AMD Quad-Father 4x4 Details Update

P

pawstar

Limp Gawd
Joined
Jul 2, 2005
Messages
465
Meh. I like AMD, but seriously, I'd rather go with a Kentsfield or a K8L. This looks like a waste of time to me.
 
Obi_Kwiet said:
Meh. I like AMD, but seriously, I'd rather go with a Kentsfield or a K8L. This looks like a waste of time to me.
Click to expand...

I agree, it's gotta be 4 on one socket then dual socket.
:(
 
The article is hinting at being able to put 2 quad cores in it when they come out. Personally, I think dual sockets is awesome, the more CPU power the better. Now the amount of PCI-E slots was insane...I still don't understand the need for SLI let alone more.
 
I wonder if it will be possible to use Opterons in the 4x4 system. I already ordered two of them; however, I know that once "K8L" quad processors come out, I'm definately upgrading so that leaves me with two opty's. If they could be used on the 4x4 platform that would be a great way to reuse 'em.
 
I don't see why people have negative reactions to this. First off, you're getting a dual socket board without having to purchase server grade equipment. Second, you will be running two desktop CPUs in SMP, which for Intel hasn't happened since the Pentium III and I'm not sure what the last desktop AMD chip to run SMP, was. Third and finally, this provides a way to run an 8-way setup ie two quad cores, again without dropping huge amounts of money on server equipment.
 
Because Intel is on the upswing ;) AMD didn't keep up their incredible upclimb, Intel came out with a chip they should have released years ago (Core Duo 2), and they fell approximately even for the first time in a while. Thus, people are turning on AMD in record numbers.

That said - if I want a dual-socket board, I want it server-grade. What reason would I have right now for an 8-core machine with no ECC? Most of the parallelized apps right now are server apps - web servers, SQL servers, distributed computing stuff, etc. None of which I'd consider running on a machine with no memory checking. Just that simple, as far as I'm concerned. Perhaps eventually, games will come out that use more than two threads, but until then it doesn't make sense to have a jillion cores.

 
I disagree on wanting to purchase server hardware in order to use duallies. I don't know how many times have I had two single-core desktop CPUs and wanted to put them in SMP, but couldn't because I need to spend an obscene amount on server grade CPUs, ram, and mobos. Now the same thing could be said about two dual-cores...they are getting popular enough that two could be bought cheaper than a quad-core.

I am also not sure why you made the assumption that server apps like web servers and database servers must be run on server grade hardware. I have dual P3 box (all desktop hardware) that runs your so called "server apps" and has be going for a steady for over 50 days. I understand the need for server grade hardware, but it isn't needed for my personal use. I am curious as to your reason for believing that dual-socket should be server grade motherboards only.

I also wonder with having a HT link per CPU would provide better throughput and bandwidth with two dual-cores in seperate sockets than one quad-core in a single socket, with a single HT link.
 
slithy said:
I am also not sure why you made the assumption that server apps like web servers and database servers must be run on server grade hardware. I have dual P3 box (all desktop hardware) that runs your so called "server apps" and has be going for a steady for over 50 days. I understand the need for server grade hardware, but it isn't needed for my personal use. I am curious as to your reason for believing that dual-socket should be server grade motherboards only.
Click to expand...
How many times did that desktop-class system make mistakes due to not having ECC? I'm going to guess you don't know. As far as I can see, if it's worth doing a particular calculation (whether it's raytracing a scene, loading a record from a table, or encoding xvid to mpeg2) it's worth doing it right in slightly longer time as opposed to wrong in slightly less time.

Fifty days is nothing. My machine is up for 26, with no UPS, no spike strip, no redundant hard drive, no nothing. I've had this machine up (with a similar lack of protection) for hundreds of days. But I wouldn't trust it to stay up with any kind of heavy load on it - Corsair says you can expect a memory error every 750 hours for 256 MB of memory. My fileserver has 2 GB of memory; about once every 4 days its ECC is catching an error, by that statistic.

