I'm starting to wonder about the marketing going on here. This was a big no-no in the GPU section without a disclaimer in the signature.
- Status
It does. In order to boost that much you need a decent amount of headroom, which you aren't going to see if your chip design needlessly uses power.
The difference is that one has a stock clock rate 400 MHz higher than the other, but has a maximum turbo boost that is 100 MHz lower. Assuming they are the same architecture and that these numbers have been arbitrarily put in place, there likely won't be much overall performance difference.
If someone has a 2600K, they could easily create such a simulation by adjusting the clock rate and turbo settings.
Depending on the chip, certain clock speeds will result in a more efficient chip depending on the amount of voltage needed to reach that speed and the delta required to gain additional clock speed. It is entirely possible that one chip performs better than the other simply because it is more efficient at a given clock rate.
Tom's Hardware did an over-clocking test on the Sandy Bridge parts and found that the processor was actually more efficient at some settings. Most chips generally can be more efficient with a slight to moderate over-clocking.
It's likely that AMD has determined the levels where their new architecture operates most efficiently and has used those as the stock setting and then determined that if they're going to turbo boost, that increasing the clock by 500 MHz is more efficient than either 400 or 600 MHz.
It's entirely within reason that the settings AMD has decided to use represent the most efficient chip within a given TDP. In your example, all of your chips will top out at 3.5 GHz, but some may be more efficient over a long period of time.
Good question. There title is prefaced by "Rumor" and JFAMD has spent more time telling everyone that none of this is official and is most likely fabricated rather than leading us on about how great things will be.
I'm not sure where this 'marketing' is because I'm not seeing it.
I'm not sure where this 'marketing' is because I'm not seeing it.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
Define headroom. Does a 65w cpu have more or less headroom than a 95w cpu?
Again, if a take a cpu and rename it, does it develop more headroom somehow?
That's just it - "stock clock rate". CPU's now days really don't have one, they run at different speeds depending upon load.
That hypothetical 2.5Ghz cpu would most likely run at the exact same clock speeds under normal conditions as that 3.4Ghz cpu would (idle clocks may be a bit lower).
The only thing you can say about a cpu that boosts or turbos 500Mhz is exactly that. You can't infer anything else. Don't fall prey to a numbers game.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
A good marketer, just like any good salesman, will appeal to logic as well as emotion. Sure, SOME people will buy based upon emotion alone, but many many more will buy based upon logic backed up with strong emotion.
Regardless, that's beside the point. John is a server dude, he has to appeal to logic much more than emotion in his job. Remember, he's not trying to sell an Opteron 887912 to little johnny who's using his lunch money to build a budget rig with spare parts, duct tape, and rubber bands; he's selling to google/M$/IBM/lawrence livermore/etc etc etc.
cpus are very different from gpus, however. neither intel nor amd has a "focus group" or other such nonsense, nor have either of them had rollo rumbling around getting lifers banned and generally stirring up trouble. If you're talking about jfamd, he's got amd in his name, plus he's got a link to his blog in his sig. AND he's not a mod like keys, which imho makes a big difference as well.
Last edited:
In terms of what, percentage of total or watts? If the first CPU dissipates 60W under normal conditions and the second 90W, both have 5W of headroom, but the first processor has a larger percentage of headroom remaining.
If you took a CPU designed to have a 65W TDP and sold it as a 95W TDP CPU, it would have more headroom by definition, but it's not designed for those conditions so in reality it might not be possible to reach 95W TDP without massive amounts of over-clocking. I really don't understand why you continue to talk about renaming a CPU. If it's designed to meet a certain level of performance, that's what it's going to be branded as. No one is going to undersell their performance and overselling it would be fraud.
I can infer that if AMD wanted to get the best performance out of the chip, they may have chosen a value that keeps the chip within its rated TDP and maximizes the efficiency of the chip.
Let's look at your example with the 2.5 GHz and 3.4 GHz CPU and assume that neither chip will boost in order to complete the work load that it's been given. Obviously the 3.4 GHz chip will complete the work more quickly since it is faster, but if it uses a disproportionately larger amount of power to complete that work, it's less efficient than the 2.5 GHz chip.
Similarly, I can speculate that boosting that chip by 500 MHz makes it more efficient than boosting it only 400 MHz or 600 MHz. That extra 100 MHz may require additional voltage that makes it less efficient and the chip may be able to get an extra 100 MHz for only slightly more voltage than the 400 MHz boost.
My guess is that the base clock speed has the highest efficiency for the rated TDP, but leaves plenty of headroom. The 500 MHz turbo gets the chip closest to its TDP while still maximizing efficiency. This is most likely done from the perspective of providing the best performance per watt and greatest chip longevity. The could easily just set the base clock 500 MHz higher, but it may not be as efficient and would probably reduce the life of the chip.
