It seems like AMD is getting crushed...

[zsdersw]

Originally posted by: Viditor

It gives the K8L much lower latency (which is the most important metric for AMD), so it should give the K8L a distinct advantage.

I didn't know K8 had a latency problem either...

 

[zsdersw]

Originally posted by: JPH1121
You know the reason they don't have memory bandwidth problems is because HT+IMC gives them more than enough right...? It's not that they don't like a lot of bandwidth, it's just that BECAUSE they have so much, increasing it more doesn't help a lot...

Yes, I already knew that.

 

[myocardia]

Originally posted by: zsdersw

Originally posted by: Viditor

It gives the K8L much lower latency (which is the most important metric for AMD), so it should give the K8L a distinct advantage.

I didn't know K8 had a latency problem either...

It doesn't, but only because of HT and the on-die memory controller (mostly the memory controller).

 

[IgNite]

There are a three major bottle necks with the k8 that need to be resolved. One is out of order execution for interger instructions. Another is floating point operands. And more importantly the memory subsystem which includes better caching.

 

[zsdersw]

Originally posted by: IgNite
There are a three major bottle necks with the k8 that need to be resolved. One is out of order execution for interger instructions. Another is floating point operands. And more importantly the memory subsystem which includes better caching.

Right.. and if those things are fixed with K8L, the fixing of those issues will be responsible for any advantage it may have over Conroe/Woodcrest.. not HT and the IMC.

 

[Regs]

Originally posted by: IgNite
There are a three major bottle necks with the k8 that need to be resolved. One is out of order execution for interger instructions. Another is floating point operands. And more importantly the memory subsystem which includes better caching.

wth lol. I posted that.

 

[myocardia]

Originally posted by: zsdersw

Originally posted by: IgNite
There are a three major bottle necks with the k8 that need to be resolved. One is out of order execution for interger instructions. Another is floating point operands. And more importantly the memory subsystem which includes better caching.

Right.. and if those things are fixed with K8L, the fixing of those issues will be responsible for any advantage it may have over Conroe/Woodcrest.. not HT and the IMC.

What you seem not to be understanding, zsdersw, is that if all three of those things were "fixed" with K8L, but it didn't have the IMC, guess where the bottleneck would then be?

 

[zsdersw]

Originally posted by: myocardia
What you seem not to be understanding, zsdersw, is that if all three of those things were "fixed" with K8L, but it didn't have the IMC, guess where the bottleneck would then be?

I'm not missing that at all. K8 and K8L will have HT and an IMC.. which makes it irrelevant to imagine them without HT and an IMC.

Viditor said that HT and IMC will be what, if anything, puts K8L ahead of Conroe/Woodcrest. I'm saying that the core improvements will be what does that (if it is indeed faster).. not HT and the IMC.

 

[FelixDeCat]

Originally posted by: HipnoticTranz
AMD will be broke in 6 months. :thumbsup:

:shocked:

 

[Viditor]

Originally posted by: zsdersw

Originally posted by: myocardia
What you seem not to be understanding, zsdersw, is that if all three of those things were "fixed" with K8L, but it didn't have the IMC, guess where the bottleneck would then be?

I'm not missing that at all. K8 and K8L will have HT and an IMC.. which makes it irrelevant to imagine them without HT and an IMC.

Viditor said that HT and IMC will be what, if anything, puts K8L ahead of Conroe/Woodcrest. I'm saying that the core improvements will be what does that (if it is indeed faster).. not HT and the IMC.

I said the K8L core will be equivalent to Conroe, and that the HT and IMC (being things that Conroe is incapable of) would put them over the top in speed because reduced latency increases IPC...

 

[dmens]

Originally posted by: Viditor
I said the K8L core will be equivalent to Conroe

how'd you figure that?

 

[dexvx]

Originally posted by: dmens

Originally posted by: Viditor
I said the K8L core will be equivalent to Conroe

how'd you figure that?

All speculation ATM.

Another key feature that'll help is the shared L3, but reading the tech specs, besides the improved HT (and maybe IMC upgrades), there isnt really much of a difference.

 

[zsdersw]

Originally posted by: Viditor
I said the K8L core will be equivalent to Conroe, and that the HT and IMC (being things that Conroe is incapable of) would put them over the top in speed because reduced latency increases IPC...

And again I ask.. how is HT/IMC going to matter in the desktop and 1P and 2P server space? K8 can't utilize all of the memory bandwidth available with HT and its IMC right now.. and K8L on the current HT spec would merely come a bit closer to using all of that bandwidth. Adding more bandwidth isn't necessarily going to bring more to the party.

I still maintain that if K8L is the better performer, it will be because of core improvements.

 

[programmer]

AMD == cheap && AMD == fast

IMO, conroe this, athlon that. I want cheap, fast processors, and that is what AMD has for me now. I applaud AMD for pushing Intel along so they are now both producing fantastic products that are [relatively] less expensive than ever.

 

[SparkyJJO]

Originally posted by: OhioState
everyone likes to discuss the merits of X2-VS-C2D but can we agree that semperon totally pwns cerleron >?

Agreed. Celeron is pretty bad. Only once CPU series I know of is worse - VIA processors 😛

 

[Viditor]

Originally posted by: dmens

Originally posted by: Viditor
I said the K8L core will be equivalent to Conroe

how'd you figure that?

