AMD Shanghai is officially out

[Hornet331]

shanghai is running on ecc ram 800mhz and old HT 1,000mhz. phenom (agena) is faster then Barcelona by some.

I doubt faster memory will bring much for consumer apps, since beside from data compression programs thers virtually no need for high bandwidth. And since most of this test seem to be quite unaffected by bandwidth they give a good impression of what to expected from desktop apps (imho).

How much the higher NB/L3 speed is, i don't know but well see about that.

edit:
ops i missed justapost reply, he pretty much confirmed that faster NB/L3 speed only affects certain apps.

btw justapost could you run a quick test on games, faster L3 should help there a bit. (low res so that you dont run into a gpu limit, different NB speeds)

[Calmatory]

I doubt faster memory will bring much for consumer apps, since beside from data compression programs thers virtually no need for high bandwidth. And since most of this test seem to be quite unaffected by bandwidth they give a good impression of what to expected from desktop apps (imho).

How much the higher NB/L3 speed is, i don't know but well see about that.

edit:
ops i missed justapost reply, he pretty much confirmed that faster NB/L3 speed only affects certain apps.

btw justapost could you run a quick test on games, faster L3 should help there a bit. (low res so that you dont run into a gpu limit, different NB speeds)

High memory bandwidth helps when the data can't be found from CPU caches. With high memory bandwidth and low latency the cost of cache miss reduces, depending on how much data has to be fetched and what kind of a miss it is. So, for smaller cached CPU's the mem bandwidth is more crucial than for high cached CPU's.

[Hornet331]

High memory bandwidth helps when the data can't be found from CPU caches. With high memory bandwidth and low latency the cost of cache miss reduces, depending on how much data has to be fetched and what kind of a miss it is. So, for smaller cached CPU's the mem bandwidth is more crucial than for high cached CPU's.

yes, thats the theory, but real desktop apps show otherwise, else there would be a noticeable differents on K8/K10 with faster mem.

[Calmatory]

yes, thats the theory, but real desktop apps show otherwise, else there would be a noticeable differents on K8/K10 with faster mem.

Yes, in practice the e.g. 15 % highher bandwidth is usually useless unless the program really needs the bandwidth. With games there are marginal gains from going from 256 KB to 1024 KB L2 cache, or from 2 MB to 4 MB etc. No wonder the increased bandwidth and decreased latency has no practical impact when the data needed to fetch from RAM is usually just few bytes.

Oh, and even 10 % increase in GFX and CPU power can barely be seen in games, in practice. ...unless compared systems are next to each other.

[Hornet331]

Oh, and even 10 % increase in GFX and CPU power can barely be seen in games, in practice. ...unless compared systems are next to each other.

But at least they are somewhat measurable and dont fall into the margine of error.

[Calmatory]

But at least they are somewhat measurable and dont fall into the margine of error.

Yes. But then again, at least half of the gains are psychological rather than actual performance gains due to overclocking(or, in this case slightly improved CPU).

At least I get the "Now it runs better" feeling after I can stretch the last few MHz out of the CPU and RAM, even when I know that the gains are barely measurable with software, let alone by human. Makes me happy with my effort on getting the last drips.

[Jazzman]

merkk1 posted a site in the AMD section that states:

For HPC applications (Using SPECfp_rate2006), Shanghai at 2.7Ghz is faster than all current Intel processors (2-socket and 4-socket).

The referenced data is obtained from an independent testing site. HERE is where the quote came from.

[gallag]

merkk1 posted a site in the AMD section that states:



The referenced data is obtained from an independent testing site. HERE is where the quote came from.

Does "All current Intel cpus" include nehalem? or will it when it hits 2 socket? Any spec results for nelly yet?

[gallag]

Also, Dont forget that AMD also has a hefty lead in something far more important than spec, virtualization could be the future and is something it could still lead nelly in. Sorry for going ot but has anyone seen any nelly results for virtualization?

[Jazzman]

Does "All current Intel cpus" include nehalem? or will it when it hits 2 socket? Any spec results for nelly yet?

I have not seen Nehalem results for virtualization. Nehalem does have the benefit of providing the user the option of frying an omelet on the heatsink while running benchmarks.

[accord99]

Does "All current Intel cpus" include nehalem? or will it when it hits 2 socket? Any spec results for nelly yet?

C't got 110 in SPECint_rate2006 base and 84.7 in SPECfp_rate2006 with a 965 EE. An optimized official submission will probably do better.

http://www.heise.de/english/newsticker/news/118430

I have not seen Nehalem results for virtualization. Nehalem does have the benefit of providing the user the option of frying an omelet on the heatsink while running benchmarks.

I'd hate to think what your Phenom would do to an egg.

[gallag]

I have not seen Nehalem results for virtualization. Nehalem does have the benefit of providing the user the option of frying an omelet on the heatsink while running benchmarks.

Sad, If this was shinti commenting on AMD like that you would be all over him.

[verndewd]

great initial numbers iMO considering the abysmal recent past, I find it curious that HKMG is pushed back so far but it is likely for good reason. also an interesting happening the IBM AMD coalition is looking into bulk, Likely due to the expected 32nm and lower transitions power envelopes? Bulk is also more easilly adaptable to hkmg processes as far as I have read.

[villa1n]

Does "All current Intel cpus" include nehalem? or will it when it hits 2 socket? Any spec results for nelly yet?