Not only that, but the "requirement" doesn't actually exist.
Keysplayr denotes that he is a member of the Nvidia Focus Group in his sig here in the AT Forums because that is a requirement of the Nvidia Focus Group, it is not a requirement of the AnandTech forums.
We have no such requirement for the simple reason that such a requirement would not be enforceable.
We do not require people to indicate their positions of employment, affiliations, associations, club memberships, etc in real life.
Could you imagine just how much more rancor and friction we would all have to step through if every member had to list their employer, their religion, and their registered political party in their sig?
People who want to divulge that info are free to do so. But we also can't verify or attest to the legitimacy of any of these claims either. Keysplayr claims he is a member of Nvidia Focus Group but we have not verified this to be true.
The poster here who presents himself here as "JFAMD" and claims to be the John Fruehe of AMD could actually be an Intel employee for all we know, as members and as moderators, as we simply do not require anyone to provide their bonafides unless they seek us out for a special title or some such. (and even then we don't really verify by way of background check or employee references, its more of an honor system)
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
This is where you are in error.
Under most normal workloads both chips will operate at the exact same clock speed. You are assuming that the base clock speed is the minimum clock speed. That is incorrect. For example, as I type the 2.53Ghz I5 in this laptop is running at 1.2Ghz.
You are confusing TDP with power consumption. A "2.5Ghz" cpu running at speed X will consume the same amount of power as a "3.5Ghz" cpu running at speed X. The label we put on the box is meaningless.
The I5 in this laptop is a 2.5Ghz cpu. I don't know what it's turbo speed is - for arguments sake let's say it's 3.0Ghz.
As I type this my cpu is running at 1.2Ghz. I can legitimately name this a 1.2Ghz cpu and say it has 1.8Ghz of boost! This cpu has the exact same operating characteristics of that 2.5Ghz I5 - same TDP, same clock speeds, same price. But it is not more efficient because it has more headroom over it's "rated" speed than the 2.5Ghz cpu.
The only place rated or base clock speeds come into play is if all power saving and turbo features are disabled.
Last edited:
btw, for anybody who is wondering, I'm a gemini.
Oops, wrong forum...
Gemini Information said:
It's all starting to make sense now! :hmm:
brybir
Senior member
- Jun 18, 2009
- 241
- 0
- 0
That's why I said over a long period of time. Typing in text to a web page that's already been rendered is not computationally intensive. However, when you submit the post and need to render the new page, the CPU utilization probably jumps up to 2.53 GHz for a brief moment before returning to rest.
No, TDP is the maximum amount of waste heat the processor will need to have dissipated and can be measured in watts. It's related to the amount of power consumption, but can vary. For example if you have a process that leaks more voltage than another processor, it will produce additional waste heat. If a processor is idle, it's drawing less voltage and producing less heat, therefor having a larger amount of headroom. In terms of TDP it doesn't matter how much or how little voltage it requires to produce that amount of waste heat, although in practical terms you'd like to have an efficient processor.
If you have two processors rated for 65W and 95W TDPs, and under regular usage they dissipate 60W and 90W of waste heat, both are capable of dissipating another 5W of waste heat before they hit their TDP limit. The first processor has a larger percentage headroom, but that's just another way of looking at it.
If both processors are the same architecture and within the same level of quality, they'll both use approximately the same amount of power. If one chip were rated for 3 GHz instead of 3.5 GHz, it's likely because the quality is lower and it's leaking more voltage which means it needs to be clocked lower to stay within the rated TDP. In this example both chips were designed to be 3.5 GHz parts ideally, but due to defects in the manufacturing process not all of them are suitable to be 3.5 GHz parts.
In that case, running the 3 GHz chip at 3.5 GHz is going to produce additional waste heat and require additional voltage. If the chip is fully capable of running at 3.5 GHz, no one is going to market it as a 2.5 GHz chip. They won't be able to sell it for as much. Any chips that are being marketed as 2.5 GHz are defective 3.5 GHz chips or a different chip designed to fall into a different TDP range.
No one would name it a 1.2 GHz chip as that's just the idle speed and it's most likely not very efficient when it operates at that speed which is why it normally operates at 2.5 GHz or higher. The only reason it reduces its clock to 1.2 GHz is because it's not actually doing anything and having it run at 2.5 GHz would just waste power. Once it actually has to do something it's not going to do it at 1.2 GHz as that wouldn't be efficient.
Well fights breaking out over favorite color Power Ranger, Teenage Mutant Ninja Turtle, etc. might be somewhat more entertaining than fights over favorite GPU manufacturer. At least for the first few weeks.
VirtualLarry
No Lifer
- Aug 25, 2001
- 56,587
- 10,225
- 126
You sure about that? I remember differently.
[/quote]
What about people that represent that they are from companies, and they get a special member title, indicating that they are from a company.
