VirtualLarry
No Lifer
- Aug 25, 2001
- 56,587
- 10,225
- 126
So when are we going to see the six-core desktop models? I want to see more WRs broken at 6.5Ghz under liquid helium, on a six-core!
Cookie Monster
Diamond Member
- May 7, 2005
- 5,161
- 32
- 86
Yea true
On a side note, I dont think desktop versions exists since these aren't exactly based on AM3 socket, but rather the existing socket F (or socket 1207).
Would be interesting if these fit on the AMD quad FX platform though.
What possible use would a 6-core desktop chip be??? (though I can empathise with the "cool" factor involved there...)
Well you also have to consider that AMD's current chips are running well within their TDP guidelines.... for example on the desktop the Phenom II 940 consumes ~90W or so despite having a 125W TDP. And that is with 3.0GHz @ 1.35V..... not power efficient at all.
Current Shanghai quads go up to 2.8GHz @ 105W ACP and the 2.7GHz chip fits into a 75W ACP.
With the real power consumption we see from Deneb @ 3.0GHz @ 1.35V....... I don't see any reason that we can't see Istanbul @ 2.8GHz if AMD can reduce the operating voltage to 1.25V or so. Not with the current C2 revision, but by the time Istanbul ships AMD is likely to have made some improvements to their process.
TDP does not equal average/typical power consumption, it is a value needed for the design of power dissipation by both the HSF as well as the computer housing equipment and the larger computer center for AC purposes.
I understand TDP's greatly exceed actual chip power consumption in the environments we mere enthusiasts use them (and measure them) at, it is true for Intel as well as AMD.
But surely we can agree the engineers and marketing teams at these companies are equally aware of the findings you site, and yet they feel compelled (for some good reason we must assume) to continue to market the CPU's with the higher TDP rating.
Whatever level of caution, engineering margin, actual computer center practicalities they are targeting and accomodating, I doubt very much those levels will change with the arrival of a 6-core chip.
The response to your question regarding 45nm Deneb power improvements is contained in my unanswered question from above:
What you argue for is not impossible, it is possible that AMD will make a 6GHz 6-core chip on 45nm using 0.9V. Anything is possible if we just want to assume it is possible. But is it probable? Is it plausible? Is it likely?
DrMrLordX
Lifer
- Apr 27, 2000
- 22,937
- 13,024
- 136
Great minds think alike!
Originally posted by: Flipped Gazelle
Would have been a great time to introduce Socket 1453!
:beer:
Ba dum, tsh.
I see your point, but remember that this was a demo in which AMD did not disclose clock speeds. We have no idea how fast the demo system was running or whether or not the clockspeeds of the demo system will be indicative of the clock speeds of products at launch. In fact, I don't think we even know what clock speeds were used on the Shanghai system (at least not based on the article).
Yes, one would assume that lower clock speeds and/or lower VIDs would be required to replace a 16-core 4xShanghai board with a 24-core 4xIstanbul board given the same enclosure, cooling, etc. unless the system had more-than-adequate cooling to begin with. Power supply output might also be an issue, though I would hope not.
To assume that AMD's tech demo was governed by any of the real-world concerns stemming from replacing a 16-core board with a 24-core board, however, is not necessarily wise. What if they made sure the cooling for their test bed was more-than-adequate for both platforms to run at the same clock speed? I'm sure they want us to think the demo represents a real-world board swap but how do any of us know for sure?
Challenging Intel's dominance in the desktop realm when it comes to apps that can utilize more than 4 cores. Core i7 can already handle 8 threads, and once octal-core Nehalem hits the desktop scene, we're looking at 16 threads. Granted, that's a pretty small niche, but when it comes to video encoding, the desire for more cores is there.
While others have answered your question well enough with that artistic rendering of the Istanbul core, I'll agree that it would appear Istanbul is not an MCM solution. AMD will be releasing a 12-core MCM chip called Mangy-Cours in 2010.
It's still far too early to presuppose anything about power usage or clockspeed IMHO...
There are too many variables that could be different...lower voltage, improved power efficiency on the various elements (mem controller, HT controller), etc...
I have yet to see that Hyperthreading on the i7 chips adds anything. It's not really possible to know for sure because you can't turn it off on the Nehalem...
And I agree that it's a very small niche...small enough that I don't think it makes sense to create a line of chips for it...
You are arguing about max-capable clockspeeds, which is not what I am talking about at all. I am strictly speaking to the GHz/watt aspect. For sure there is nothing here to preclude an upper limit on absolute GHz.
But let's not delude ourselves into thinking that AMD's 45nm with Shanghai is so seriously bad that the power-consumption can be improved by some 30-40% when they release Istanbul so that the GHz/watt remains the same while doubling the number of cores.
I am merely attempting to appeal to common sense here, I am not trying to make any ground-breaking statements or conclusions or assumptions or statements regarding AMD's future clockspeeds with Istanbul.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
VirtualLarry
No Lifer
- Aug 25, 2001
- 56,587
- 10,225
- 126
DrMrLordX
Lifer
- Apr 27, 2000
- 22,937
- 13,024
- 136
Of course. But there's no reason to believe that the AMD guys didn't just pop in an Istanbul system that ran at the same clock speed as the Shanghai system. They most likely were not limited by thermals.
It does make sense that an actual system integrator would have to think about heat and power consumption when selecting Istanbul drop-in replacements for Shanghai, but it also makes sense that the AMD guys running the demo would just throw in a similarly-clocked Istanbul system to put Istanbul in the best possible light.
I'm sure anyone viewing the demo would like for them to have put in a system with similar thermals to emulate what could be anticipated in the real world, but I am not so confident that the demo was actually set up that way.
In the end, I guess it boils down to whether or not you really trust the guys running the demo to provide useful data. The above-linked article's author obviously does, but . . .
Indeed. I think the author kind of jumped the gun when started making assumptions about clockspeeds.
Originally posted by: Viditor
I have yet to see that Hyperthreading on the i7 chips adds anything. It's not really possible to know for sure because you can't turn it off on the Nehalem...
And I agree that it's a very small niche...small enough that I don't think it makes sense to create a line of chips for it...
Phynaz's link provides some interesting rebuttal, though I wasn't thinking about games as much as high-quality video encoding, 3d-rendering, and any other apps that can load all 4 physical and all 4 virtual cores on an HT-enabled Core i7. Stuff that can load all 8 cores is probably going to run faster with the extra virtual cores.
HT aside, once octal-core Nehalem hits the scene on the desktop/1P Workstation market, AMD will want to fire back with something. A Socket AM3 Istanbul-based Opteron would be a good place to start.
I was going from a report using an early bios, and I'm glad you caught me on that...
Interesting reading in the Techreport review!
1. In games, HT appears to slow the i7 down in at least 2 cases (by about 7.5%), speed it up in only one case (by 11%), and stay about the same (+/-2%) in the majority of cases.
2. The most interesting datum of all...TR have a few cases where they compare the same software using progressively more threads (1, 2, 4, 6, and 8 threads).
a.) Using 1 and 2 threads, HT and non-HT are about the same
b.) Using 4 threads, HT is significantly SLOWER than non-HT
c.) Using 6 threads, HT is slightly faster than non-HT
d.) Using 8 threads, HT is significantly FASTER than non-HT
While a, c, and d are to be expected, b was quite a surprise to me...
In MyriMatch the difference was ~14%
In STARS Euler3d the difference was ~27%
Also, even in many of the media encoders...x264 HD video pass 1, Windows media encoder (which coincidentally uses only 4 threads), videowave, Lame, etc..., HT decreases performance.
So again, I reiterate that HT isn't always a good thing...
Yeah, that's one of my favourite jokes as well...(sorry, Mark)
The Techreport Phynaz linked (linked again here) has it listed...
Originally posted by: Viditor
I was going from a report using an early bios, and I'm glad you caught me on that...
Interesting reading in the Techreport review!
1. In games, HT appears to slow the i7 down in at least 2 cases (by about 7.5%), speed it up in only one case (by 11%), and stay about the same (+/-2%) in the majority of cases.
2. The most interesting datum of all...TR have a few cases where they compare the same software using progressively more threads (1, 2, 4, 6, and 8 threads).
a.) Using 1 and 2 threads, HT and non-HT are about the same
b.) Using 4 threads, HT is significantly SLOWER than non-HT
c.) Using 6 threads, HT is slightly faster than non-HT
d.) Using 8 threads, HT is significantly FASTER than non-HT
While a, c, and d are to be expected, b was quite a surprise to me...
In MyriMatch the difference was ~14%
In STARS Euler3d the difference was ~27%
Also, even in many of the media encoders...x264 HD video pass 1, Windows media encoder (which coincidentally uses only 4 threads), videowave, Lame, etc..., HT decreases performance.
