Jump to content
SceptreCore

AMD conFusion? Forget the hype!

Recommended Posts

Well just done some research and there was 1 year 1 month and 12 days between Phenom and Phenom II. So if we see the same turn around here then pencil in early December 2012 in the calendar. Not say ing it will be a repeat, but thats how long it took them to turn fail into success last time.

 

Of course to go from the terrible stuff up phemon B2 to the fixed up phenom B3 took 4 moths 8 days. So maybe AMD will give us some love for valentines day next year.

 

 

and in terms of what I'd like to see from a bd review. Grab k10stats (if it works on bd) drop voltages, pump the stock and turbo clocks as far as they will go on stock clocks and see where everything really falls on power consumption.

 

From my perspective if power consumption was more reasonable and the clocks were around 4.4ghz stock then it would have been a damn nice chip. So maybe they can do a phenom B2/B3 fix and have us something nice in 4 months. Or maybe the problems are deeper and call for a more phenom to phenom II fix and the time required to do that.

Thats a good point and I hope that is true. It can only get better!

Share this post


Link to post
Share on other sites

As per my post in another thread:

 

I was going to get a Bulldozer or two for review in a few days, but I 'm not sure I see the point now. On top of everyone and his dog having reviewed the thing... if performance was ambiguous enough I could investigate where it does excel, maybe get some insight into the architecture and it's future directions but honestly, and I know a lot of you disagree with me, I just think it's sufficiently unimpressive as compared to Phenom II as for there to be nothing to see.

 

However there may be some value for AMD fans or those curious - can you folks think of anything you'd like investigated such as simulated Opteron performance (think of it as a reacharound and penance for my rather harsh clarion calls about BD)?

 

Man if you can get one for a review I say do it.

Share this post


Link to post
Share on other sites

and in terms of what I'd like to see from a bd review. Grab k10stats (if it works on bd) drop voltages, pump the stock and turbo clocks as far as they will go on stock clocks and see where everything really falls on power consumption.

Prob K10STAT won't work, however doing a stock vcore O/C via BIOS should accomplish the same thing? An overall analysis of how power consumption scales with clock speed would be interesting, I'm guessing some severe leakage goes on at higher volts.

 

The point is that it isnt that good now. If threading does become the way of the future (and I think it is) then Bulldozer may be the beast that we all hoped for but can we all really see games being properly threaded in the near future?

And even if that is the case, fully-threaded performance isn't generally above the competition.

 

 

Imagine sandybridge cores in a bulldozer config - impossible I know but THAT would be some real performance.

Goodness yes, would be a thing to see and we could fit eight cores (and 16 HT threads!) into a slightly larger die and power envelope. Edited by philo-sofa

Share this post


Link to post
Share on other sites

On that opinion article - Yes I believe they are guilty of over hypeing their processor if they were actually aware of how poorly the cpu performed overall. It doesn't consistently beat intels cpus in multithreaded tasks - it does do very well but from the hype I would expect it to dance on the graves of the poor single thread optimised architecture.... unless something went wrong; maybe despite the initial results looking good, the process just wasn't refined enough and they really haven't completely fixed the B0 stepping problems? Or maybe they need to write drivers that retreat single threaded tasks to make them perform better on this cpu,

 

I don't think we were expecting too much I just think that AMD has fallen short of what it said its goal is. I would have called it a success if the numbers were all the same except that single threaded got (like they promised) +50% on phenom II, actually I would have settled for +30%, at least then we could just say yeah the power consumption is high but at least its doing something. Being beaten by a phenom II 980 just seems inexcusable and the only way I can see them justifying this is if there was a mistake and it just didn't work in this revision and with another stepping the problem would be solved but isn't that going to cut it very fine to the Q2 launch of piledriver (has this date changed?)

Edited by UberPenguin

Share this post


Link to post
Share on other sites

^^ agree. Given it's significantly slower in single threaded apps than a Thuban and between a 2500 and 2600 in multithreaded apps I'd say it's a failure for the consumer market (tho it might be completely different in servers). My criteria for success if the performance/die size and performance/watt were roughly in line with Sandy in multithreaded as opposed to way over.

 

However as you say, hopefully a combination of process improvements and reengineering can rescue the fundamental architecture. I too wonder if they pushed it out of the door too early - however then again they're only forecasting, ~15% more speed from Piledriver.

Share this post


Link to post
Share on other sites

If your current new top line 8 core CPU can't even match a 4 core hyperthreading 2600K, something went wrong I think.

Share this post


Link to post
Share on other sites

Maybe they learnt their lesson about over hyping, or are trying to pull a conroe - "only 15%" in actual benchmarks turns out to be 50%.... maybe I'm being naive.

 

@philo-sofa just out of curiosity why do you care so much about die size?

Edited by UberPenguin

Share this post


Link to post
Share on other sites

Maybe they learnt their lesson about over hyping, or are trying to pull a conroe - "only 15%" in actual benchmarks turns out to be 50%.... maybe I'm being naive.

 

@philo-sofa just out of curiosity why do you care so much about die size?