ECC support seems so cheap and so little impact on performance that I don't know why it hasn't been made a requirement yet. Really. Perhaps the live-on-the-edge types can live with an occasional error in their systems, but I like putting in data and getting the same data out.

 
I think that the non-server enthusiast dual-cpu board is a great idea. When I was building my system 5 years back, I wanted to get an Athlon MP board which unfortunately was delayed and ended up being priced high. I am certain there are others out there just like me who just dream about having a dually system for a reasonable price. There are other uses for a nice system rather than just games. Amateur photographers and movie director wannabies are thrilled about this (at least the computer literate portion) as Photoshop & Premier are definately multithreaded as are many plugins. Even internet surfing power users are probably thrilled. Imagine being able to open a whole bunch of windows at the same time without any lagging as other pages are loading while working on something else. Having a server board for this would be a waste of money, but with the dually it becomes more feasible.

But I think people are also forgetting a key possibility that in the future a lot of co-processors are going to be made available on the Hypertransport bus so this board might be an icebreaker for the platform to emerge in the market and once quad or octa core cpus are available there will be other intereting ships to plug into that spare socket. What if AMD were to license the PhysX core and the AiPU and maybe a RAID, Crypto, etc accelerator that fits in nicely into that other socket!
 
unhappy_mage said:
Fifty days is nothing. My machine is up for 26, with no UPS, no spike strip, no redundant hard drive, no nothing. I've had this machine up (with a similar lack of protection) for hundreds of days. But I wouldn't trust it to stay up with any kind of heavy load on it - Corsair says you can expect a memory error every 750 hours for 256 MB of memory. My fileserver has 2 GB of memory; about once every 4 days its ECC is catching an error, by that statistic.
Click to expand...

My forementioned box actually is under the same conditions. No UPS, no RAID, no backup, and no ECC. It actually survived a power surge about a week ago that restarted all my other machines. Keep in mind this is just my home server if it goes down, I don't really those that much. No way in hell would I take risks like that with a real server in a data center.

Also, I wouldn't have a problem using ECC if it wasn't priced for the elite few.
 
I would much rather have real server hardware as well. I'll never go back to regular consumer stuff :cool:

I think this whole 4x4 fiasco is just that. It's not going to take off very well, and once AMD's new architecture hits they're going to drop 4x4 like a midget in the NHL.
 
Jason711 said:
anybody ever find out if you have to use fx cpu's or not?
Click to expand...

Well the CPU's must be for Socket F ie 1207 - So mainstream AM2 wont work for sure. But I too am wondering if regular Opty's will work (as asked earlier)
 
well, its been my opinion all along if that it is fx only, then the platform is total horse shit.

but hey, thats me.
 
I think this is a hoax. Seems like a desperate move to me. Also a good way to screw over the customer, why have 1 when you can have 2 for twice the price? Lame.
 
I think a few of you are missing a rather glaring piece of information...this will not work with standard desktop chips.

You cannot stick two 3800+ processors into one of these boards/configuations.

There have been a few key points released from AMD that this will be compatible with a FX-type chip only...expensive, high end consumer stuffs...

For the price of buying two current FX processors (AM2), you could buy a dual socket 940 board, with two dual core processors & ram. At best this will be priced just below dual socketed opterons, in fact that stuff I am hearing from AMD indicates these chips will simply be Opterons that are able to run with non-ecc/non-registered RAM. If that is the case WHO CARES.

I would love to buy a platform that allowed me to run dual X2 3800s, but that isn't going to happen, so unless AMD goes crazy on the pricing, I would rather spend the expected trivial difference in price and just buy the server grade stuffs.
 
hardwarephreak said:
I think a few of you are missing a rather glaring piece of information...this will not work with standard desktop chips.

You cannot stick two 3800+ processors into one of these boards/configuations.

There have been a few key points released from AMD that this will be compatible with a FX-type chip only...expensive, high end consumer stuffs...

