compared to 4MB L2 Conroes.
No they don't. You are reading too much into cache dependancy for gaming. It has a negligible impact on performance.
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2795&p=4
2MB vs 4MB
Quake 4 1.2 6.7%
Half Life 2 Ep 1 1.9%
Oblivion (Dungeon) 5.4%
Oblivion (Town) 1.0%
F.E.A.R. 1.03 1.5%
Battlefield 2 1.22 1.8%
Rise of Legends 1.6%
Avg difference: 2.84%
2.84% is hardly 5 - 10%. Stop making numbers up.
Oh, and here are some X2 3600+ vs X2 3800+ numbers:
http://www.xbitlabs.com/articles/cpu/display/athlon64-x2-3600_7.html
2x256KB vs 2x512KB
Far Cry
X2 3800+: 113.3
X2 3600+: 109.1
Half Life 2
X2 3800+: 84.4
X2 3600+: 80.8
F.E.A.R.
X2 3800+: 140
X2 3600+: 136
Quake 4:
X2 3800+: 98.2
X2 3600+: 94.6
Average difference: ~3.6%
CONCLUSION: Cache sizes have negligible impact on gaming performance. A 5% clockspeed gain is all that is required to negate a doubling in cache size, for both C2D and A64 platforms.
Yeah I would say it depends on what games tested, but judging from the results listed and seen here as well as TechReport here.
http://techreport.com/reviews/2006q3/e6300-vs-sff/index.x?pg=5
The cache helps somewhats, but the bulk of the performance is coming form the architecture itself. Some games are dependent on cache size, but not all.
http://techreport.com/reviews/2006q3/e6300-vs-sff/index.x?pg=5
The cache helps somewhats, but the bulk of the performance is coming form the architecture itself. Some games are dependent on cache size, but not all.
It actually depends on the video card you use. If GPU bound then your framerate differences are negligible. If looking strictly at CPU performance, however, cache makes a huge difference. Just because most video cards out there do not allow you to perceive the difference does not mean that it doesn't exist. Otherwise you can just say that an FX-62 performs the same as an X6800, since you can just set the game to 1600x1200 with 4xAA/16xAF on a single card to achieve equal performance. Yes, there are some games where you won't see much of a difference at all but, for the most part, gaming applications are more likely to be cache-bound compared to other applications simply because of the volume and variety of data they handle.
Regardless, my point was that adding any cache at all will bring benefits, and this is true.
Regardless, my point was that adding any cache at all will bring benefits, and this is true.
Anandtech was using a Crossfire setup, even though it was at 1600x1200, AA/AF was disabled so I'm not entirely convinced those results are predominantly GPU bound. 1600x1200 in CF is like 1024x768 on a single GPU, is it not?
As a person who still has a Abit BP6 with a couple of Celeron 333s humming along at 500MHz, I look forward to the return of dual sockets on the general desktop. Hopefully, it will turn out to be just as inexpensive in the end as the Intel solution was six years ago. However, if AMD limits it to the FX series as originally planned then it will just be a toy for the upper-end market.
AMD is reinventing the FX "series" from what I understand. The FX series will probably be priced from $375-$1000 per chip considering AMD has said there will be a 4x4 for under $1000 which leaves $250 for the motherboard; $375 for each processor.
Kougar
Senior member
- Apr 25, 2002
- 398
- 1
- 76
Regardless of whether Reverse hyperthreading exists or not, here is why AMD's 4x4 is only for show or those users that can afford Quad-SLI to go with it.
You can't take any two X2 CPU's and throw them together into the motherboard. You have to buy special CPU's that are designed for 2P systems. If you take a look at the cheapest dual-core 2xx model Opteron, it is showing as $316 right now. http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_609,00.html AMD has already publicly stated they will make 2 processor "kits" available for "under $1,000" for the 4x4 platform, so I am not sure how reassuring that is to hear...
Secondly with 4x4 you are required to buy TWICE the memory. Lets say you want a computer with 2GB. You would be foreced to by 4x1GB of RAM just to get a total of 2GB of system memory, because each processor must have it's own dedicated RAM.
So, expensive not massively produced server-like 4x4 motherboard: $300?
2 of the cheapest 2xx model processors: 2x$316
Twice the RAM: 2x$150
Extra beefy PSU to run everything will only cost more.
And the Core 2 Duo E6600 only costs $400 if you want to pay that much. Fry's is selling them for $319, although they are currently out again. More than a few places are selling the E6300 for $210 shipped now, I have already gotten mine and OCed to 3.33ghz absurdly easily. I stuck all the details in another thread in this subforum though.