Well, as dexvx points out, it is just speculation at the moment and we should keep that in mind. But the core improvements of note for K8L (grabbed hastily from Wiki) are:

More aggressive prefetching (16 bytes to 32 bytes)
Out of order loads
128 bit wide Floating point units
Larger Out of Order (OoO) buffers
Greater number of entries in Branch Target Buffer
Probable new additions to micro-ops ROM

For Conroe, the major advantages appear to be:

Can decode and execute 4 commands per clock cycle (though I have yet to see an example where it actually does this, it has the ability and hence the headroom)

Process 128-bit SSE3 instructions without slowing down


While we certainly won't know how these 2 play out until samples are released, my own opinion is that K8L comes pretty darn close to Conroe in performance.
Add to that the reduced latency of HT and ODMC, and it seems to me that K8L will be crowned the new champ next summer...

 

[Viditor]

Originally posted by: zsdersw

Originally posted by: Viditor
I said the K8L core will be equivalent to Conroe, and that the HT and IMC (being things that Conroe is incapable of) would put them over the top in speed because reduced latency increases IPC...

And again I ask.. how is HT/IMC going to matter in the desktop and 1P and 2P server space? K8 can't utilize all of the memory bandwidth available with HT and its IMC right now.. and K8L on the current HT spec would merely come a bit closer to using all of that bandwidth. Adding more bandwidth isn't necessarily going to bring more to the party.

I still maintain that if K8L is the better performer, it will be because of core improvements.

For some reason you are focusing only on bandwidth and not on latency...latency is the key advantage here for the desktop systems.

 

[zsdersw]

And you think the latency situation for K8L is going to make a significant difference? We'll see...

 

[dmens]

Originally posted by: Viditor
For some reason you are focusing only on bandwidth and not on latency...latency is the key advantage here for the desktop systems.

well imo neither bandwidth nor latency is really that important when viewed relative to other factors. it'd be a waste for amd to spend resources to tune that metric.

as for the list of improvements, that's all nice, but how high can it clock, and how high will be power be at given frequencies? i noticed the stallion codename cores have really high power targets compared to intel planned products in the same time frame. merom is a prime example of frequency scalability of a part with a low power design target.

whatever, data should be coming along in 6 months.

 

[Regs]

Originally posted by: dmens

Originally posted by: Viditor
For some reason you are focusing only on bandwidth and not on latency...latency is the key advantage here for the desktop systems.

well imo neither bandwidth nor latency is really that important when viewed relative to other factors. it'd be a waste for amd to spend resources to tune that metric.

as for the list of improvements, that's all nice, but how high can it clock, and how high will be power be at given frequencies? i noticed the stallion codename cores have really high power targets compared to intel planned products in the same time frame. merom is a prime example of frequency scalability of a part with a low power design target.

whatever, data should be coming along in 6 months.

Cache latency is very important. Conroe's memory sub-system can beat AMD's IMC with "brute force" cache as AMD calls it. Whatever AMD wants to call it, Intel's system seems to be working more efficiently than AMD's. AMD has grown too large now to ignore the desk top market and they can no longer depend on Opteron revenue for future growth. They will have to deliver a new product the will satisfy their customers across the board. They know this and they plan to deliver. The only question is when.

There is no problem with brand loyalty. It's perfectly natural and it shows that they are doing their jobs trying to keep their customer base. Though to wait over a year for a upgrade seems like a severe penalty to pay for loyalty.

 

[Viditor]

Originally posted by: zsdersw
And you think the latency situation for K8L is going to make a significant difference? We'll see...

We have already seen an example of this when the ODMC was first introduced on the Athlons...
The decreased latency was estimated to increase performance by ~20%...
Ars-Technica

 

[harpoon84]

Originally posted by: Viditor
While we certainly won't know how these 2 play out until samples are released, my own opinion is that K8L comes pretty darn close to Conroe in performance.
Add to that the reduced latency of HT and ODMC, and it seems to me that K8L will be crowned the new champ next summer...

And you really think Intel will do nothing from now until Q3 07? Does 45nm or Yorkfield ring a bell?

You keep going on about how K8L will beat Conroe, but you fail to mention that it won't even be competing with Conroe (65nm) when it is launched but rather Yorkfield @ 45nm.

According to the latest reports, it'll be Yorkfield @ 3.4 - 3.7GHz vs K8L @ 2.7 - 2.9GHz, K8L better well be significantly faster than Conroe (and Yorkfield) clock for clock if it wants to compete.

Of course, I'll be expecting you to say you are sceptical of a Q3 Yorkfield release, that the slated clockspeeds are unrealistic, etc etc.

 

[zsdersw]

Originally posted by: Viditor
We have already seen an example of this when the ODMC was first introduced on the Athlons...
The decreased latency was estimated to increase performance by ~20%...
Ars-Technica

20% improvement in performance compared to what preceded it; no IMC. The difference between K8's IMC and K8L's IMC doesn't seem to be as significant as the IMC was for the Athlon.

 

[harpoon84]

Originally posted by: zsdersw

Originally posted by: Viditor
We have already seen an example of this when the ODMC was first introduced on the Athlons...
The decreased latency was estimated to increase performance by ~20%...
Ars-Technica

20% improvement in performance compared to what preceded it; no IMC. The difference between K8's IMC and K8L's IMC doesn't seem to be as significant as the IMC was for the Athlon.

I think what Viditor is trying to say is that K8L's core features are just as impressive as Conroe, and with the added benefits of an IMC will enable it to exceed Conroe levels of performance.

However, it seems K8L will be competing against Yorkfield, rather than Conroe, so that is a totally different ballgame.

Yorkfield will be slightly faster than Conroe clock for clock, and will clock higher due to the 45nm process. That will be a stiff challenge for a K8L chip @ 65nm.

 

[dexvx]

Originally posted by: SparkyJJO

Originally posted by: OhioState
everyone likes to discuss the merits of X2-VS-C2D but can we agree that semperon totally pwns cerleron >?

Agreed. Celeron is pretty bad. Only once CPU series I know of is worse - VIA processors 😛

No, Celeron-M's squarely and soundly beat Sempron's.