For the price of buying two current FX processors (AM2), you could buy a dual socket 940 board, with two dual core processors & ram. At best this will be priced just below dual socketed opterons, in fact that stuff I am hearing from AMD indicates these chips will simply be Opterons that are able to run with non-ecc/non-registered RAM. If that is the case WHO CARES.

I would love to buy a platform that allowed me to run dual X2 3800s, but that isn't going to happen, so unless AMD goes crazy on the pricing, I would rather spend the expected trivial difference in price and just buy the server grade stuffs.
Click to expand...


The question here is whether regular (non FX) Socket F Opterons will be able to run on the motherboard. If so, I am sure we will once again see a wave of people buying the cheapest opty's and overclocking them like crazy. As for the memory, the non-registered ram will save users half the cost of ram or so depending on brand and quality.
 
Only 1xxx series opterons lack the requirement for ecc/reg RAM (unless something has changed that I am not aware of), so the fact that you would need a 2xxx series processor kind of kills the "buy cheap opterons' option.
 
i was looking more along the lines of eventually being able to nab two (low end) k8l's...
 
The real question is what form factor will this be? If it is bigger than ATX (12x9.6) they will have some issues in being limited to certain cases. Cases for E-ATX are fewer in choices and not everyone wants to be limited to E-ATX cases.
 
Hmmm, the more i read it, the more I like it. It's very very simple: 2 is better than 1. With 2 sockets you have capabilities for more. Sure at launch you may be limited to dual core CPUs, but 2 is still better than 1. Dual socket and single socket are apples and oranges. If you want a [H]ardcore system, it's gotta be dual socket. In the past there was an excuse for not having dual socket systems, and that they are server platforms and not ideally suited for HL2 and 2142 at the same time (but they do it very well when asked). Server platforms are limited on OC capabilities.

Now with 4x4 there is no excuse for the wanna be [H]ardcore, it's go dual sockets or go home. (the truely [H]ardcore already have dual sockets).
 
well from what i am looking at there is a simple workaround here.

yes it will use fx chips
yes it will use opteron socket f as well.

the memory controller is on the cpu not the mobo.
all the mobo manufacturer needs to do is add support for registered dimms.
i.e. 2 registered dimm sockets
2 non registered dimm sockets

problem solved

new can of worms opened.
enthusiast mobo with only 2 dimm slots.

another possibility is registered dimms for all slots and if you drop in an fx chip they work anyway. this is a product for a very narrow market i.e. AMD.

this is nothing more than AMD selling damn few systems with the most they can throw at it and calling it the fastest platform is offered by AMD. in other words this isn't a product looking for a market this is a marketing gimic looking to be a product.


the initiative to move to 4 or 3 pci-e slots is a long battle between the chipset/vid card vendors and the mobo makers/consumers. one camp wants it the other doesn't. mobo makers don't want it cause it increases r&d costs and manufacturing costs. consumers don't want it cause it is a waste of money right now with no software to take advantage of a 3rd vid card to act as a ppu (parallel processing unit) to accelerate floating point calculations. soon we will have a few games and apps that will but right now all we have is a special build of folding@home that only needs one x1900 and it can be your primary and only vid card not a secondary or tertiary card. although the havok engine is going to use the second vid card or a third if your doing sli or crossfire and hence super mega power users will need 3 slots. also 4 slots could net you octal sli or crossfire (with the new connector in the 1950 series cards and dual gpu cards).


this is a marketing gimmick right now though and a few folks with cash will buy the few boards made and run 4 cores (2 dual core cpus) along with 4 crossfire x1950's or 4 g80's or 2 7950gtx's
 
They need to drop the whole ECC / Registered support from the board. No ECC / Registered DIMMS. That's for server boards. Period.

all things being equal, 2 sockets are better than 1 all day everyday. I have 2 sockets (K8WE and dual opterons) and I will never go back.

The whole 4x4 thing is a bit of an annoyance as previously only the truely [H]ardcore had dual sockets, now 4x4 is going to diminish the exclusive dual socket club.

It makes sense, with dual sockets you have a huge advantage. Now intel has to get 8 cores per chip to compete against an AMD 4 core offering, and continually have double the amount of cores just to stay equal. Intel will never have the more cores per system crown if they don't go with dual sockets. I imagine intel will offer a dual socket (non-server) board soon.
 
hardwarephreak said:
Only 1xxx series opterons lack the requirement for ecc/reg RAM (unless something has changed that I am not aware of), so the fact that you would need a 2xxx series processor kind of kills the "buy cheap opterons' option.
Click to expand...

No, no AMD64 including no Opterons require ECC. All can use it, though.

Opteron 1x for socket 940 do require registered.

The requirement for registered or unbuffered is in the socket (before 4x4), not the name.
 
did i miss it, or has no one mentioned the increase in memory bandwidth afforded by dual sockets?

It seems to me that bringing multiple sockets to the masses (at least the non server market) is the fruition of amd's whole architecture in athlon 64.
 
Watch my quad-core intel run faster, cooler, making less noise, using less power than this non-sense 4x4.

Interesting how laptops with core 2 duo perform far better than dual-cpu single core systems. Get the point?

We need more cores and not more sockets.
 
uOpt said:
No, no AMD64 including no Opterons require ECC. All can use it, though.

Opteron 1x for socket 940 do require registered.

The requirement for registered or unbuffered is in the socket (before 4x4), not the name.
Click to expand...


I knew that 940 had the requirement, I just thought that from 939 forward (socket AM2 1xxx series currently), all single socket Opterons lacked the registered ram requirement. And yes, ECC is not required, however, most (granted not all) ECC ram is also registered.
So kill me for making a generalization :D

Clarification is never a bad thing...
 
floodo1 said:
did i miss it, or has no one mentioned the increase in memory bandwidth afforded by dual sockets?

It seems to me that bringing multiple sockets to the masses (at least the non server market) is the fruition of amd's whole architecture in athlon 64.
Click to expand...

No you haven't missed it, just all the naysayers have. They don't understand what a dual socket system is all about.... And probably never will.

spectrumbx said:
Watch my quad-core intel run faster, cooler, making less noise, using less power than this non-sense 4x4.

Interesting how laptops with core 2 duo perform far better than dual-cpu single core systems. Get the point?

We need more cores and not more sockets.
Click to expand...

We'll see about that. I can't say which will be faster, etc, but I can say that it won't be that cut and dry. You silly notion about laptops being faster? totally false, they are still limited by 5400 (sometimes 7200) RPM drives. My dual socket (opteron) system has a 4 drive RAID 0 15,000 RPM disk array and it's way faster than my dual core laptop.

Regarding CPU's... I've tested my dual socket system with a single dual core opteron (2.2 GHz) And with two single core opterons (again 2.2 GHz) and they perform the SAME!!

So get over it, 2 cores x 2 sockets is the SAME and will perform the SAME as 4 cores on 1 socket (2 sockets will give you twice the memory bandwidth and NUMA).

NUMA makes a huge difference and you CAN'T get it with a single socket!!!!

More sockets IS more cores, silly (along with MORE of alot of other things).
 
wildcard442 said:
NUMA makes a huge difference and you CAN'T get it with a single socket!!!!
Click to expand...

In all those years that the Opterons with NUMA are on the market I have never seen a benchmark show a case where it gives you a huge advantage. Nothing > 3 % real world performance except for one test where nothing but one designated database server was running it's own benchmarking.

Care to share some links?
 
uOpt said:
In all those years that the Opterons with NUMA are on the market I have never seen a benchmark show a case where it gives you a huge advantage. Nothing > 3 % real world performance except for one test where nothing but one designated database server was running it's own benchmarking.

Care to share some links?
Click to expand...


You say 3% is nothing? lol
 
It's not that 3% is nothing, but it's still pretty marginal. Besides, that 3% increase from NUMA is pretty much the maximum, I would bet that it averages out to be even lower, because pretty much nothing is written to make use of NUMA.
 
Slartibartfast said:
It's not that 3% is nothing, but it's still pretty marginal. Besides, that 3% increase from NUMA is pretty much the maximum, I would bet that it averages out to be even lower, because pretty much nothing is written to make use of NUMA.
Click to expand...

Thats right, the problem is that you need the OS to support NUMA as well as the application itself to support NUMA. Now, therer is an advantage to using NUMA and AMD is pushing it dead right center: multitasking. Assuming that you are using a NUMA aware OS say WinXP64bits and you are running concurent programs, each application does not have to share memory access which will boost performance.
 
wildcard442 said:
No you haven't missed it, just all the naysayers have. They don't understand what a dual socket system is all about.... And probably never will.



We'll see about that. I can't say which will be faster, etc, but I can say that it won't be that cut and dry. You silly notion about laptops being faster? totally false, they are still limited by 5400 (sometimes 7200) RPM drives. My dual socket (opteron) system has a 4 drive RAID 0 15,000 RPM disk array and it's way faster than my dual core laptop.

Regarding CPU's... I've tested my dual socket system with a single dual core opteron (2.2 GHz) And with two single core opterons (again 2.2 GHz) and they perform the SAME!!

So get over it, 2 cores x 2 sockets is the SAME and will perform the SAME as 4 cores on 1 socket (2 sockets will give you twice the memory bandwidth and NUMA).

NUMA makes a huge difference and you CAN'T get it with a single socket!!!!

More sockets IS more cores, silly (along with MORE of alot of other things).
Click to expand...

Blah blah blah!
I remember clearly seeing benchmarks of dual core opteron 165 beating dual cpu 265. NUMA certainly did not help the 265... did it?

I do not know why you are trying to introduce hard disk performance in the equation, when my point about the laptop was about cpu efficiency (performance/power consumption).

Multiple cpu sockets are nothing new, multiple cpu cores are. Hence, let's save all excitement for more cores on a single socket.
 
spectrumbx said:
Blah blah blah!
I remember clearly seeing benchmarks of dual core opteron 165 beating dual cpu 265. NUMA certainly did not help the 265... did it?
Click to expand...
If you don't program for the new technology, you don't get the benefit of it. Nobody complains that (faster) single cores get better scores than (slower) multiple cores, because they have more scaling potential. There are more FLOPS happening on the multi-core CPU, so the potential to go faster than the single-core is there. Likewise, there are more memory reads and writes happening with two memory controllers.

What do you think 16 cores with one memory controller would be like? My guess is, pretty slow. At some point you're gonna need more memory bandwidth, so at some point you need to have more memory controllers. It's a good idea to get that framework (and the programming support for "local" memory) in place now, rather than hitting a brick wall later.
spectrumbx said:
Multiple cpu sockets are nothing new, multiple cpu cores are. Hence, let's save all excitement for more cores on a single socket.
Click to expand...
Why? If you can get more cores on a dual-socket board, how is that not exciting? It may not be cost effective, but there are plenty of people who don't care about cost-effective, just speed.

That's my two cents, anyways. The computer industry is gonna go there eventually, so why not now?

 
Uh, 165 and 265 are both dual core at 1.8 ghz. Assuming it was a single 165 vs a single 265, the 265 would be running ecc/reg memory. I'd be curious to see if they bothered running the memory at the same speed/timings.
 
Slartibartfast said:
Uh, 165 and 265 are both dual core at 1.8 ghz. Assuming it was a single 165 vs a single 265, the 265 would be running ecc/reg memory. I'd be curious to see if they bothered running the memory at the same speed/timings.
Click to expand...

I meant the 244 (single core conterpart).
I can't keep up with all that cpu numbering. :)

Those benchmarks were all out there when dual core was first out.
You couldn't have missed it.

When a multi-core cpu is created, its multi-socket conterpart is also created.
So, again, nothing new.
A quad-socket system is even at the reach of who ever cares enough to own it. So, why are we getting excited about dual-sockets?

This gimmick should be driven by a mobo manufacturer and not AMD.

I need AMD to stop with the gimmicks and work on what matters.
 
You must log in or register to reply here.
Back
Top