Now while I do expect AMD's K8L to sweep away the performance crown, I don't think it will be by a large margin. Simply because K8L does not share it's L2 cache. It's basically a 2x-"2x512kb" L2 design, or maybe a 2x-"2x1mb". Only the L3 cache will be FULLY shared exactly like the L2 cache in Conroe, and initially AMD is reporting the L3 cache pool will = 2mb, and later upgrade to 4mb versions. Considering they are so starved for silicon wafer real-estate that they have cut their 1mb parts off the books and instead cut a 2x512kb X2 into a 2x256KB X2 3600+, I think 4mb K8L's will be long while in coming.
What I find astounding though, is that while Intel will be ramping up 45nm by this time next year, AMD will be matching the 45nm transition by the first half of 2008, far ahead of their original plans. If they can match Intel's fab process size, then Intel is going to be in deep trouble... That is all a good ways off still, however, and the Core 2 Duo is here and now. Kentsfield will even be out this fall, while K8L will be here just before the first half of 2007 ends.
Even so, some food for thought... wasn't everyone bashing Intel for their heater cores that offered sub-sub par performance while straining PSU's and power bills? So how is sticking 2 not as power-efficient dual-cores together to create a quad-core system, then doubling the RAM, all on a motherboard that will require some juice of it's own to support everything just to beat a single Conroe processor any more efficient? And before anyone mentions the High efficiency line of dual-cores, I suggest you go look at the prices for the 2xx 2P High-Efficiency model Opterons linked above, and double that figure since 4x4 will need two.
You can't take any two X2 CPU's and throw them together into the motherboard. You have to buy special CPU's that are designed for 2P systems. If you take a look at the cheapest dual-core 2xx model Opteron, it is showing as $316 right now. http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_609,00.html AMD has already publicly stated they will make 2 processor "kits" available for "under $1,000" for the 4x4 platform, so I am not sure how reassuring that is to hear...
Secondly with 4x4 you are required to buy TWICE the memory. Lets say you want a computer with 2GB. You would be foreced to by 4x1GB of RAM just to get a total of 2GB of system memory, because each processor must have it's own dedicated RAM.
So, expensive not massively produced server-like 4x4 motherboard: $300?
2 of the cheapest 2xx model processors: 2x$316
Twice the RAM: 2x$150
Extra beefy PSU to run everything will only cost more.
And the Core 2 Duo E6600 only costs $400 if you want to pay that much. Fry's is selling them for $319, although they are currently out again. More than a few places are selling the E6300 for $210 shipped now, I have already gotten mine and OCed to 3.33ghz absurdly easily. I stuck all the details in another thread in this subforum though.
Now while I do expect AMD's K8L to sweep away the performance crown, I don't think it will be by a large margin. Simply because K8L does not share it's L2 cache. It's basically a 2x-"2x512kb" L2 design, or maybe a 2x-"2x1mb". Only the L3 cache will be FULLY shared exactly like the L2 cache in Conroe, and initially AMD is reporting the L3 cache pool will = 2mb, and later upgrade to 4mb versions. Considering they are so starved for silicon wafer real-estate that they have cut their 1mb parts off the books and instead cut a 2x512kb X2 into a 2x256KB X2 3600+, I think 4mb K8L's will be long while in coming.
What I find astounding though, is that while Intel will be ramping up 45nm by this time next year, AMD will be matching the 45nm transition by the first half of 2008, far ahead of their original plans. If they can match Intel's fab process size, then Intel is going to be in deep trouble... That is all a good ways off still, however, and the Core 2 Duo is here and now. Kentsfield will even be out this fall, while K8L will be here just before the first half of 2007 ends.
Even so, some food for thought... wasn't everyone bashing Intel for their heater cores that offered sub-sub par performance while straining PSU's and power bills? So how is sticking 2 not as power-efficient dual-cores together to create a quad-core system, then doubling the RAM, all on a motherboard that will require some juice of it's own to support everything just to beat a single Conroe processor any more efficient? And before anyone mentions the High efficiency line of dual-cores, I suggest you go look at the prices for the 2xx 2P High-Efficiency model Opterons linked above, and double that figure since 4x4 will need two.
