AMD Zambezi news, info, fans !

[Baam]

Well,

The latency results for simple sequental access show that prefetching is disabled in this ES....

In Sandy Bridge preftching reduces the L3 latency by a factor 2 or so. Latency reduction
is responsible for most of Sandy Bridge's IPC increase over Nehalem.

Regards, Hans

So pretty much what JF-AMD was saying. A lot of the ES chips have features disabled on purpose.

[flyck]

These quotes from JF's FAQ seems to confirm my anticipations, or lack of there of:



The way he try to mock people who care about IPC and single thread performance makes me believe he is trying to justify bad performance in these areas. We know frequency range, so IPC is important! And single thread performance is important to everyone who isn't building a server. It doesn't matter if you have a million cores, if single thread performance matches a K6-2 you won't be able to play any games today, just using windows would suck

IPC will be low, maybe lower than Phenom II, and that goes for single thread performance as well.

What logic are you using?

Since we know the frequency range you should also know the fastest BD has a rather big frequency advantage compared to intel i7 series.

1 thread -> +400MHz (10%)
2 thread -> +500MHz (13.5%)
3 threads -> +600MHz (16,6%)
4 threads -> +700MHz (20%)

From that point the differences get smaller between them.

ipc was said to be higher then K10 multiple times, so it should be somewhere between C2D and infinity.
if ipc is lower then FX4 and FX6 have no reason to exist. since the rumoured FX4 and FX6 models have no clockspeed advantage compared to deneb/thuban.

Also it is impopssible to deduct ipc/application/clock for BD from current known designs.

[maltrabob]

I can understand why AMD might not want to announce to the world an exact date like Sept 19th.

They want people buying as many of their other CPUs as possible. I've seen many folks buying "tide me over" CPUs in the past. AMD may want that to continue as long as possible. If people know an exact concrete launch day, I'd imagine AMD's other CPU sales may drop off rather well as said date approaches.

That`s my point of view too. If we forget the speculations about weak performance for a while it seems to be a good reason for AMD not to announce Bulldozer`s release date. I am very tempted to buy a Phenom II X4 955 BE for as low as 90 quid myself, especially when some online stores bundle it with a new Deus Ex game. In my oppinion that indicates they want to get rid of them in anticipation of a new AMD`s arrival. I will wait and see from trusted reviews if it was worth waiting for, but I understand that some weak individuals would already opt for 955 BE ;-).

[Evantaur]

Well i bought 1100t to warm my socket for bulldozer.

[LowRun]

So, we are one week past AMD's worst case scenario for BD's availability but they don't feel like communicating about the delay

I suppose AMD must be removed from the reliable sources list for AMD's products launch dates

[Michaelius]

What logic are you using?

Since we know the frequency range you should also know the fastest BD has a rather big frequency advantage compared to intel i7 series.

1 thread -> +400MHz (10%)
2 thread -> +500MHz (13.5%)
3 threads -> +600MHz (16,6%)
4 threads -> +700MHz (20%)

From that point the differences get smaller between them.

ipc was said to be higher then K10 multiple times, so it should be somewhere between C2D and infinity.
if ipc is lower then FX4 and FX6 have no reason to exist. since the rumoured FX4 and FX6 models have no clockspeed advantage compared to deneb/thuban.

Also it is impopssible to deduct ipc/application/clock for BD from current known designs.

1. Who cares about stock frequency ? Phenom II used higher frequency card against i5 750 and while they managed to equal it at stock it still was beaten badly when overclocked.
2. FX4 and FX6 have very simple reason for existance- saving partially faulty silicon from garbage bin and turning it into some cash (absolutly obvious when we look at ugly TDP those chips have)

[flyck]

1. Who cares about stock frequency ? Phenom II used higher frequency card against i5 750 and while they managed to equal it at stock it still was beaten badly when overclocked.
2. FX4 and FX6 have very simple reason for existance- saving partially faulty silicon from garbage bin and turning it into some cash (absolutly obvious when we look at ugly TDP those chips have)

I believe stock is what matters... when you talk oc you have absolutely no reference for BD. its more like, who cares about overclocking? overclocking, or the possiblity of it does not break or make a product.

you mean those ugly TDP ratings equal to those of SB?

