AMD to integrate PCIe

[Peter]

Originally posted by: humey
1st Peter, i didnt say you said fsb, i just included the info.

2nd Yes sure 940 is still around and for sale but its being replaced on desktops you now get 939 FX's not 940

3rd I dont give a toss what your job is ive heard it all before, im a engineer and it dont mean all i worked with were good or knew wtf they were doing.

One of my buddies in my irc channel that works together with MS on beta stuff used to work with AMD (fixed machines that made wafers), so i dont need you or any others apart from him to edu me on the tech, i take his word over anyone elses.

He seems to know whats going on well into 2007 and all he has said in past is now real so i dont doubt him.

If you so clued up on mobos you should have known 940 pin is being phased out and all FX's now are 939pins and use normal ram.

940 has never been intended for desktop. It's alive and doing extremely well in its intended segment, servers, blades, workstations. ECC RAM is highly welcome and even a must there, "registered" DIMM tech is required for high chip count, thus letting 940-pin CPUs drive four times as much RAM as "unbuffered" 939. Speed doesn't always matter the most ...

Sure, the 940-pin FX chips for high end enthusiast gaming were just a stopgap until 939 was ready. No news there.

754 isn't dead either, in fact, this is what the big OEMs are shifting into the budget PC market in massive quantities, and it's the ONLY platform for AMD notebooks. So we got mobile Semprons, Athlons and Turions, and desktop Athlon and Sempron processors. So the "Athlon" brand is phasing out on this socket, but Turion and 64-bit enabled E6 step Semprons keep 64-bit support on this platform alive. Dead? AMD doesn't think so. But of course, you're the expert.

Others have already commented on your general attitude, so I'll spare myself from repeating what they said.

 

[imported_humey]

Jeff7181, Your tha ass so STFU and butt out something i didnt address to you.

Just like the n00b i did address it to you are inventing stuff i never said.

I said the 940 pins FX is dead is now repalced by 939 and it does jsut as well on non regged ECC ram and ram is cheaper.

I said 940 is still used for opterons but will be replaced by 1207.

OEM, who cares, they sell sh1t thats been on shelf for long enough, most in the know wont buy parts to build a 754 socket rig now.

Your the one insulting me calling me a "ass" and "my buddy" knows 100x more than anyone here and i wont learn from reading your b0ll0cks.

The two of your are right n00bs, the FX was a desktop cpu and was 940 pins so it was made for desktop even if it was to fill a gap till 939 came out it still meant you had pay premium for regged ECC ram.

Personally i dont really care nothing i post here is made up or guesswork its as good as it gets, i will try get my buddy here to post but you prob think he is a making it all up.

He prob aint intersted in getting into flames in forums anyhow, as he knows what others dont and dont need to prove himself, i aint talking some 20 year old punk kid here BTW and i aint either.

Peter i dont care what you and Jeff think of my attitude, you 1 fags can go rent a room, i post better info here then 905 of peeps and sometime i dont kn ow why i bother but i suppose its for the good 10% of users.

---

You are not that good, and you appear to be preoccupied with your own sexual inadeqacies. Your account on our forums is locked.

Good bye.

AnandTech Moderator

 

[Jeff7181]

Originally posted by: humey
Jeff7181, Your tha ass so STFU and butt out something i didnt address to you.

Just like the n00b i did address it to you are inventing stuff i never said.

I said the 940 pins FX is dead is now repalced by 939 and it does jsut as well on non regged ECC ram and ram is cheaper.

I said 940 is still used for opterons but will be replaced by 1207.

OEM, who cares, they sell sh1t thats been on shelf for long enough, most in the know wont buy parts to build a 754 socket rig now.

Your the one insulting me calling me a "ass" and "my buddy" knows 100x more than anyone here and i wont learn from reading your b0ll0cks.

The two of your are right n00bs, the FX was a desktop cpu and was 940 pins so it was made for desktop even if it was to fill a gap till 939 came out it still meant you had pay premium for regged ECC ram.

Personally i dont really care nothing i post here is made up or guesswork its as good as it gets, i will try get my buddy here to post but you prob think he is a making it all up.

He prob aint intersted in getting into flames in forums anyhow, as he knows what others dont and dont need to prove himself, i aint talking some 20 year old punk kid here BTW and i aint either.

Peter i dont care what you and Jeff think of my attitude, you 1 fags can go rent a room, i post better info here then 905 of peeps and sometime i dont kn ow why i bother but i suppose its for the good 10% of users.

Thanks for proving my point.

 

[imported_humey]

No problem wanker im happy to help and im going to for once just ignore any further posts by you 2 lamers as this is my thread and its a good topic and you 2 are fooking it up for good convo by others, you are so brave online aint you ?.

 

[biostud]

always fun when people actually have the same opinions, but misunderstand eachother and start spewing their anger towards eacheother :roll:
What we all agree on I guess :
Workstations/Servers: S940 -> S1207 (Socket F)
Desktops: S754 -> S939 -> M2 (939 or 940 cant remember which, but not compatible with current CPU's)
FX line: S940 -> S939 -> M2? (Most likely IMHO)

Was that so hard?

 

[cKGunslinger]

Originally posted by: humey
You are not that good, and you appear to be preoccupied with your own sexual inadeqacies. Your account on our forums is locked.

Good bye.

AnandTech Moderator

<Nelson> Ha-ha! </Nelson>

 

[Zebo]

Originally posted by: Fox5

Originally posted by: Megamixman
God, CPU's are slowly becoming microcontrollers on steroids.

Which has some problems...the bugs are now on the cpu and not on the chipset, and there's limits to what they can do. AMD's cpu already has some memory problems, what if it has problems with this? That, and it is slowly removing the point of a motherboard, they're going to be little more than connectors soon, well, I guess anybody will be able to make a decent AMD mobo then.

BTW, didn't intel want to make an all in one cpu at one point?

Will this really improve performance much though? Wasn't aware there was that much cpu->gpu communication anymore.

Well, this gives me a reason to hold off on an upgrade for even longer, and I thought an X2 system with fast DDR would be enough for a while...on the other hand, graphics bus performance is another area where AMD has lagged behind Intel, despite faster cpus, so this will at least look good for theoretical performance.

Links to any of that crap you just spouted?

Memory problems? You mean it's a problem to perform better?

It is clear that DDR memory manufacturers are paying close attention to the Athlon 64 platform, since we are seeing familiar memory reaching further on Athlon 64 than what we saw on our Intel 478 platform

http://www.anandtech.com/printarticle.aspx?i=2226

Seriously Fox5 don't be spreading FUD, makes you look like an idiot fanboy. Real links, real sources, real facts.

 

[ProviaFan]

The only thing I would be concerned about is if they didn't put enough PCI-E lanes on the socket for future expansion, and ended up changing the socket later on to compensate... I suppose they could add extra "lower performance" lanes on the *bridge, kind of like workstation boards today often have two PCI-X busses, one faster at 133MHz and the other at 100MHz. Anyway, I'm :thumbsup: for progress, even if it means that what I have now becomes obsolete (because I just get to buy something new a few years from now)

 

[biostud]

Originally posted by: ProviaFan
The only thing I would be concerned about is if they didn't put enough PCI-E lanes on the socket for future expansion, and ended up changing the socket later on to compensate... I suppose they could add extra "lower performance" lanes on the *bridge, kind of like workstation boards today often have two PCI-X busses, one faster at 133MHz and the other at 100MHz. Anyway, I'm :thumbsup: for progress, even if it means that what I have now becomes obsolete (because I just get to buy something new a few years from now)

With the nvidia professional chipset you can add more PCIe lanes.

http://anandtech.com/cpuchipsets/showdoc.aspx?i=2327&p=4

 

[Fox5]

Originally posted by: ZobarStyl

Originally posted by: Fox5
Which has some problems...the bugs are now on the cpu and not on the chipset, and there's limits to what they can do. AMD's cpu already has some memory problems, what if it has problems with this?

What problems? Links to back up such claims should go without saying.

Will this really improve performance much though? Wasn't aware there was that much cpu->gpu communication anymore.

Oh yeah, they hardly ever speak anymore. They don't send each other any information, the image just magically appears on the screen.

graphics bus performance is another area where AMD has lagged behind Intel, despite faster cpus, so this will at least look good for theoretical performance.

Again, link? I wasn't aware of any graphical performance area (games) where AMD has lagged behind Intel. Considering how long SLi solutions have available for AMD mobo's vs Intel's, I wouldn't say the bus performance has been an issue for AMD.

I'm not trying to hop on your case, but you post looks like a lot of misinformation to me.

AMD's memory controller used to downclock to 333mhz if there were too many ram chips, though now it just goes to 2t command rate.

Currently the cpu and gpu don't communicate enough to saturate an AGP 8x bus, so it doesn't appear that's a limitation right now, so there may be little to no performance boost from lowering the latency. Plus, graphics are more bandwidth limited than latency aren't they? Having PCI-E on the chip could bring along problems that can't be fixed without a cpu revision from AMD, and if there is no real performance benefit from doing this(in games at least, I'm sure there would for 3d rendering programs which seem to tax the bus more) then it would be a negative.

And Intel seemed to generally outperform AMD in the 3dmarks when they were purely testing the video card.

 

[Drayvn]

If they were purely testing the video card in 3dmarks.... then wouldnt it be the video card thats making the difference...

And anyway does 3dmark show real world performance of gpus? No.

 

[imported_michaelpatrick33]

Originally posted by: Fox5

Originally posted by: ZobarStyl

Originally posted by: Fox5
Which has some problems...the bugs are now on the cpu and not on the chipset, and there's limits to what they can do. AMD's cpu already has some memory problems, what if it has problems with this?

What problems? Links to back up such claims should go without saying.

Will this really improve performance much though? Wasn't aware there was that much cpu->gpu communication anymore.

Oh yeah, they hardly ever speak anymore. They don't send each other any information, the image just magically appears on the screen.

graphics bus performance is another area where AMD has lagged behind Intel, despite faster cpus, so this will at least look good for theoretical performance.

Again, link? I wasn't aware of any graphical performance area (games) where AMD has lagged behind Intel. Considering how long SLi solutions have available for AMD mobo's vs Intel's, I wouldn't say the bus performance has been an issue for AMD.

I'm not trying to hop on your case, but you post looks like a lot of misinformation to me.

AMD's memory controller used to downclock to 333mhz if there were too many ram chips, though now it just goes to 2t command rate.

Currently the cpu and gpu don't communicate enough to saturate an AGP 8x bus, so it doesn't appear that's a limitation right now, so there may be little to no performance boost from lowering the latency. Plus, graphics are more bandwidth limited than latency aren't they? Having PCI-E on the chip could bring along problems that can't be fixed without a cpu revision from AMD, and if there is no real performance benefit from doing this(in games at least, I'm sure there would for 3d rendering programs which seem to tax the bus more) then it would be a negative.

And Intel seemed to generally outperform AMD in the 3dmarks when they were purely testing the video card.

Once you start having massive amount of database and server data moving on that PCI-ex bus through the raid cards etc. then we will talk about performance latencies. This isn't about a home user, this is about dual/quad core Opterons running datacenters with all low latency components. Hypertransport along with integrated memory and now PCI-ex.

Intel outperforms AMD in 3dmarks? Show many any web review site that show Intel winning gaming performance at a non video card bound resolution. AMD did have memory downclock with the older cpus with 4dimms on the 939. I don't know if the ECC 940 chips where more memory was utlized had that same issue.

 

[Zebo]

The guys is not all there... usually I loose all intrest when some noob mentions 3dmark and go ::sigh clueless:: moving on to the next thread.. but I'm still interested in this less memory performance and CPU "bugs" AMD supposedly has.

 

[Fox5]

Originally posted by: Drayvn
If they were purely testing the video card in 3dmarks.... then wouldnt it be the video card thats making the difference...

And anyway does 3dmark show real world performance of gpus? No.

No, but it's an area where an Intel system and an AMD system with the same video card will show the Intel system winning by a small margin. Intel also tends to have slightly better performance in extremely gpu limited cases. Seems to indicate intel has better graphics interface performance, it's not significant at all but it's something, just like people who used to buy Intel systems because they tested higher on memory benchmarks.
Also the fact that this advantage will almost never show in any real situation makes it doubtable that amd improving on it would either, the limit will generally be the cpu or the video card, not the interface between the two.

Once you start having massive amount of database and server data moving on that PCI-ex bus through the raid cards etc. then we will talk about performance latencies. This isn't about a home user, this is about dual/quad core Opterons running datacenters with all low latency components. Hypertransport along with integrated memory and now PCI-ex.

Intel outperforms AMD in 3dmarks? Show many any web review site that show Intel winning gaming performance at a non video card bound resolution. AMD did have memory downclock with the older cpus with 4dimms on the 939. I don't know if the ECC 940 chips where more memory was utlized had that same issue.

Well I was talking about the home user, didn't realize this pci express integration is only applying to the server market and that socket M doesn't have it. I think it will have a benefit in the non gaming market.

For that matter, the fact that Intel often does win at video card bound resolutions shows an advantage. Of course, you could always overclock the agp or pci express bus and outperform in those situations, but it's not worthwhile to do so at all.

but I'm still interested in this less memory performance and CPU "bugs" AMD supposedly has.

The guy above you noted they existed. Memory downclock used to happen with too many ram chips(not the number of sticks, that was just a coincidence), and still drops down to 2t command rate with too many. The performance benefit of the integrated memory controller outweighs these disadvantages, but if a similar thing happened with the pci express bus, well the integration of the bus might not outweigh the disadvantage.

 

[Peter]

Memory downclock is not a CPU bug, no matter how often this is repeated. If you look at the JEDEC specification for PC3200u, you'll find that it isn't even SUPPOSED to run with two DIMMs per channel. PC2700 is what's guaranteed for two-DIMM setups, and that's exactly what the older revision AMD64 CPUs achieved. The latest ones squeeze PC3200 (with degraded timings) out of it, that's a PLUS, not a bugfix.

As for CPU bugs, well yes, it has some. So does Intel's. No news there.

 

[Elcs]

Fox has a point somewhere in there.

Looking at the first A64 compared to the latest A64, the biggest 'problem' so to speak has been its on-die memory controller. Whilst its not buggy or holding back performance too much, it did take AMD a few goes to get it right for the end user. With the transition of games starting to shift the balance towards 2 Gb of RAM, 4x512 @ pc3200 speeds is probably much more fun than 4x512 @ pc2700 speeds.

AMD's PCI-E on CPU is another good idea by AMD but it 'can' present problems just as the memory controller did. Not enough lanes, how the lanes communicate with the cpu, lanes not fast enough,.... this route seems a lot more difficult than putting an on-die memory controller on (Im non-technical).

Kudos for AMD's effort. I'll be awaiting more informations though

 

[Zebo]

. The performance benefit of the integrated memory controller outweighs these disadvantages, but if a similar thing happened with the pci express bus, well the integration of the bus might not outweigh the disadvantage.

Well there you go so what's the problem? *If* you drop to 2T still performs better than any intel solution so what's the problem? And BTW many run 1T with 2 Gigs so it's mute anyway you just have to know which sticks to buy (crucial Micron D). Thrid we;ve shown 2T hardy impacts performance at all! Look here:
http://forums.anandtech.com/messageview...atid=28&threadid=1577496&enterthread=y
Fouth Intel chipset drop CMD when you add more ram and speed so I don't know why you think this is unique to AMD.

PCIe bus? Don't worry your little head AMD engineers got all this stuff fiqured out... Intel P4 was the only chip(s) in history slower than thier predecessors mhz to mhz.. I don't expect history to repeat itself. They would'nt include it unless it's benefits outwieghed whatever disavantages your imagining at the moment.

 

[Fox5]

We'll see what happens.

BTW, P4 is not the only chip in history to perform worse than its predecessor mhz to mhz, though most usually eventually overcome it as software gets optimized for them. It's kind of a moot point that P4 never beat P3 mhz per mhz since P4 reached much higher speeds, performance levels P3 could never reach. Even if you took the P3 design as it was and shrunk it down until it could reach 3ghz it would never perform as well as a P4, mainly due to its slow memory bus, and this is an equally unimportant point as trying to say P4 was a bad design because it didn't mhz per mhz outperform the P3.

 

[Drayvn]

I thought the Pentium M was designed from a P3 architecture tho... And the P-Ms are pretty damn good.

 

[Zebo]

Oh uh..we might be waiting a while FOX5 This is the socket 754/939 blunder all over again.

 

[coldpower27]

Interesting so that why we have Socket F (1207) for Opterons in the future with DDR2-??? and Integrated PCI-E

Right now this is what I see:

S754 On Desktop it's for Semprons, on Mobile it's for both.

S939 Mainstream/Enthusiaist Socket, for Athlon 64's and Athlon FX on Desktop.

S940 Opteron Server/Workstation

In the future:

Socket S1 for Mobile Processors.

Socket M2 (940) Dual DDR2 Note: Not the electrically the same as currently S940 used on Opterons.
Designed for the Sempron, Athlon 64, Athlon 64x2, Athlon FX lines.

Socket F (1207) Registered Dual DDR2 with Integrated PCI-E as well. Opterons Server/Workstation