I thought that kentsfield will be out in 4Q'06 I thought that it wass orginally slated for 2Q'07 but Intel ramped it up early. Also I know that Anand's test between AM2 and Conroes memory Controllers showed that the Coroes Branch prediction pretty much negated AM2's on-die Memory Controller. Also AMD 4x4 will be 2 seperate chips with the Mobo Seperate HTT's and even memory for each Processer but kentsfield is all on one chip with unified memory and FSB so would that be faster that AMD's 4 core solution? Or would only having one FSB @ 1333Mhz prove to be a bottleneck? Unless Intels FSB proved to be a major bottleneck I would think that AMD's quad core would suffer the same problems that Intel's 8XX series did by requiring the FSB to communicate between cores. In which case the Conroe is already faster than AM2s significantly. and Kentsfield would be even quicker then AMD's 4x4. i know that Kentsfield is poised to be the Extreme Edition Intel, but with requiring a special MoBo and two expensive Procs it seems to me "4x4" is just a desperate stopgap/ bandaide for AMD.
Originally posted by: Kougar
Regardless of whether Reverse hyperthreading exists or not, here is why AMD's 4x4 is only for show or those users that can afford Quad-SLI to go with it.
You can't take any two X2 CPU's and throw them together into the motherboard. You have to buy special CPU's that are designed for 2P systems. If you take a look at the cheapest dual-core 2xx model Opteron, it is showing as $316 right now. http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_609,00.html AMD has already publicly stated they will make 2 processor "kits" available for "under $1,000" for the 4x4 platform, so I am not sure how reassuring that is to hear...
The problem is that you are applying the pricing scheme of S940 2xx Opterons towards the AM2 FX chips...that really doesn't work. I agree that $1000 for 2 CPUs and a mobo is above the price range of most users, but Kentsfield probably will be as well.
Secondly with 4x4 you are required to buy TWICE the memory. Lets say you want a computer with 2GB. You would be foreced to by 4x1GB of RAM just to get a total of 2GB of system memory, because each processor must have it's own dedicated RAM.
That's not how the system works...each processor has all the ram available, but you can only have 4 slots per processor. So if you have 2GB on each processor, the system will show and allow you to use 4GB. The difference is that you can have a max of 16GB in a 2 socket system as opposed to 8GB for a single socket.
I don't know why you think that the C2D's shared cache has an advantage over the Directly Connected cache of AMD, but it doesn't. There is an advantage to the SIZE of Intel's cache, but it isn't that much in most cases (say 5%).
What I find astounding though, is that while Intel will be ramping up 45nm by this time next year, AMD will be matching the 45nm transition by the first half of 2008, far ahead of their original plans. If they can match Intel's fab process size, then Intel is going to be in deep trouble... That is all a good ways off still, however, and the Core 2 Duo is here and now. Kentsfield will even be out this fall, while K8L will be here just before the first half of 2007 ends.
Even so, some food for thought... wasn't everyone bashing Intel for their heater cores that offered sub-sub par performance while straining PSU's and power bills? So how is sticking 2 not as power-efficient dual-cores together to create a quad-core system, then doubling the RAM, all on a motherboard that will require some juice of it's own to support everything just to beat a single Conroe processor any more efficient? And before anyone mentions the High efficiency line of dual-cores, I suggest you go look at the prices for the 2xx 2P High-Efficiency model Opterons linked above, and double that figure since 4x4 will need two.
Again, I think you are confusing 4x4 with 2xx Opterons. There are also HE X2s, and the 4x4 will be using the X2 FX AM2 line. In addition, AMD has already stated that their quad cores will be using the same TDP as their present dual cores, so I don't really see this as an issue...
AMD's 4x4 will not have unified RAM since the MoBo will have Seperate HTT's (FSB's) thus would be controlled by each Processors built in Memory controller would control its own Memory They might be able to share between banks but I would think that it would require using the Northbridge, but the AMD sides whole claim to Fame is that since 4x4 has twin HTT's and independant Memory banks that Kentsfields Achillies heel will that one FSB will be a bottleneck, while Intel boys claim that 1067Mhz isn't even close to being Saturated anyway.
Kougar
Senior member
- Apr 25, 2002
- 398
- 1
- 76
I used them as an example, because they are the closest thing I know to what the 4x4 chips might be. If 4x4 requires FX brand chips then I'm not sure how AMD can keep the price tag of a 2P kit under $1,000, but I can't claim to know everything about AMD's 4x4 platform. Still, the lowest of the low end models won't get cheaper than the Operton's own prices, so that pretty much negates the cost saving arguement against any Core 2 Duo, price gauging or not. (Turning a $208 E6300 into a 3.33ghz chip isn't exactly bad though... cheap RAM and mainboard here too)
I'll stand corrected then. But you would still be required to buy twice the memory, as I don't think you can leave one processor without any?
