Barcelona faster than expected?

[ObscureCaucasian]

While were doing the whole rumor thing:

http://www.fudzilla.com/index.php?option=com_content&task=view&id=649&Itemid=35

http://www.fudzilla.com/index.php?option=com_content&task=view&id=874&Itemid=35

I for one hope these are true.

 

[DrMrLordX]

Originally posted by: mamisano


Intel's already has dual Quad-Core parts available.

In server/workstation configurations, yes. In the desktop sector? No. AgenaFX will give you 8 cores on a desktop board supporting non-registered DIMMs which would be a first for either company.

 

[MarcVenice]

I wonder how much overclocking headroom those barcelona/agena cpu's will have. Can you imagine pushing a stockspeed of lets say 2.5ghz up to 3.4ghz or so, whilst 2.5ghz beats c2d's running at 3ghz ? I bet those things will run fairly hot though, and tuniq might see an increase in sales

 

[Viditor]

Originally posted by: keysplayr2003
Yes, I remember some SGI "desktop" systems with only 2 processors about 5 or six years ago where I used to work. Those small systems were close to $16,000 each. I was baffled at the price tags on those things. I couldn't imagine a 16 proc system, or whatever they have to offer these days. These were specialized systems of course. Proprietary software for very specific tasks, and they did them well.

You're right about quad core being the standard for a while though. I can't see any average to enthusiast gamer needing any more than that for a bit.

If you want a killer Folding rig however, the sky is the limit (actually the individual's wallet is) on the number of cores. Then again, simply use a few rigs with G80's and R600's to get the folding power they need. Would be cheaper. Funny that.

To give you an idea on how much cheaper things have become, that SGI system I used originally cost $1.2 Million (not counting software)...to get superior performance from a similarly configured hypothetical 16 core Opteron system would most likely be closer to the $30-40k range (and most of that would be the Ram cost).

 

[ObscureCaucasian]

Originally posted by: MarcVenice
I wonder how much overclocking headroom those barcelona/agena cpu's will have. Can you imagine pushing a stockspeed of lets say 2.5ghz up to 3.4ghz or so, whilst 2.5ghz beats c2d's running at 3ghz ? I bet those things will run fairly hot though, and tuniq might see an increase in sales

That CPU-Z screen from Fudzilla shows a 1.9 ghz Agena running at 3.05 with a minor bump in Vcore. That was on an engineering sample, so I'm not sure how that will compare to the retail parts.

 

[amenx]

The big question is how Barcelona will perform against Intels Penryn, no point in looking at it from what Intel currently has on offer.

 

[ObscureCaucasian]

Well really Penryn is just Conroe with SSE4 and 45nm, so it won't be much faster (if at all). It may be able to clock higher and run at lower voltages, but its performance should be in the ballpark of the current C2D offerings. We won't see a big change until Nehalem.

 

[Regs]

Originally posted by: BlameCanada
Well really Penryn is just Conroe with SSE4 and 45nm, so it won't be much faster (if at all). It may be able to clock higher and run at lower voltages, but its performance should be in the ballpark of the current C2D offerings. We won't see a big change until Nehalem.

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2972&p=2

We are also well into the 2nd quarter with no preview of what the Barcelona has to offer. So you can't blame people for being less than optimistic at this point. The delay of the x2900 was the icing on the cake.

 

[Griswold]

Originally posted by: amenx
The big question is how Penryn will perform against K10.

Corrected it

 

[Viditor]

Originally posted by: Regs

Originally posted by: BlameCanada
Well really Penryn is just Conroe with SSE4 and 45nm, so it won't be much faster (if at all). It may be able to clock higher and run at lower voltages, but its performance should be in the ballpark of the current C2D offerings. We won't see a big change until Nehalem.

http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=2972&p=2

We are also well into the 2nd quarter with no preview of what the Barcelona has to offer. So you can't blame people for being less than optimistic at this point. The delay of the x2900 was the icing on the cake.

Actually, it's that very point that (to me at least) gives Charlie's story a tad more credence...
If AMD has indeed found that the current re-spun stepping gives K10 a much bigger speed range, then it makes sense that they would hold off showing the chip until they had enough of the new stepping for a marketing preview.