He wants to put a FX-8150 into a laptop? LOL

Share this post


Link to post
Share on other sites

just my 2c:

 

judging of of reviews only i think that bulldozer has a market to fill but i think if some of the results are right then wait till the next stepping or piledriver would be the option. I really want to try out AMD but i guess this isnt the case..

 

with all these boring new CPU's coming out or released i think im happy to stay on my 950.

my opinion on the current CPUs (and this is off what ive read or used):

 

AmD bulldozer: from review it looks decent but results show doesn't perform..looks like it could be a good overclocker though..and might be decent as a small reder system for max work.

 

Intel Sandy bridge: decent performance and overclock ability..after using a stock i7 2600 it doesnt really as fast as my current 950..as others have said when SB came out - if your one s1366 now its now worth upgrading. I would have to agree with this.

 

Sandy bridge -E: early results indicate that the extreme OC capability of this is only 5.5ghz so if air is anything to go my it looks like the OC headroom isnt going to be a big difference no matter what cooling is used. Also i have read reports/forum post that say it has a high power draw so its not a ccool chip...including the price of the chips and boards once released is going to be expensive it doesn't look worth it until a few steppings/revisions in or price is cheaper..

 

i think with games the way they are and many applications not really taking advantage of faster CPU's staying on any recent CPU setup (AM3,S1366,1156/1155) is a good idea at this current time.

Edited by jdog

Share this post


Link to post
Share on other sites

Maybe they learnt their lesson about over hyping, or are trying to pull a conroe - "only 15%" in actual benchmarks turns out to be 50%.... maybe I'm being naive.

 

@philo-sofa just out of curiosity why do you care so much about die size?

It’s a good question. In short because die size is a good measure of ‘how much CPU it takes’ to produce a given level of performance, as well as a reasonable-ish measure of how much it costs relatively speaking, to make a given chip. All-up Sandy is 216mm^2 iirc, and Bulldozer 315mm^2 – so all other things being equal Intel could produce one and a half 2600K’s for every FX-8150 (implying a commensurately higher cost to produce the FX-8150).

 

Bearing in mind that CPUs are produced on 300mm diameter wafers, I say ‘all other things being equal’, 'implying' and ‘arguably’ as this depends on how much each ‘wafer’ of silicon costs to produce - AMD’s SOI methodology may be cheaper per wafer (and thus cheaper per mm^2 of silicon). Yields (number of failed chips per 300mm wafer) also affect the calculation. The penultimate caveat is that it depends on how much ‘extra stuff’ is packaged with the CPU, for example a good 20% of Sandy is actually graphics, and it also packs an on-die PCIe controller. The final wrinkle to all this is that die-size, complexity, yield and the actual processor design all interrelate in a very complex way.

 

What is certain is that if you take the actual ‘CPU area’ of a Sandy Bridge processor, it’s about half the size of the ‘CPU area’ of a Bulldozer processor, meaning (subject to most of the complications above) that AMD appears to be using twice as much ‘chip’ to get the same performance as Intel (under ideal conditions), which is a good part of the reason Bulldozer consumes so much power. Subject to the remaining caveats (cost per wafer and yields), and given the reality that Intel has packaged a GPU and PCIe controller in with Sandy, it suggests that an FX-8150 could cost 50% more to make as compared with a 2600K – however I really have to learn more about how much each 300mm wafer of CPUs costs before I could say that with certainty; it’s possible AMD’s process is much cheaper than Intel’s and that ameliorates a lot of the extra cost.

 

The other possible equivalent metric to use is ‘transistor count’ (about 1bn for Sandy and 1.4bn for Bulldozer) however that’s ambiguous as transistors are sometimes ‘packaged’ together, and this you can’t always directly compare transistor counts between architectures and manufacturers. Thus I tend to concentrate on die-size as it’s a good measure of ‘how much CPU’ is giving the performance levels we see, and a broad proxy for the cost-per-CPU.

Edited by philo-sofa

Share this post


Link to post
Share on other sites

 

@philo-sofa just out of curiosity why do you care so much about die size?

It’s a good question. In short because die size is a good measure of ‘how much CPU it takes’ to produce a given level of performance, as well as a reasonable-ish measure of how much it costs relatively speaking, to make a given chip.

 

Ya, but why does that matter to you?

You are not paying for the production of the chip, so it's kind of meaningless to rate a CPU by it?

Share this post


Link to post
Share on other sites

 

@philo-sofa just out of curiosity why do you care so much about die size?

It’s a good question. In short because die size is a good measure of ‘how much CPU it takes’ to produce a given level of performance, as well as a reasonable-ish measure of how much it costs relatively speaking, to make a given chip.

 

Ya, but why does that matter to you?

You are not paying for the production of the chip, so it's kind of meaningless to rate a CPU by it?

 

It shows how effective a given architecture is, and that's interesting. It also shows how much profit the manufacturer may be making, which has direct implications for how the company will fare.

 

I agree it's cetainly not something for the consumer to rate a chip on when deciding to buy it however - all that matters there is consumer price and performance. Die size analyses are purely interesting to gain insights into the architectural efficiency of a CPU and its profitability.