So again, I reiterate that HT isn't always a good thing...
How do you recon HT will play out in the market AMD will position Istanbul in?
Originally posted by: jones377
Originally posted by: Viditor
I was going from a report using an early bios, and I'm glad you caught me on that...
Interesting reading in the Techreport review!
1. In games, HT appears to slow the i7 down in at least 2 cases (by about 7.5%), speed it up in only one case (by 11%), and stay about the same (+/-2%) in the majority of cases.
2. The most interesting datum of all...TR have a few cases where they compare the same software using progressively more threads (1, 2, 4, 6, and 8 threads).
a.) Using 1 and 2 threads, HT and non-HT are about the same
b.) Using 4 threads, HT is significantly SLOWER than non-HT
c.) Using 6 threads, HT is slightly faster than non-HT
d.) Using 8 threads, HT is significantly FASTER than non-HT
While a, c, and d are to be expected, b was quite a surprise to me...
In MyriMatch the difference was ~14%
In STARS Euler3d the difference was ~27%
Also, even in many of the media encoders...x264 HD video pass 1, Windows media encoder (which coincidentally uses only 4 threads), videowave, Lame, etc..., HT decreases performance.
So again, I reiterate that HT isn't always a good thing...
How do you recon HT will play out in the market AMD will position Istanbul in?
A good question...the truth is that I don't know yet because we haven't seen those benchmarks. The big test will be how it benches with VMWare as that is increasingly the fastest growing server market area. I don't really know if HT helps or hinders it yet...
Originally posted by: Viditor
Originally posted by: jones377
Originally posted by: Viditor
I was going from a report using an early bios, and I'm glad you caught me on that...
Interesting reading in the Techreport review!
1. In games, HT appears to slow the i7 down in at least 2 cases (by about 7.5%), speed it up in only one case (by 11%), and stay about the same (+/-2%) in the majority of cases.
2. The most interesting datum of all...TR have a few cases where they compare the same software using progressively more threads (1, 2, 4, 6, and 8 threads).
a.) Using 1 and 2 threads, HT and non-HT are about the same
b.) Using 4 threads, HT is significantly SLOWER than non-HT
c.) Using 6 threads, HT is slightly faster than non-HT
d.) Using 8 threads, HT is significantly FASTER than non-HT
While a, c, and d are to be expected, b was quite a surprise to me...
In MyriMatch the difference was ~14%
In STARS Euler3d the difference was ~27%
Also, even in many of the media encoders...x264 HD video pass 1, Windows media encoder (which coincidentally uses only 4 threads), videowave, Lame, etc..., HT decreases performance.
So again, I reiterate that HT isn't always a good thing...
How do you recon HT will play out in the market AMD will position Istanbul in?
A good question...the truth is that I don't know yet because we haven't seen those benchmarks. The big test will be how it benches with VMWare as that is increasingly the fastest growing server market area. I don't really know if HT helps or hinders it yet...
Well, it won't be long until we find out. The only real test we have so far is SAP-SD from 3 different vendors running Gainestown and it smokes everything else including Shanghai. While I have no ultimate proof that you seem to demand, I think that HT makes a BIG difference in that one test.
Any server test, using primarily the integer units, that maxes out the number of threads should see a noticeable boost from HT, that's just common sense. I think you know this already but it's easier to say I don't know when it's something you don't want to know
Originally posted by: Viditor
I was going from a report using an early bios, and I'm glad you caught me on that...
Interesting reading in the Techreport review!
1. In games, HT appears to slow the i7 down in at least 2 cases (by about 7.5%), speed it up in only one case (by 11%), and stay about the same (+/-2%) in the majority of cases.
2. The most interesting datum of all...TR have a few cases where they compare the same software using progressively more threads (1, 2, 4, 6, and 8 threads).
a.) Using 1 and 2 threads, HT and non-HT are about the same
b.) Using 4 threads, HT is significantly SLOWER than non-HT
c.) Using 6 threads, HT is slightly faster than non-HT
d.) Using 8 threads, HT is significantly FASTER than non-HT
While a, c, and d are to be expected, b was quite a surprise to me...
In MyriMatch the difference was ~14%
In STARS Euler3d the difference was ~27%
Also, even in many of the media encoders...x264 HD video pass 1, Windows media encoder (which coincidentally uses only 4 threads), videowave, Lame, etc..., HT decreases performance.
So again, I reiterate that HT isn't always a good thing...
I agree scenario "b" is perplexing as one might first pass over these results as just more proof that context switching (thread migration) is a bad thing, but the performance degradation from thread migration should be most amplified when running single-threaded apps while comparing HT vs. no HT.
So what causes the marked performance degradation for 4 threads when HT is on vs off? Is it really forced stalls in the pipeline because two threads are making requests to utilize the same physical hardware simultaneously? (I can't think of any other reason that would not also make 1-2 thread apps run horribly bad with HT on as well)
HT needs a hardware-based manager that dynamically disables and enables the HT aspects of the chip as predominately active threads either fall below 4 or above 4. (or 8 on beckton, 2 on clarkdale, etc)
I agree that a hardware switch might be needed to avoid these performance drops (or better OS thread management). It appears that the OS sees the 4 cores as 8 cores, and will migrate threads to a core that already in use processing something else instead of moving it to one that is not in use due to HT.
It would of course be better if the OS sees the 4 cores as 4 cores, and only uses the additonal resources when those 4 cores are being used. I honestly don't know how the I7 and Windows XP/Vista work with this situation, but I have a feeling that the next iteration of the processor or OS will have a fix for this to do exactly that.
After reading through my post, I don't feel that I managed to communicate my oppinion clearly, but alas I don't have time to figure out a better way to write my thoughts on this matter.
Based on my work in other fields, I would guess it is for allowing the processor to run in places like Iraq on a 130F day. (I am just trying to help with a real world example of the design constraints that might be used.)
Originally posted by: jones377
Well, it won't be long until we find out. The only real test we have so far is SAP-SD from 3 different vendors running Gainestown and it smokes everything else including Shanghai. While I have no ultimate proof that you seem to demand, I think that HT makes a BIG difference in that one test.
Any server test, using primarily the integer units, that maxes out the number of threads should see a noticeable boost from HT, that's just common sense. I think you know this already but it's easier to say I don't know when it's something you don't want to know
The whole point is that "common sense" isn't always what really happens.
I refer of course to instances like the results from scenario "b" I mentioned above...
Will utilizing 4 threads always have those poorer results?
Does it change with 2 sockets?
Common sense says that a 24 core Xeon 7460 system should be faster in VMMark than a 16 core Opteron 8386 system...but it's not (Opteron 20.43, Xeon 19.99). While we can certainly begin to guess why that happened (now that the results are in), I don't know anyone who would have predicted it.
Originally posted by: Viditor
Originally posted by: jones377
Well, it won't be long until we find out. The only real test we have so far is SAP-SD from 3 different vendors running Gainestown and it smokes everything else including Shanghai. While I have no ultimate proof that you seem to demand, I think that HT makes a BIG difference in that one test.
Any server test, using primarily the integer units, that maxes out the number of threads should see a noticeable boost from HT, that's just common sense. I think you know this already but it's easier to say I don't know when it's something you don't want to know
The whole point is that "common sense" isn't always what really happens.
I refer of course to instances like the results from scenario "b" I mentioned above...
Will utilizing 4 threads always have those poorer results?
Does it change with 2 sockets?
Common sense says that a 24 core Xeon 7460 system should be faster in VMMark than a 16 core Opteron 8386 system...but it's not (Opteron 20.43, Xeon 19.99). While we can certainly begin to guess why that happened (now that the results are in), I don't know anyone who would have predicted it.
Common sense works if you know a bit of the things you talk about. I had never even heard of VMMark before a while back when it was shown with various configs. But it has been known for a long time that Opterons have had much lower virtualisation overhead than contemporary Xeons.
Thing is though, VMMark measures the performance of the various virtual servers running on the machine. So even if Nehalem-EX does nothing to reduce the virtualisation overhead compared with Dunnington (something I find very hard to believe since Intel has improved it in every processor update before) it will STILL perform better in VMMark because the apps within the virtual enviroments will see the improvements brought by the Nehalem architecture, including HT. Don't you agree?
Right now Shanghai is only a hair ahead of Dunnington in VMMark. Will Istanbul improve more over Shanghai than Nehalem-EX over Dunnington? I really doubt it. Again no ultimate proof, just an educated guess...
btw a tip to you guys interested in following AMD and Istanbul. The one feature it brings that you should remember is the probe filter. It may seem like just another insignificant checklist item to the untrained eye but believe me it won't be, especially for 4 and 8 socket systems
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: 411
-
Facebook
Twitter