[TESKATLIPOKA]

Michaelius

2. FX4 and FX6 have very simple reason for existance- saving partially faulty silicon from garbage bin and turning it into some cash (absolutly obvious when we look at ugly TDP those chips have)

how can you sell something worse than your older product and cost probably more, everyone would buy Deneb instead.
are you saying TDP95W is ugly, SB has the same TDP so It must be ugly too, right.



Shouldn't they release a native 2module too, I don't think they would have enough chips just from faulty 4module if they need to sell 3modules too.

[Michaelius]

I believe stock is what matters... when you talk oc you have absolutely no reference for BD. its more like, who cares about overclocking? overclocking, or the possiblity of it does not break or make a product.

you mean those ugly TDP ratings equal to those of SB?

I'd assume something like huge majority of people who read this site or [H] or other forums

Last non overclocked CPU i runned was PIII 500

[Michaelius]

Michaelius


how can you sell something which is worse as your older product and cost probably more, everyone would buy Deneb instead.
are you saying TDP95W is ugly, SB has the same TDP so It must be ugly too, right.

Shouldn't they release a native 2module too, I don't think they would have enough chips just from faulty 4module if they need to sell 3modules too.

I never said it will be worse than older cpus (that would be amazing engineering achievement to make something slower than phenom II architecture in 2011). Even if only IPC increase comes from bigger cache it should be faster than Phenom II.

And seriously do you even understand basic concept of performance per watt ?
95W TDP on SB is great because it has performance to match it.
Yet FX4 comes with 95W TDP compared to 125W for top-end FX8 while offering half it's performance.

PS. Radeon 5770 says hello. How do you sell slower card at higher price point ? Simply you just stop producing 4870 and 4890 so there's no competition from your older better product.

[Mats]

TDP is a limit. You should read it ≤95 W.

[TESKATLIPOKA]

Michaelius

And seriously do you even understand basic concept of performance per watt ?
95W TDP on SB is great because it has performance to match it.
Yet FX4 comes with 95W TDP compared to 125W for top-end FX8 while offering half it's performance.

Yeah I understand It, but this is the first time you mentioned it, I can't find anything about it in your previous comment
And wouldn't you say BD FX 8120 is way more awesome than SB because It has the same TDP yet it should be faster in multithread.

PS. Radeon 5770 says hello. How do you sell slower card at higher price point ? Simply you just stop producing 4870 and 4890 so there's no competition from your older better product.

i didn't hear anything about stopping the production of deneb

I never said it will be worse than older cpus (that would be amazing engineering achievement to make something slower than phenom II architecture in 2011). Even if only IPC increase comes from bigger cache it should be faster than Phenom II.

yeah, I have the same opinion yet you can still find people who think the IPC will be lower than the previous generation.

[Mats]

And wouldn't you say BD FX 8120 is way more awesome than SB because It has the same TDP yet it should be faster in multithread.

Who cares, compare power consumption instead.

[Apokalipse]

And seriously do you even understand basic concept of performance per watt ?
95W TDP on SB is great because it has performance to match it.
Yet FX4 comes with 95W TDP compared to 125W for top-end FX8 while offering half it's performance.

FX4 = 4 cores, SB = 4 cores
what's the problem here?

Also, AMD's TDP != Intel's TDP

PS. Radeon 5770 says hello. How do you sell slower card at higher price point ? Simply you just stop producing 4870 and 4890 so there's no competition from your older better product.

5770 has more/better features (does Direct3D 11, has lower power consumption, eyefinity, better crossfire scaling), the same performance, and overclocks better.
And it's not supposed to be faster than the 4870. The 5870 is.

I would go with a 5770 over a 4870 and overclock it.

[flyck]

I never said it will be worse than older cpus (that would be amazing engineering achievement to make something slower than phenom II architecture in 2011). Even if only IPC increase comes from bigger cache it should be faster than Phenom II.

And seriously do you even understand basic concept of performance per watt ?
95W TDP on SB is great because it has performance to match it.
Yet FX4 comes with 95W TDP compared to 125W for top-end FX8 while offering half it's performance.

PS. Radeon 5770 says hello. How do you sell slower card at higher price point ? Simply you just stop producing 4870 and 4890 so there's no competition from your older better product.

FX4 would only offer half the perfornance in case of massive parralelism. on average it would be around 25-30% slower. (maybe less). Or in the facinity of i5 2300-2400. all that within the same TDP. Given FX6 can be made with same clocks in same TDP, FX4 could very well consume considerably less than its TDP also (just like SB does).
So i see absolutely no basis for low performance/W ratio for BD.... except ofcourse the major issues with the 32nm process...

[Solus Corvus]

Michaelius is trolling, don't fall for it.

A) The vast majority of the market cares about stock performance.
B) OC performance isn't known, but if he-who-shall-not-be-named is to be believed, it can OC well. And he's pathetic at overclocking, for what it's worth.
C) TDP != power consumption.

It's best to ignore the bait.

[freeloader]

So, we are one week past AMD's worst case scenario for BD's availability but they don't feel like communicating about the delay

I suppose AMD must be removed from the reliable sources list for AMD's products launch dates

Hehehe....I like that. Can I sig it?

[LowRun]

Hehehe....I like that. Can I sig it?

Unfortunately, yes you can.

[Michaelius]

Michaelius is trolling, don't fall for it.

A) The vast majority of the market cares about stock performance.
B) OC performance isn't known, but if he-who-shall-not-be-named is to be believed, it can OC well. And he's pathetic at overclocking, for what it's worth.
C) TDP != power consumption.

It's best to ignore the bait.

A. It says "Extreme Systems" in the top of my browser not "Average joe rigs". if I'm building rig for my uncle I'm giving him parts that will perform at stock
B. It better be considering 2600K can score 1Ghz+ overclocks easily using AC for 24/7 operation
C. Yes and your point is? It's still an indicator of real world usage. If they could stick lower rating on it then they would. Also SB TDP is given for situations when GPU is used alongside CPU part

And lol i guess i must be hitting some nerve with my arguments when the best you can do is try to call me a troll.

@ Apocalypse - yes typically Intel parts with same tdp have lower power usage in real world benchmarking even if official definition would suggest otherwise
@ Teskatlipoka - there were some rumors that AMD will phase out phenoms II very fast http://www.guru3d.com/news/amd-to-ph...henom-ii-fast/

[Solus Corvus]

And lol i guess i must be hitting some nerve with my arguments when the best you can do is try to call me a troll.

Nope, it's not all I could do. I could have responded in depth. But it's not worth the time when it is fairly clear that your only purpose is to "hit a nerve".

A) Irrelevant to the market success of bulldozer.
B) Unknown, until we get our hands on it.
C) Wrong, TDP is not an indicator of real world usage.

And that's about all the effort I'm going to put into it when your intent is so clear.

[TESKATLIPOKA]

Mats You are right but I can't compare a product versus another when its still not released, only TDP is known.

[Particle]

A. It says "Extreme Systems" in the top of my browser not "Average joe rigs".

You're missing the point though. Sure, overclocked performance matters to most (but not all) of us here, but we as an overclocking community might account for what...10, 1K trays of CPUs? Like it or not, we're a barely significant niche which you're overstating the importance of with your sweeping statement that assumes what matters to us matters to a significant degree overall. It doesn't. I think that's what he is trying to tell you.

[CrazyNutz]

PS. Radeon 5770 says hello. How do you sell slower card at higher price point ? Simply you just stop producing 4870 and 4890 so there's no competition from your older better product.

5770 replaced 4770 (5770 is faster)

5870 replaced 4870/90 (5870 is faster)

Your comparing a mid-range to a high-end part. Also if the 4870/90 was ever cheaper than a 5770 (I don't think it was) it would have only been to deplete inventory.

[jimbo75]

@ Apocalypse - yes typically Intel parts with same tdp have lower power usage in real world benchmarking even if official definition would suggest otherwise

"13W" D525 http://ark.intel.com/products/49490

"18W" E-350 http://products.amd.com/%28S%28h2ine...ookieSupport=1



http://techreport.com/articles.x/20401/5

[AKM]

So, we are one week past AMD's worst case scenario for BD's availability but they don't feel like communicating about the delay

I suppose AMD must be removed from the reliable sources list for AMD's products launch dates

I'm not making any illusion that I'll find any BD chips available sooner that Q1 2012.