Last I heard, that requires some sort of official company confirmation to get that title.
VirtualLarry
No Lifer
- Aug 25, 2001
- 56,587
- 10,225
- 126
I disagree. Intel does this all the time. They don't strictly bin as per performance metrics, if they did, they wouldn't have many lower-end chips to sell.
http://www.theregister.co.uk/2011/02/24/amd_bulldozer_core_isscc/page2.html
As an example.
I assume that on a Linpack type load, a Bulldozer will run at 3.5 GHz.
On an Integer type load, it will run at 3.5 + GHz.
All times upclocking to stay within the power limit.
What exactly is the problem?
As an example.
I assume that on a Linpack type load, a Bulldozer will run at 3.5 GHz.
On an Integer type load, it will run at 3.5 + GHz.
All times upclocking to stay within the power limit.
What exactly is the problem?
They probably didn't like to do it and if you look at the changes they've made to SB to all but prevent over-clocking on anything but the K parts it makes sense. I imagine that they're not going to bin a perfectly capable part. If they have several thousand they may just bin the worst X percentage of parts if they want to have a certain quantity of parts.
If one were to get multiple CPUs for testing purposes, I'm sure that many of the binned parts would still be able to over-clock well, but that they would likely require additional power in order to reach those same levels. There're also other reasons that the chips could have been binned such as defects in the cache or other feature areas such as hyper threading execution.
Generally though they'd rather not sell you a $100 CPU that can be over-clocked to bring it close to the performance of their $300 CPU. If the process is mature and reliable enough that binning can't be based on performance, they'll just put arbitrary restrictions on the chips.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
Just about everything you posted above is wrong.
Normally operates at 2.5Ghz or higher? Okay, define normal.
Also please show evidence that a cpu operating at 1.2Ghz isn't efficient.
Anyway, you still haven't said why this chip, which I can claim has 1.8Ghz of boost, is any more or less efficient that a chip with 500Mhz or 100Mhz of boost, based solely upon the amount of boost.
whats the scope of the clocking domain? fine grained domains mean efficenty can go way up based on the needed resouces of the workload.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
Yes I am quite sure.
What you are remembering is in relation to the banned member Rollo who was banned as Rollo and then allowed to come back as nRollo with one of the terms of conditions being that he disclose his focus group association in his sig.
This stipulation was unique to him, we've never made such stipulation for other members.
Yes we require them to provide a minimum level of credentials, which we verify, before they get their custom titles.
I'll see if I can find someone who's done analysis on over-clocking where they've massively under-clocked the chip. There's a certain amount of overhead you're going to see no matter what speed the CPU is running at. Someone posted a Register article that detailed the power draw of the new Bulldozer chips. When it's idle is still uses 20% of it's power budget to deal with leakage.
Edit: Found some efficiency results for an over clocked i7 920.
When you're running at an incredibly low clock rate the chip is much less efficient than running at it's stock rate. Ideally, that's an area of high efficiency so the amount of power it takes to complete the work is lower than if it were running at the bare minimum amount.
If we assume the above graph is an accurate portrayal of a Bulldozer chip and that it has a base clock of 3.5 GHz and an idle clock rate of ~1.2 GHz, it can accomplish roughly three times as much work while only using ~70% more power.
Here's a Tom's Hardware efficiency curve for SB over-clocking:
Notice that it varies. The graph is slightly exaggerated due to the way it's displayed, but there's a ~6% difference in efficiency depending on the clock rate used. If they were to extend it, running at the lowest clock rate possible would be highly inefficient and it's likely that continuing past a certain point will continue to see decreases in efficiency.
If you want your CPU to maximize efficiency it's best to target peak areas and avoid the valleys. Efficiency scales close to linearly, but not exactly. For server parts efficiency is important so it makes sense that AMD would try to select configurations where their chips perform in the most efficient manner possible. I don't have access to Bulldozer chips or any performance data, but this is simply an educated guess on my part.
Why you would want to market the chip in such a way is beyond me considering that it will generally avoid operating at idle unless it has almost nothing to do.
Edit: Found some results for an over clocked i7 920.
At best, there's slightly more than a 12% increase in efficiency.
Last edited:
well,
different workloads have differnet usage patterns, the more granular your clock domains the more head room you can create by down clocking those parts of the core that are under utilised and the parts of the core that are bottlenecking the workload can be clocked higher.
in the hotchip presentation the speaker said they had done a lot of work to decouble different stages, so if the clock speed is also variable between the indervidual stages then you can have your base clock set for the workload that is the heaviest/most power intensive workload(within a given TDP) and scale up from there (keeping within TDP).
the processor that has less ability to scale clock and clocking domains wont be able to scale as high on workloads that are bottlenecking somewhere within the core while being under TDP.
- Status
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter