It depends on the layout of the GPU. If you have 4 shader units tied to each pixel pipe then one cluster would be one core. If each pixel pipe is tied to a single shader unit, that would be a core.
I guess the simplest way of looking at any GPU, if they were using a die for multiple different configurations by disabling sections of the GPU based on yield, what options were available to them to utilize. When you figure that out, you can determine how many cores a given part has.
I guess the simplest way of looking at any GPU, if they were using a die for multiple different configurations by disabling sections of the GPU based on yield, what options were available to them to utilize. When you figure that out, you can determine how many cores a given part has.
So (like I said the previous time and now it is clear for anyone) you original intention asking all those questions was not to learn but to argue and you wouldn't even admit it.
Imagination Technologies categorize their GPUs exactly like I implied.
They categorize one SGX543 GP-GPU as one core that has 4 pipelines.
Like Imagination Technologies care about your objections!
What has this to do with how Nvidia categorize their own GPUs?
They have the same architecture?
No, they are different!
That's exactly what I implied with:
ATI's HD4XXX or NV's GXXX series are heterogeneous multi-core processing systems, I already told you that:
but what you can't understand is that the architecture in a model like the SGXMP16 is different.
apoppin
Lifer
Originally posted by: MODEL3
Originally posted by: apoppin
Originally posted by: MODEL3
Originally posted by: apoppin
What IS this thread about?
(It became) not exactly what I had in mind in the first place!
Originally posted by: apoppin
No, they did not
I agree. JPR did. :laugh:
No .. they did not
.. they mean what they say:
now i think they are talking new PCs; not half the PC in the world will be upgraded to multiGPU
Yes they are talking for new PCs and discrete GPU business.
Maybe the future will prove you right my friend, but like phexac pointed out it is highly unlikely that 50% of the new users will need a hardware solution like ATI CrossFire or Nvidia SLI technics.
Unless ATI & NV use in the future the label CrossFire & SLI as a software label for multi GPUs solutions that look like SGX543MP, then I really don't get JPR report.
i disagree
they have been quite accurate in the past; they made a similar prediction about multicore CPUs - which is true for '09, if i remember right
Originally posted by: apoppin
i disagree
they have been quite accurate in the past; they made a similar prediction about multicore CPUs - which is true for '09, if i remember right
Well like I said in my previous reply:
Maybe the future will prove you right my friend...
One clarification I want to make is that when I said:
I was joking of cource (that's why i used the laughing symbol)
Essentially I meant that my prediction is, that ATI or NV (or both) will use multi core (in the sense of "SGX543MP multi core") GPUs solutions in the future and they will use their well known brands "CrossFire" & "SLI" as a label for the Software side for these solutions, and I guess it was too early for JPR to reveal the future plans of ATI/NV.
apoppin
Lifer
i don't think JPR knows Nvidia/ATi's future planOriginally posted by: MODEL3
Originally posted by: apoppin
i disagree
they have been quite accurate in the past; they made a similar prediction about multicore CPUs - which is true for '09, if i remember right
Well like I said in my previous reply:
Maybe the future will prove you right my friend...
One clarification I want to make is that when I said:
I was joking of cource (that's why i used the laughing symbol)
Essentially I meant that my prediction is, that ATI or NV (or both) will use multi core (in the sense of "SGX543MP multi core") GPUs solutions in the future and they will use their well known brands "CrossFire" & "SLI" as a label for the Software side for these solutions, and I guess it was too early for JPR to reveal the future plans of ATI/NV.
- i believe they think that multi-GPU is "the future"
AMD is staking their graphics future on CrossFire-X, AFaiK
We are talking near-future = 2012
- if the Mayans are wrong -we will find out pretty shortly
:Q
Originally posted by: apoppin
i don't think JPR knows Nvidia/ATi's future plan
- i believe they think that multi-GPU is "the future"
AMD is staking their graphics future on CrossFire-X, AFaiK
Probably, but he collect informations from various sources as well from inside ATI & NV and maybe ATI & NV didn't clarify to him what these future CrossFire & SLI solutions will be like in order not to reveal their future plans.
Originally posted by: apoppin
We are talking near-future = 2012
- if the Mayans are wrong -we will find out pretty shortly
:Q
:laugh:
Up to that point I was not sure what you meant by a GPU core. Once I did I could start to argue that your OP statement is wrong.
Let me ask you differently. The current PC GPUs; are they multi-core or not?
No you were not arguing in the later posts you made only.
You were arguing since your second reply and you were trying not to show it that is why I told you the following back then:
I don't want you to ask me differently.
I disproved your claims:
with what I wrote in my previous reply:
So I don't know what else to say. I think I was clear enough in my answer.
And please no more questions!
You are ruining my topic!
If you have a question make a topic about it!
With this title: The current PC GPUs; are they multi-core or not?
If you have a specific argument about why I didn't disprove your above claims please state it, and I will reply you.
My specific argument: It's generally accepted that modern GPUs have been essentially multi-core for many years and therefore more than a half are multi-core already.
When JPR said multi-GPU they meant multi-GPU and not multi-core as you are trying to imply.
taltamir
Lifer
- Mar 21, 2004
- 13,576
- 6
- 76
graph makers need to understand that trends cannot continue forever. There is a thing called market saturation, not to mention individual circumstances.
The assumption that 1/2 will have multi GPU is based on the trend of increase continuing at the same rate. This ignores a few facts:
1. A recent major jump was due to major improvements in implementation and cost, especially by AMD. It is unreasonable to expect identical technology improvements. (it might improve more, it might improve less)
2. Eventually the market will be saturated and it will level off, it is BOUND to happen. In every product out there people go assuming an infinite continuation of the same growth rate, and they are wrong every single time
I would go so far as to say that in regards to #1, that the only reason we have been seeing growth in MGPU vs SGPU is because of technological improvements, as it becomes cheaper and more effective, more people are willing to buy it. If it remained the same as today for a couple of years, it will not grow any, at all.
The assumption that 1/2 will have multi GPU is based on the trend of increase continuing at the same rate. This ignores a few facts:
1. A recent major jump was due to major improvements in implementation and cost, especially by AMD. It is unreasonable to expect identical technology improvements. (it might improve more, it might improve less)
2. Eventually the market will be saturated and it will level off, it is BOUND to happen. In every product out there people go assuming an infinite continuation of the same growth rate, and they are wrong every single time
I would go so far as to say that in regards to #1, that the only reason we have been seeing growth in MGPU vs SGPU is because of technological improvements, as it becomes cheaper and more effective, more people are willing to buy it. If it remained the same as today for a couple of years, it will not grow any, at all.
This is not a specific argument about why I didn't disprove your above claims at all!
What you are saying is what I told you 100 times aleady (that modern GPUs are multi core, thanks for the news) :
I will say it again, What you can't understand is that although all solutions are multi core, there is a difference with the way NV & ATI offerings are multi core with the way SGXMP16 is.
About JPR he didn't use words like you say as multi-GPU or multi core in his chart.
The term he used is AIB. (add-in board)
AIB is a standard term and in the GPU space means that a graphics card is added to the motherboard.
Essentially in the GPU space this term is used to categorize the discrete GPU market.
He used the term multi AIB systems in his chart, which means systems that have many add-in boards.
What I was trying to say all along from the start is:
1.JPR chart is not indicative about what he meant. (I explain below)
2.XbitLabs misunderstood the JPR report (because they wrote that there will be nearly half of PCs powered by multi-GPU technology, such as ATI CrossFire or Nvidia SLI, in 2012)
3.JPR said the exact opposite in his report from what XbitLabs implying he said (He specifically said that for the future, hardware solutions like ATI CrossFire or Nvidia SLI is not fit because of perf. scaling and extra cost)
4.It seems that the main reason JPR made his report is to promote the LucidLogix Technologies chip solution as a viable tech. for the future.
I didn't even got a reply from anyone regarding this central JPR view, did anybody who argued read the report?
5.I disagree with him regarding the LucidLogix Technologies chip solution mainly because I think NV & ATI have the technological capability to make "SGXMP16 multi core type" GPUs so they will not need the Lucid tech for scaling,
and I think the internal cost for that is way lower than ATI & NV to pay Lucid from now (every freakin year for every chip) money.
It is clear enough from this reply & from your previous replies that you don't even know what JPR report is.
So in my point of view, JPR should have said multi core technologies (like the one is implemented in a model like the SGXMP16) instead of multi AIB systems (JPR meant multi GPU solutions with Lucid tech. solution) .
Please don't make another question, don't you understand that it is clear to anyone that you don't have clue about what JPR report is?
Like I said, what you are doing is, to arguing for the sake of argument!
Please read the JPR report and if you like make your own Topic with this title:
Why Imagination technologies should have listen to me!
regarding your disagreements about how Imagination technologies defines its GPUs!
SGXMP16 is a single SoC. I guess it's one die.
ATI's GPU is one die and NV's GPU is one die as well.
All of them have different cores but other than that I don't see what is so special about SGXMP16. ATI and NV can scale their chips as well.
No, the method that ATI & NV use to scale their cores (universally, not only scale the shaders or the texture units or what ever..., but scale everything to improve performance in all the aspects of the design) has certain disadvantages in relation with SGXMP16 in certain things.
I will give one example:
If ATI & NV scale 2X universally its cores they will need nearly twice the bandwidth
but the SGXMP16 scalling will bring way less bandwidth overhead than ATI's & NV's offerings.
I'm no expert here, just trying to wrap my head around this stuff. Correct away. (And forgive me, but I only speak ATI-ese...but somebody can probably translate.)
Looking at the architecture for the 4870, there are several levels of processors. The big question is, what should call a "core"? The stream processors (800), the MIMD vec5s (160), or the 16-way SIMDs (10)?
The term "processor" is rather ambiguous, but the Indiana University UITS knowledge base suggests a decent modern definition for a "core". According to them, a core is "a logical execution unit containing an L1 cache and functional units", which would seem to be the 16-way SIMD according to this site:
http://kb.iu.edu/data/avfb.html
http://www.anandtech.com/video...c.aspx?i=3341&p=4&cp=5
I don't know, does that seem right?
---
Anyway, it seems like the original article is suggesting more people will have multi-chip video cards like the 4870X2 by 2012, using either onboard Crossfire or SLI. Perhaps they'll make weaker versions with more GPUs, like a 4350X2 or something, for the masses.
Looking at the architecture for the 4870, there are several levels of processors. The big question is, what should call a "core"? The stream processors (800), the MIMD vec5s (160), or the 16-way SIMDs (10)?
The term "processor" is rather ambiguous, but the Indiana University UITS knowledge base suggests a decent modern definition for a "core". According to them, a core is "a logical execution unit containing an L1 cache and functional units", which would seem to be the 16-way SIMD according to this site:
http://kb.iu.edu/data/avfb.html
http://www.anandtech.com/video...c.aspx?i=3341&p=4&cp=5
I don't know, does that seem right?
---
Anyway, it seems like the original article is suggesting more people will have multi-chip video cards like the 4870X2 by 2012, using either onboard Crossfire or SLI. Perhaps they'll make weaker versions with more GPUs, like a 4350X2 or something, for the masses.
We have no increased load coming from games in the near future and we haven't seen an increase in the past 2 years while we have still been seeing rapid increases in GPU power, we also are no longer seeing increases in resolution eliminating another area that could continue to ramp demands for GPU power. In all, the benefits of multiple die graphics solutions is in a torrid state of decline, only a very small handful of games at the absolute highest settings have any real visible benefit for the most part, and this is right before a major new round of launches.
As single GPU solutions keep improving rapidly and demands for that power is not increasing at all, the realistic assesment at this point is that if anything is going to change it will be the extreme marginalization of everything beyond low/mid range single GPU solutions.
Like it or not at this point we are tied to the consoles in terms of graphics advancements, and on that note due to the example of Nintendo this generation it wouldn't be shocking if Sony and MS backed off a bit on their ramping up computational power(not that the next gen won't be considerably more powerful, but I would fully expect them to have much better margins on the hardware side next generation).
I can tell you right now, I will not have a multi-GPU setup in 2012. That's 100% negative. End of thread!
They claim:
Till they have a real solution and it's properly measured by independent sites I don't believe their marketing talk.
Their statements are too generic to draw any conclusion.
You have to understand that this topic is not about your beliefs.
What you are doing so far is to ruin my topic. (and now it's too late)
My original topic was about JPR report, what he ment or not, and if Xbitlabs misunderstood.
You were starting to make questions (becauce you said that you wanted to learn) that were not exactly the theme of my topic.
(it were related of cource but not what I asked in the main theme)
When I replied to you the first time answering your question, you immediately start to ask "argumentative" questions!
When I every time disproved your objections you came with another question!
Enough is enough!
Don't get me wrong, I love to communicate with people, but the way you do it, is not my style, this method for me is meaningless!
So back to the original topic.
The JPR report is called "Multi GPUs The Needs, Issues, and Opportunities". The whole document is about multi-GPU. That's what Xbitlabs reported.
The words "multi-core GPU" is used in the report only once, and it is in the summary at the end of the document.
Have you misunderstood the report? Why you implied they meant multi-core GPUs?
Originally posted by: Janooo
So back to the original topic.
The JPR report is called "Multi GPUs The Needs, Issues, and Opportunities". The whole document is about multi-GPU. That's what Xbitlabs reported.
The words "multi-core GPU" is used in the report only once, and it is in the summary at the end of the document.
Have you misunderstood the report? Why you implied they meant multi-core GPUs?
No I didn't misunderstood the report.
And anyway how is it possible for you to know, since before my previous reply you didn't even know what it says and you kept arguing without knowing anything for the report.
The Focus of my Topic was about JPR report in the sense of what he ment or not, if he was right or not and in the sense of if Xbitlabs misunderstood or not the report!
Your Focus until now was about multi core GPus definitions and why didn't Imagination technologies used your definition for their GPUs. (like they care)
And no, Xbitlabs didn't report correctly what JPR said in his research!
I didn't implied they meant multi-core GPUs I implied what I told you the previous time:
but you didn't even know before my previous reply, that in JP research, LucidLogix Technologies play a big role!
My friend, this is the last time I'm replying to you!
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: 22K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
News NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products- Started by poke01
- Replies: 403
-
Facebook
Twitter