• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

I wish AMD would do this

C

colinstu

2[H]4U
Joined
Oct 11, 2007
Messages
3,563
I have never owned an AMD system, but it is nice to have competition.

I'd like them to stick with the K8, it's been used for years and continued to work.

They should do this to these: 5200+, 5600+, 6000+, 6400+

1) Make them 65nm
2) SSE4

3) increase the L2 some more?
4) Put two dies on chip, hey it may be a "fake" quad core experience but 2 is better then 1
5) Bump up the HT? Bump up the clocks?

I want my Q9450!
 
last i heard they ARE keeping the X2 K8 cpus around... not because its a good move but because they have to. they dont have the k10 yields to completely drop them.

I think AMD fucked up by dropping their 939 line. I'd have bought a top of the line X2 if they were still available for 939. Now there's no point for me to upgrade to AM2/+ since it'd be the same price and less performance as upgrading to an intel setup. Even though I really really want to buy AMD over intel , the prices are too close to justify it. Im hoping for another 50 dollar drop in phenom 9500s in a couple months or else im going Intel
 
colinstu said:
I have never owned an AMD system, but it is nice to have competition.

I'd like them to stick with the K8, it's been used for years and continued to work.

They should do this to these: 5200+, 5600+, 6000+, 6400+

1) Make them 65nm
2) SSE4

3) increase the L2 some more?
4) Put two dies on chip, hey it may be a "fake" quad core experience but 2 is better then 1
5) Bump up the HT? Bump up the clocks?

I want my Q9450!
Click to expand...
You want a Q9450 yet post all the improvements AMD should perform on their current line of processors?
 
Intel isn't going to release their new chips if they don't have too... there is no threat from AMD right now so they are going to milk the Q6600 as long as they can.
 
colinstu said:
Intel isn't going to release their new chips if they don't have too... there is no threat from AMD right now so they are going to milk the Q6600 as long as they can.
Click to expand...
Yes, just like NVidia has done with their 8000-series cards. There hasn't been such a lack of competition in the computer industry in a long time. Hopefully 2007 will be remembered as a bad year and things will return to a more competitive market in 08 and beyond.
 
To be honest I see no reason to dump my 939 setup. Far more likely my next upgrade will be a GPU, not a CPU.
 
osalcido said:
I think AMD fucked up by dropping their 939 line. I'd have bought a top of the line X2 if they were still available for 939. Now there's no point for me to upgrade to AM2/+ since it'd be the same price and less performance as upgrading to an intel setup.
Click to expand...

LstOfTheBrunnenG said:
To be honest I see no reason to dump my 939 setup. Far more likely my next upgrade will be a GPU, not a CPU.
Click to expand...

Both QFT - my 939 Opty 170 at 2.6Ghz does the job, and I'm upgrading the pair of X1900s to 3870s next (no point going nVidia as I'd need a new board as well). I'd still buy a better socket 939 CPU if one was released though. Not much point picking one up as is since 2.6 was as fast as they ever got, and an FX-60 that might give me more headroom is way overpriced.
 
I think AMD fucked up by dropping their 939 line. I'd have bought a top of the line X2 if they were still available for 939. Now there's no point for me to upgrade to AM2/+ since it'd be the same price and less performance as upgrading to an intel setup. Even though I really really want to buy AMD over intel , the prices are too close to justify it.
Click to expand...

QFT!!!!
 
colinstu said:
I have never owned an AMD system, but it is nice to have competition.

I'd like them to stick with the K8, it's been used for years and continued to work.

They should do this to these: 5200+, 5600+, 6000+, 6400+

1) Make them 65nm
2) SSE4

3) increase the L2 some more?
4) Put two dies on chip, hey it may be a "fake" quad core experience but 2 is better then 1
5) Bump up the HT? Bump up the clocks?

I want my Q9450!
Click to expand...

The Phenoms are faster than the K8s (clock for clock) - why would you want them to take a step backwards?
 
The K8s could beat the phenoms if they upgraded them. Unless they fix all the bugs with K10...

Wish AMD would stop f'ing around.
 
My top 5 things AMD needs to do in 2008.

1) Quickly rectify the anal/cranial inversion
2) Fix the TLB problems with Phenom
(Whether or not this is an issue for most desktop users is a moot point, it's so overblown that something needs to be done. Hopefully that fix is B3.)
3) Die shrink and push out the MHZ and/or Cores.
4) Create a Mobile platform that will rival lntel's Centrino line
5) Turn a damn profit

That's my short list, I'm sure there's many more (Hector Ruiz being #6) but those are the most important I feel.
 
SulSeeker said:
My top 5 things AMD needs to do in 2008.

1) Quickly rectify the anal/cranial inversion
2) Fix the TLB problems with Phenom
(Whether or not this is an issue for most desktop users is a moot point, it's so overblown that something needs to be done. Hopefully that fix is B3.)
3) Die shrink and push out the MHZ and/or Cores.
4) Create a Mobile platform that will rival lntel's Centrino line
5) Turn a damn profit

That's my short list, I'm sure there's many more (Hector Ruiz being #6) but those are the most important I feel.
Click to expand...
Personally, I think the #5 should have been #1 a long time ago and no comment on the proposed #6. There's a whole thread already devoted to that one...
 
There is no point increasing the cache on a processor with a memory controller slapped into it and with HT technology. That'd just be dead weight.
 
VanFanel89 said:
There is no point increasing the cache on a processor with a memory controller slapped into it and with HT technology. That'd just be dead weight.
Click to expand...
Actually there's differing opinions on additional cache and how that would impact performance with AMD processors. AMD has often stated it wouldn't make a big impact in order to bolster their architecture, but I imagine a little extra cache wouldn't hurt performance. It definitely would increase the cost and probably reduce yields, but whether all the negatives are worth it remains to be seen.
 
VanFanel89 said:
There is no point increasing the cache on a processor with a memory controller slapped into it and with HT technology. That'd just be dead weight.
Click to expand...

Can't say I agree there. While true that extra cache isn't as benifical to Athlon/Phenom as it is to the Core architecture, it does help, more so in some applications than in others.
 
RamonGTP said:
Can't say I agree there. While true that extra cache isn't as benifical to Athlon/Phenom as it is to the Core architecture, it does help, more so in some applications than in others.
Click to expand...

Indeed, cache is still much faster than memory, even when you do have the controller built into the CPU. On Core2 you have so much cache that some routines can operate almost entirely from cache. It goes way beyond just storing the most recently used items. It's now acting as a sort of superfast 'scratch-pad', where nearly all of the processing is done. If you hit the 'sweet spot', you can even see a big difference on K8 between the 2x512 kb and 2x1 mb L2 models. But in most cases even 2x1 mb is far too small to truly act as a 'scratch-pad', and the CPU still needs to rely on memory a lot, so many people seem to be fooled into thinking K8 cannot benefit from larger caches.
The K8 has three problems:
1) The cache is rather small
2) The cache is rather slow
3) The cache has rather low associativity (basically it can't store data coming from many different locations well)

If they made the cache bigger AND faster, it'd have quite some impact. People often seem to overlook how much faster the Core2 cache is, besides just being larger. I've posted figures on this in the past. In some cases the Core2 cache is nearly three times as fast (mostly because it can read and write at the same time, making copy operations much faster).
Another thing I've discussed many times is that Core2 has a shared L2 cache, while NOT having higher latency than the K8's cache.

K10 has improved the cache architecture of K8 a bit, but it's not enough. It's still too small, still too slow. Not that much of a problem today, but in less than a year, Intel will also have an onboard memory controller. AMD isn't going to be ready when that day comes.
 
Why not take out L3 cache and improve L2 cache and put more HT and intergrated memory controller for each core and run numa. I get around ~9gb/sec for dual channel. Why not added a 2nd controller or 3rd and 4th get 18gb/sec, 28gb/sec 36gb/sec. Or add ddr 5 x 4 controllers 80gb/sec?
 
VanFanel89 said:
There is no point increasing the cache on a processor with a memory controller slapped into it and with HT technology. That'd just be dead weight.
Click to expand...


Generally true,but,would it not make sense to spend more transistors on adding more execution units,and beefing up the fpu even more ?

Core is still ahead in these areas,even after all the tweaking AMD did with K10.Increased clock speeds are not enough.
 
Manny Calavera said:
Generally true,but,would it not make sense to spend more transistors on adding more execution units,and beefing up the fpu even more ?
Click to expand...

When I was optimizing my code for Athlon64/Opteron, it seemed like the cache was the limiting factor. I couldn't saturate the execution units fully, because the cache couldn't deliver the data quickly enough. I was bandwidth-starved from the cache. The 2x1 mb models also did considerably better than the 2x512 kb models on my code.

So I'd be inclined to say that AMD had to solve the cache bottleneck first, before moving on to extra execution units. The SSE improvements aside, because those were a relatively poor feature on the K8. I think on the whole, AMD's execution core is now about as efficient, perhaps more efficient than Intel's Core2.
I think it's mainly limited by the cache design and the fact that it cannot be clocked at the same clockspeeds as Intel.
Intel has an advantage there, because they get their high IPC with a relatively long pipeline. They will probably always have a slight advantage in clockspeed because of that, even when AMD moves to 45 nm. The Core2 architecture is just designed to scale to higher clockfrequencies than K10 is. In a way it's an added bonus that they can match the K10's IPC, I think. Perhaps not even Intel expected that.
 
Higher clock speed won't be enough to beat Intel, but it is a very big step in the right direction and will make Phenom much more competitive. When Phenom was introduced, it was not a viable option at all IMO. It had a slower IPC in addition to lower clock, it was more expensive than the C2Q, and of course, there is the erratum issue.

Lowering the price was a good starting point for AMD, it made it a more viable alternative to many people, while still unacceptable to others. If they can fix the erratum and ramp up the clock speed on the next revision while still maintaining a lower price point, then Phenom, while still not the C2Q killer it was once hoped to be, will become a very viable alternative to C2Q.
 
RamonGTP said:
Higher clock speed won't be enough to beat Intel, but it is a very big step in the right direction and will make Phenom much more competitive. When Phenom was introduced, it was not a viable option at all IMO. It had a slower IPC in addition to lower clock, it was more expensive than the C2Q, and of course, there is the erratum issue.

Lowering the price was a good starting point for AMD, it made it a more viable alternative to many people, while still unacceptable to others. If they can fix the erratum and ramp up the clock speed on the next revision while still maintaining a lower price point, then Phenom, while still not the C2Q killer it was once hoped to be, will become a very viable alternative to C2Q.
Click to expand...

Agreed. I think you hit the nail on the head with this post.
 
Dan,has AMD related to either you,or Kyle when they plan to fix the L3 speed issue ?

And no I am not talking about the TLB crap.

The NB/+L3 on Phenom is supposed to be running at a higher speed then the cores.I think many here may not be aware of that... I am surprised so many reviewers glossed over it,or didnt bother to mention it at all.


Mega2 said
Why not take out L3 cache?
Click to expand...

That would only serve to slow it down in an even greater way.
 
Here's what I think AMD needs to do.
1) Oust Hector Ruiz
2) The other execs need to all take fat pay-cuts until the company becomes profitable, and that money should be re-directed to useful causes within the company.. (Paying off debt, R&D, Fab upgrade expenses, etc.)..

I was going to write more but the other suggestions have already been mentioned..
 
They need to increase the speed of their cache and the size. Once clock speeds increase the should be fine.

Remember intel is a micro process ahead of amd. The core 2s are on a much more mature process than amd's 65nm and now they are oving to 45nm
 
I don't care if they don't ramp clockspeeds, but the damn things are way too hot. No excuse for AMD's single-die quad using more energy than Intel's dual-die quad. This is especially true at idle, where AMD's independent scaling of the clock speed and voltage among the cores was supposed to make all the difference.
 
LstOfTheBrunnenG said:
I don't care if they don't ramp clockspeeds, but the damn things are way too hot. No excuse for AMD's single-die quad using more energy than Intel's dual-die quad. This is especially true at idle, where AMD's independent scaling of the clock speed and voltage among the cores was supposed to make all the difference.
Click to expand...


What an engineer plans years in advance on the drawing board @ AMD,and what the leave's the fab years later to go into our computers are a far,far cry from each other it
seems ! :)
 
SulSeeker said:
My top 5 things AMD needs to do in 2008.

1) Quickly rectify the anal/cranial inversion
2) Fix the TLB problems with Phenom
(Whether or not this is an issue for most desktop users is a moot point, it's so overblown that something needs to be done. Hopefully that fix is B3.)
3) Die shrink and push out the MHZ and/or Cores.
4) Create a Mobile platform that will rival lntel's Centrino line
5) Turn a damn profit

That's my short list, I'm sure there's many more (Hector Ruiz being #6) but those are the most important I feel.
Click to expand...

AMD is toast....Nehalem is coming... and what will AMD HAVE to fight it?

3.0 GHZ phenoms?

THe k11 is so far off... its not even mentioned

curtains...
 
Hey now that the OP mentioned it. why hasn't amd made a cpu with a FSB (or HTT) higher than 200? Intel's been moving on up every dang year. Although it does make me NOT buy expensive ram just to overclock if you know what i mean.
 
POPEGOLD said:
AMD is toast....Nehalem is coming... and what will AMD HAVE to fight it?

3.0 GHZ phenoms?

THe k11 is so far off... its not even mentioned

curtains...
Click to expand...

Man do you have an agenda or something? If you don't like AMD, go buy Intel then post on the Intel side about how much you hate AMD. Is AMD in a bad way? Yes. Does it help to have someone like you who craps on the post with something like 'AMD is going to DIE 2008 INTEL OWNZ U ALL!!11!!'? Not just no, but hell no, if fact it pisses people off and verges on trolling.

To the OP which I am sure someone has already said. I think AMD just got too ahead of themselves and shot themselves in the foot. The ATI aquition was needed but streched them too thin and we are seeing the aftermath of it. In the long run tho they will pull out and have a better product for the future because of it.
 
I'd like them to make a single-core K8, with a multiplier of 16, and a whole lot of cache (lots of cache, since there's room for it because it's only one core). Maybe 65nm.
 
I was hoping to buy a higher clocked 939 for my/a nvidia platform... :(
so I bought Intel instead, that just sounds so wrong
 
POPEGOLD said:
AMD is toast....Nehalem is coming... and what will AMD HAVE to fight it?

3.0 GHZ phenoms?

THe k11 is so far off... its not even mentioned

curtains...
Click to expand...
The Inq was reporting last week that Bulldozer was taken off the roadmap, so things at AMD must seem dismal at this point if the article holds any water. Intel is also delaying parts, but any 'issues' they're experiencing are nowhere near as serious and will require much less time to sort out.

I think AMD took a much bigger bite than they could chew when they acquired Ati. Retrospectively speaking, it was a huge mistake. I hope they can climb out of this one, but I for one will wait for the 45nm parts before I even think of purchasing a new processor from the green camp.
 
I love AMD, I really do. I really don't want to go back to Intel.

But, when Nehalem does come out, and Intel has an integrated memory controller, what will AMD do? If AMD keeps going on their pace, they will just get K10 cleaned up by the time that Nehalem comes out. If k10 is having a hard time keeping up with core2, how will it compete with Nehalem?

I used to be an Intel guy, and then I bought an opty 165 and overclocked, it, and I've NEVER looked back at Intel again. But, I am worried about what is going to happen to AMD this time next year.

What AMD needs to do, is to take as much FPU technology from ATI as they can, and put it into a CPU. I'd almost say, a tri-core CPU, but one core is a lot like a GPU, and all FPU operations would be handed to it. But how on earth would you do something like that? I have a feeling that x86 and GPUs don't play nice together, but there has to be some way to rape ATI. I mean, most of the stuff in games revolves around floating point math. Even if AMD turned out a processor that blew ass in everything else, but was good in floating point, I'd imagine it'd sell very well with gamers and scientists and researchers and professional modelers.

I guess that's why I'm CS, and not comp engi.

AMD needs to do something radical with ATI's technology. The way I see it is, that AMD is going to be in trouble this time next year if they keep going the way they are, so they should take some chances and not be afraid. They can take a big chance, because they have ATI, and they have technology that Intel clearly doesn't have, looking at their GPUs.
 
APOLLO said:
The Inq was reporting last week that Bulldozer was taken off the roadmap, so things at AMD must seem dismal at this point if the article holds any water. Intel is also delaying parts, but any 'issues' they're experiencing are nowhere near as serious and will require much less time to sort out.
Click to expand...
I wouldn't read too much into an INQ story. What was in development is still in development.

sdlvx, the eventual goal of Fusion is just that, heterogeneous cores that can be customized for different applications. But realize a brand new micro-architecture takes 3-4 years to develop, so people should not expect to see something radically new 1-2 years after the merger. Most project aren't even announced to the public until at least 1.5 years into developments.
 
That's good to hear.

We really need to give this guy a tag of some kind to lend weight to his words.
 
You must log in or register to reply here.
Back
Top