Don't think there are any 2p i7's out there atm, and the server side isnt expected until 2h 2009.. so maybe in a few months.

[Shintai]

Please no _rate benchmarks.

And yes, an i7 beats a 2 socket Shanghai in rate benches. Only due to trichannel and fast memory (up to 2000Mhz). With 2 socket Xeon 5500s you gonna see insane rate benches. But they are utter crap. Its abit like running sandra memory bench on the systems and claim thats how you measure scaling of computational power.

Rates benches are an ancient leftover that basicly means nothing. Specially for the "normal" server park/datacenters.

great initial numbers iMO considering the abysmal recent past, I find it curious that HKMG is pushed back so far but it is likely for good reason. also an interesting happening the IBM AMD coalition is looking into bulk, Likely due to the expected 32nm and lower transitions power envelopes? Bulk is also more easilly adaptable to hkmg processes as far as I have read.

How big was this 32nm SRAM part? Intels was 291Mbit and 2billion transistors running at 3.8Ghz. TSMC was 2mbit. What was AMDs?

AMD also showed 45nm SRAM year(s) ago. There is just along way from SRAM and SRAM sizes to actual products and ramping the facility. And the NFC (Like TSMCs) roadmaps are ramp, not production.

Its easy to get fooled by it. AMDs own roadmap is perhaps the best indication when with 32nm desktop parts in 2011.

[Miss Banana]

How big was this 32nm SRAM part? Intels was 291Mbit and 2billion transistors running at 3.8Ghz. TSMC was 2mbit. What was AMDs?

Dude, seriously, you need to chill with the constant pro intel rambling.
Vern only said it is interesting that the AMD IBM coalition is looking into bulk in regards to 32nm, there is no justification whatsoever for these attempts to turn the thread into an intel vs amd thread.

[Calmatory]

Dude, seriously, you need to chill with the constant pro intel rambling.
Vern only said it is interesting that the AMD IBM coalition is looking into bulk in regards to 32nm, there is no justification whatsoever for these attempts to turn the thread into an intel vs amd thread.

What, he is just pointing out the obvious and telling the truth that AMD is not competitive, the slides and marketting will get people disapointed once again.


Tbh I don't know if I was being sarcastic even. Reader may decide.

[demonkevy666]

I'd hate to think what your Phenom would do to an egg.

Hex had a icore7 on a TRUE hitting 83C overclocked.

phenom over clocked only hit's about 60C and lower wit ha decent cooler.

[Donnie27]

Sad, If this was shinti commenting on AMD like that you would be all over him.

True, just as Miss Banana showed LOL!

Anyway, come on AMD, bring it to the Desktop at a fair price.

[Donnie27]

Hex had a icore7 on a TRUE hitting 83C overclocked.

phenom over clocked only hit's about 60C and lower wit ha decent cooler.

But Phenom is still slower so it should be cooler I wonder how hot it would get at 3.8 or 4GHz?

http://www.tbreak.com/articles/13/1/...ing/Page1.html

No surprises with the temperatures for the 130W CPU’s especially the 965EE hitting 4GHz on a stock air cooling solution.

HardOCP (Video) showed 92 with stock cooler. Who'd use a stock cooler to overclock? 83 on the Ultra 120 and 80 with water for them but that's a 2.66@3.8GHz"We Ran them for days without problems at these these *temps." Except for the stock cooler, 3.8GHz still over a 1GHz on the processors many said wouldn't overclock. 80C with water and 83 with at T-120.

AMD still has their work cut out for them.

[Jazzman]

But Phenom is still slower so it should be cooler I wonder how hot it would get at 3.8 or 4GHz?

Oh, a Deneb running at 4.0Ghz about 53C on AIR..... (100% core load of course)

[Stukov]

Oh, a Deneb running at 4.0Ghz about 53C on AIR..... (100% core load of course)

There is a C1 stepping (if it aint fake that is). Isn't this the SS that was argued about for weeks/months on end?

[Macadamia]

But Phenom is still slower so it should be cooler I wonder how hot it would get at 3.8 or 4GHz?

http://www.tbreak.com/articles/13/1/...ing/Page1.html



HardOCP (Video) showed 92 with stock cooler. Who'd use a stock cooler to overclock? 83 on the Ultra 120 and 80 with water for them but that's a 2.66@3.8GHz"We Ran them for days without problems at these these *temps." Except for the stock cooler, 3.8GHz still over a 1GHz on the processors many said wouldn't overclock. 80C with water and 83 with at T-120.

AMD still has their work cut out for them.

Indeed.

Deneb Rev C1 got very close to MM's i7 965 air record (both on air), the one I'm talking about is REAL and from someone at MSI. We just don't know what temperatures, but it's manageable on a Oorochi.

http://forum.coolaler.com/showpost.p...4&postcount=50


ITOCP had some temps on the 3.2Ghz (supposedly 1.5V) Deneb too, was <45C, on air but no idea of the cooler.

[demonkevy666]

that Oorochi was 4.3ghz.

there are many benchmarks but most are synthetic, haven't seen much real game ones.

I know there was q9650 vs an Icore7.

[villa1n]

There is a C1 stepping (if it aint fake that is). Isn't this the SS that was argued about for weeks/months on end?

I thought was was a photochopped one, since the 4006 and 4009 are quite obviously different sizes.