Share this post


Link to post
Share on other sites

philo-sofa, you think it's the APU (formerly a GPU) on the Bulldozer bloating the die size on the Bulldozer?

Share this post


Link to post
Share on other sites

Bulldozer doesn't have an APU... If it does I really need to redo my 'CPU performance per mm^2' calculations lol.

Sorry, got confused with Fusion/LLiano ... BLEH

Share this post


Link to post
Share on other sites

Although it would explain both the massive transistor count and the high power consumption if a GPU that they never got working was still hiding in the die. I don't expect that is true, but it is something to think about. Have we seen a bulldozer die shot?

 

Remembering back to the start of all this AMD had big hopes for the APU. I imagine that the first bulldozer design probably was an apu. but time may have changed that.

 

Anyway we will see trinity with piledriver cores soon, that will give us a significant idea about what the next 8 core chip will be like.

Share this post


Link to post
Share on other sites

However there may be some value for AMD fans or those curious - can you folks think of anything you'd like investigated such as simulated Opteron performance (think of it as a reacharound and penance for my rather harsh clarion calls about BD)?

I'd be interested in a virtualised server setup with one of these, see how it performs running a few VM's

Dunno that any of the reviews have covered this area at all yet.

 

 

 

If your current new top line 8 core CPU can't even match a 4 core hyperthreading 2600K, something went wrong I think.

Yes but AMD have consciously taken a different approach to threading than Intel. Besides a single "core" in bulldozer is smaller and less complex compared to a single SB core.

 

Remember:

Posted Image

Intel has started bottom left (SMT), AMD has gone for top left (CMP). And you gotta say, as bad as the single threaded performance is, when working together it does rather well.

Remember how Hyper threading was when it first was introduced? Almost useless. CMP may have a ways further yet before it's perfected.

And every 10% AMD add in performance to each core that'll be multiplied 8 fold over the other 8 cores in the multi threaded applications, so potentially might scale better.

 

It shows how effective a given architecture is, and that's interesting. It also shows how much profit the manufacturer may be making, which has direct implications for how the company will fare.

 

.... Die size analyses are purely interesting to gain insights into the architectural efficiency of a CPU and its profitability.

+1

 

IPC per mm2 or something is interesting to look at. Efficiency wise SMT helps when it comes to die size, but perhaps doesn't scale as well under intense multi threaded loads as the hyper threading relies on the pipeline not being full of work to do to shove a 2nd thread into it.

Theoretically AMD may have a better long term solution. Practically I'm not sure how chock a block AMDs piplines get with work. Intels cores would be kept well fed and thus better utilised having the 2nd thread in there.

 

 

Although it would explain both the massive transistor count and the high power consumption if a GPU that they never got working was still hiding in the die. I don't expect that is true, but it is something to think about. Have we seen a bulldozer die shot?

There's no GPU/IGP in bulldozer. Half of it is just cache.

Posted Image

Share this post


Link to post
Share on other sites

I really do think this design was aimed more toward the server market and probably will have advantages there with some updates.

Share this post


Link to post
Share on other sites

I guess Bulldozer was really a server chip and AMD thought it could make an awesome desktop one but that didn't turn out.

Share this post


Link to post
Share on other sites

Damn, always been an AMD fan since my old K2 chip.

 

This however is a pathetic effort accompanied by a stupid price tag.

 

I'm getting rather bored with AMD's offerings,

 

I jumped ship to Nvidia a while back maybe's it's time to give up on AMD entirely.

Share this post


Link to post
Share on other sites

So wait... I dont understand why no-one has posted benches at OC levels versus the i5 2500K OC'ed which is seems to be most suitably compared with...

 

or have I missed an article somewhere?

 

I vaguely remember Tom's saying they only got BD around 4.7Ghz stable but I sure it could match the i5's norm of ~5Ghz on a mobo with good power yeah?

 

I think gamers, value-minded techies and just mid-range enthusiasts would want to see the most would be

a/ How well it performs OC versus other similarly priced units OC versus more premium units stock

b/ If it is actually possible to "unlock" the extra cores on the FX-4, FX-6 (creating value also)

(AMD also said that the 6950's were impossible to unlock even the BIOS locked GPU's, and same for the Phenom II cores)

Edited by hicsy

Share this post


Link to post
Share on other sites

Actually it seems I missed one of the (now 100+) pages somewhere before this thread got turned into a fanboi flamewar by an AMD distributor...

but HardOCP did touch on some overclocking (if you could call it that) we just need some more comprehensive benches by perhaps Tom's (using decent RAM might I add - CMON guys... how cheap is the sh!t nowadays!)

Share this post


Link to post
Share on other sites

I guess Bulldozer was really a server chip and AMD thought it could make an awesome desktop one but that didn't turn out.

Majority of AMD's CPUs were server first then desktop. The Stars core was done in Barcelona and Magny-Cours before they moved onto the Bulldozer architecture with Interlagos and Valencia.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×