The Core 2 Duo's two cores use one single pool of L2 cache. Worst case scenario, if you buy an Allendale with half of the L2 cache, either of the cores can still use the entire 2mb L2 cache with the other core helping or sitting idle. It is an inherent physical advantage. I expect the private L2 cache approach by AMD will work, but it won't work as efficiently as Intel's fully shared cache. Then again AMD's chips don't really need cache, so the 2mb L3 pool of cache might work great, I can't really say.
Yes, but HE processors cost MORE, above and beyond the already more expensive 2P Opterons. I was simply trying to make a point about the prices again... But if AMD uses the FX line, then I guess $1,000 for a two processor kit would be a bargain. According to: http://www.xbitlabs.com/news/cpu/display/20060731233200.html They will start out at $1,000 for a kit of two.
You're right in that 4x4 isn't a low-end-then-overclock platform...the closest Intel comparison would be Kentsfield (quadcore), not Conroe. As they will be FX chips, the price you need to compare to (if you only have today's models) is 2 x e6600s plus a motherboard for $1,000. That said, I'm sure that both the FX line and the e6600s will have very different price points when 4x4 is launched, making all of this moot.
Not really...configuration is not capacity (either min or max). Optimal for a 2GB system would be 4 x 512MB scenario (2 dimms for each CPU, from ~$160 total), which means that you could use the lower latency 512MB sticks and still have 1T settings...that's about as good as it gets on a 2GB system, and it's cost effective as well. If you need the memory though (as in a 64bit OS), you could use 8 x 2GB sticks and run with a 16GB system (again, this is unmatchable by Intel on a consumer "non-FBDIMM" system), though at $2000 for ram, you better need it.
The Core 2 Duo's two cores use one single pool of L2 cache. Worst case scenario, if you buy an Allendale with half of the L2 cache, either of the cores can still use the entire 2mb L2 cache with the other core helping or sitting idle. It is an inherent physical advantage. I expect the private L2 cache approach by AMD will work, but it won't work as efficiently as Intel's fully shared cache. Then again AMD's chips don't really need cache, so the 2mb L3 pool of cache might work great, I can't really say.
There really is no advantage to a shared cache over 2 directly connected caches. Unlike the seperate caches found on the 8xx and 9xx Pentium Ds, AMD has a crossbar within the processor for direct cache coherency (the PDs need to go through the FSB and back).
There are many more differences between Opteron and X2 than most in the consumer market would understand (even enthusiasts). Most of this has to do with what is called RAS (reliability, availability, and serviceability), and it's VERRRRRY important to mission critical servers (but not at all to consumer systems). Basically it is various "hooks" that allow a system administrator to monitor and adjust the system remotely at highly granular levels.
This is why a Xeon 5130 (2GHz Woodcrest) sells for $350, and the e6400 (2.13GHz C2D) sells for $264...only the Opteron has many more RAS features than the Woodcrest, so the difference is even greater (Intel assigns many of their RAS features to the chipset rather than the CPU).
Kougar
Senior member
- Apr 25, 2002
- 398
- 1
- 76
Ahh, did not know about the crossbar on the AMD's. And I'll admit I'm not giving the direct connect design enough credit. But there is still some kind of performance gain from having a directly shared pool, because there isn't a reason for AMD implementing in a directly shared pool of L3 otherwise.
Very good info about RAS, I also didn't know any of that.
As far as pricing, yer likely right. Prices are nuts, and once Intel releases the X6900 they will drop at least a little, then with the release of Kentsfield they will drop a good bit more I think. As far as how well Kentsfield works.. well that is a whole 'nother thread. But I am running a E6300 with a 501FSB, so I would be willing to bet a little OCing would drastically reduce the bandwidth bottlenecking, even if it would not be enough to "fix" it. And everyone knows how well a single Core 2 Duo runs... A little OT, but anyone know if Kentsfield will be called the Core 4 Extreme?
Very good info about RAS, I also didn't know any of that.
As far as pricing, yer likely right. Prices are nuts, and once Intel releases the X6900 they will drop at least a little, then with the release of Kentsfield they will drop a good bit more I think. As far as how well Kentsfield works.. well that is a whole 'nother thread. But I am running a E6300 with a 501FSB, so I would be willing to bet a little OCing would drastically reduce the bandwidth bottlenecking, even if it would not be enough to "fix" it. And everyone knows how well a single Core 2 Duo runs... A little OT, but anyone know if Kentsfield will be called the Core 4 Extreme?
Jeez, these last few posts are long... I didn't really read them, to tell the truth, I just kind of skimmed through them.
First, the CPUs in 4x4 will be connected to EACH OTHER directly (via an HT link).
Second, there is a latency penalty when accessing another CPU's memory but the difference is probably around 15-20ns (just a guess, by the way). Of course, you will have twice the available bandwidth as a single AM2 setup if your OS can use it.
Third, the reason to implement a shared cache is to lessen crossbar traffic when dealing with more than 2 cores. That and to reduce contention for the memory controller, since you'll have a bigger pool of data to hit before actually going out to the memory controller.
Fourth, while Kentsfield will indeed be an MCP, it will use the northbridge as an arbiter, so there's no performance benefit to having the two conroe dies on the same package since there's little to no direct communication between the two. The only benefit is that the motherboard footprint is significantly lower.
Also remember that initial Kentsfields WILL be XE CPUs and, thus, will have the $1000 price tag and their power draw will be 2x that of similar Conroes (since there ARE two Conroe dies in the package).
First, the CPUs in 4x4 will be connected to EACH OTHER directly (via an HT link).
Second, there is a latency penalty when accessing another CPU's memory but the difference is probably around 15-20ns (just a guess, by the way). Of course, you will have twice the available bandwidth as a single AM2 setup if your OS can use it.
Third, the reason to implement a shared cache is to lessen crossbar traffic when dealing with more than 2 cores. That and to reduce contention for the memory controller, since you'll have a bigger pool of data to hit before actually going out to the memory controller.
Fourth, while Kentsfield will indeed be an MCP, it will use the northbridge as an arbiter, so there's no performance benefit to having the two conroe dies on the same package since there's little to no direct communication between the two. The only benefit is that the motherboard footprint is significantly lower.
Also remember that initial Kentsfields WILL be XE CPUs and, thus, will have the $1000 price tag and their power draw will be 2x that of similar Conroes (since there ARE two Conroe dies in the package).
Originally posted by: Kougar
A little OT, but anyone know if Kentsfield will be called the Core 4 Extreme?
No, Core2 Quadro Extreme.
A shared cache is both better and worse...obviously the speed is better, but you also have to keep things more "organized" which takes some work. All in all, it's pretty much a wash.
Very good info about RAS, I also didn't know any of that.
As far as pricing, yer likely right. Prices are nuts, and once Intel releases the X6900 they will drop at least a little, then with the release of Kentsfield they will drop a good bit more I think. As far as how well Kentsfield works.. well that is a whole 'nother thread. But I am running a E6300 with a 501FSB, so I would be willing to bet a little OCing would drastically reduce the bandwidth bottlenecking, even if it would not be enough to "fix" it. And everyone knows how well a single Core 2 Duo runs... A little OT, but anyone know if Kentsfield will be called the Core 4 Extreme?
Your point on increasing the FSB speed makes sense at first blush, but you are forgetting that you are also adding more cores and data. This means that the FSB has to do much more work with a quad core than it does on a dual core. In addition (as Furen pointed out), the Kentsfield is 2 Conroes glued together...this means that cache coherency between the 2 shared caches must go through the FSB as well. While I have no doubt that this will be no problem, it will most likely limit the amount you will be able to increase the FSB (and all things being equal, be a bit slower than a native quad core).
Link?
Fred Weber's presentation at the 2003 Microprocssor forum. Here's his feature list:
Threaded architectures;
Chip level multiprocessing;
Huge scale MP machines;
10GHz operation;
Much higher performance superscalar, out of order CPU core;
Huge caches;
Media/vector processing extensions;
Branch and memory hints;
GHz performance IO;
Security and virtualization;
Static and dynamic power management
http://www.xbitlabs.com/news/cpu/display/20031016140742.html
1. Then were is it? Assuming development started when K9 was cancelled it would be a 2007 chip. There is no mention of it in AMD roadmaps through 2008.
2. No, it's not.
3. A few megs are huge?
4. Actually it's silly of you not to Google before you reply and are proven wrong. Google just a little bit more and you can actually find the slides from his preseataion saying AMD would be at 10ghz by now.
He also said that this would be part of FUTURE microarchitectures, as in K9+. There was no timeline given for these improvements, nothing like "We'll have 10GHz by 2004". Also, many of these improvements have already been implemented or will be implemented soon enough.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 22K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 399
Facebook
Twitter