 

[dmens]

Originally posted by: Viditor
Actually, it's that very point that (to me at least) gives Charlie's story a tad more credence...
If AMD has indeed found that the current re-spun stepping gives K10 a much bigger speed range, then it makes sense that they would hold off showing the chip until they had enough of the new stepping for a marketing preview.

unlikely, there's always top bin parts to do previews and brags, even on first spins. top bin a-steps can be faster than upper range productized steps.

for example, the frequency spread of a certain p4 family on first spin was over 1ghz, and the best parts were the same as the eventual high end market segment.

 

[Viditor]

Originally posted by: dmens

Originally posted by: Viditor
Actually, it's that very point that (to me at least) gives Charlie's story a tad more credence...
If AMD has indeed found that the current re-spun stepping gives K10 a much bigger speed range, then it makes sense that they would hold off showing the chip until they had enough of the new stepping for a marketing preview.

unlikely, there's always top bin parts to do previews and brags, even on first spins. top bin a-steps can be faster than upper range productized steps.

for example, the frequency spread of a certain p4 family on first spin was over 1ghz, and the best parts were the same as the eventual high end market segment.

True, but the reason for the excitement (according to Charlie) is a rework of the memory controller. If that is the case, then the top bins of the B0 parts could very well be substantially better...

Edit: This just in...
Agena FX and Stream demo

Again, no numbers available...but they did do a 720p conversion in realtime on the 8 core system (all 8 cores were maxed).

 

[lopri]

Until Intel's 45nm chips arrive, their current 65nm quad-core parts seem to max out around 3.0~3.3GHz, and that's already quite pushing it. (Just like 90nm A64 tops out around 2.8~3.0GHz) I could imagine one more speed bump over current QX6800, which will probably be dubbed QX6900, and that will not be a mass-quantity SKU. (Needs cherry-picking)

If, I mean IF (since I don't know what the real story is), the initial batch of K10 which is native quad-core can do 2.7~2.8GHz, and possibly with a revision if it can reach ~3.0GHz, I'd say AMD will have a comfortable performance crown till the end of year. After that, the battle will turn to a clock speed race until Nehalem arrives. Yorkfield's performance increase over Kentsfield (clock for clock), to my eyes, comes mainly from added L2 as well as revised platform (X38). Suppose a same clock speed, same amount of L2, and same platform, I'd venture to guess that Yorkfield's performance increase over Kentsfield would be <5%. So basically Yorkfield is a die-shrink of Kentsfield, and some tweaks and extra L2. (SSE4.. well, I doubt we'll see any app that'll take advantage of it any time soon) But it should certainly clock higher than Kentsfield thanks to improved manufacturing process.

Overall, if the rumours bare any truth, AMD and Intel will be more on even battlefield, the major selling factor being the price of total platform. Intel has an advantage here because their new chipset is out already and will mature over time, not to mention their decade-long expertise. On the other hand, I am not sure what AMD has in store for AM2+/AM3. (See, now we know why AMD so wanted and had to buy ATI. AMD has been way too much reliant on NV ever since Socket 939 was introduced) Then again, fortunately K10 will have integrated memory controllers just like K8, so it shouldn't be too hard to produce a solid chipset/boards. Performance/feature-rich chipset is another story, though.

Not only that, under the assumption of superior performance to Kentsfield/Yorkfield, I doubt we'll see the 'bargain-bin' prices from Barcelona. Can't really blame AMD on that front since they're probably losing money on many chips they're selling right now so they need to recover the financial damage. Expect to see the return of pre-Conroe prices from AMD.

 

[GFORCE100]

Originally posted by: lopri
I could imagine one more speed bump over current QX6800, which will probably be dubbed QX6900, and that will not be a mass-quantity SKU. (Needs cherry-picking).

Expect the QX6850 G0 1333MHz FSB 3GHz version in Q3. This will the last 65nm quad core part from Intel. Anything after will be a 45nm part.

 

[erwos]

Very nice. Anything that can get Intel to push up the price drop schedule is welcome by me. Hell, maybe I'll even buy a Barcelona...

 

[dmens]

Originally posted by: Viditor
True, but the reason for the excitement (according to Charlie) is a rework of the memory controller. If that is the case, then the top bins of the B0 parts could very well be substantially better...

Edit: This just in...
Agena FX and Stream demo

Again, no numbers available...but they did do a 720p conversion in realtime on the 8 core system (all 8 cores were maxed).

a to b would not include major changes. if the part did gain a lot from a to b, the only cause is a-step was horridly broken in an unexpected way. so, if inq is even accurate to begin with on the improvement:

- they're not excited, they're relieved.
- the target frequency is now met with b-step, not exceeded (no design team tries to exceed targets on a first spin... just doesnt happen)
- manager heads have already rolled.

still waiting for benchmarks here.

 

[ObscureCaucasian]

It could be advancements in AMD's 65nm process that contributed to the 500mhz gain.

 

[Stoneburner]

I still don't have any idea on what date i can go and buy a barcelona core to upgrade my system.

 

[the Chase]

Originally posted by: Stoneburner
I still don't have any idea on what date i can go and buy a barcelona core to upgrade my system.

http://www.dailytech.com/article.aspx?newsid=7232

Christmas- if you take their word for it. Hmmm... no solid dates at all... If Penryn and X38 chipset with DDR3 is somewhat reasonably priced around Sept./Oct. I am NOT going to wait on AMD to execute. Still be waiting May of 2008.

 

[DrMrLordX]

If you have an AM2 system, probably not until October/November, but it's still up in the air.

 

[SexyK]

Originally posted by: lopri
Until Intel's 45nm chips arrive, their current 65nm quad-core parts seem to max out around 3.0~3.3GHz, and that's already quite pushing it. (Just like 90nm A64 tops out around 2.8~3.0GHz) I could imagine one more speed bump over current QX6800, which will probably be dubbed QX6900, and that will not be a mass-quantity SKU. (Needs cherry-picking)

If, I mean IF (since I don't know what the real story is), the initial batch of K10 which is native quad-core can do 2.7~2.8GHz, and possibly with a revision if it can reach ~3.0GHz, I'd say AMD will have a comfortable performance crown till the end of year. After that, the battle will turn to a clock speed race until Nehalem arrives. Yorkfield's performance increase over Kentsfield (clock for clock), to my eyes, comes mainly from added L2 as well as revised platform (X38). Suppose a same clock speed, same amount of L2, and same platform, I'd venture to guess that Yorkfield's performance increase over Kentsfield would be <5%. So basically Yorkfield is a die-shrink of Kentsfield, and some tweaks and extra L2. (SSE4.. well, I doubt we'll see any app that'll take advantage of it any time soon) But it should certainly clock higher than Kentsfield thanks to improved manufacturing process.

Overall, if the rumours bare any truth, AMD and Intel will be more on even battlefield, the major selling factor being the price of total platform. Intel has an advantage here because their new chipset is out already and will mature over time, not to mention their decade-long expertise. On the other hand, I am not sure what AMD has in store for AM2+/AM3. (See, now we know why AMD so wanted and had to buy ATI. AMD has been way too much reliant on NV ever since Socket 939 was introduced) Then again, fortunately K10 will have integrated memory controllers just like K8, so it shouldn't be too hard to produce a solid chipset/boards. Performance/feature-rich chipset is another story, though.

Not only that, under the assumption of superior performance to Kentsfield/Yorkfield, I doubt we'll see the 'bargain-bin' prices from Barcelona. Can't really blame AMD on that front since they're probably losing money on many chips they're selling right now so they need to recover the financial damage. Expect to see the return of pre-Conroe prices from AMD.

Do you have any factual basis for saying Yorkfield's performance increase over Kentsfield will be <5%, when in the article linked in this very thread AT shows a 5-10% gain clock for clock at a minimum in testing with the actual chips in hand (as opposed to your mere speculation).

 

[lopri]

Originally posted by: SexyK
Do you have any factual basis for saying Yorkfield's performance increase over Kentsfield will be <5%, when in the article linked in this very thread AT shows a 5-10% gain clock for clock at a minimum in testing with the actual chips in hand (as opposed to your mere speculation).

Suppose a same clock speed, same amount of L2, and same platform,

Under the hypothetical condition, the only variable being some tweaks on the core and increased FSB. (something like 2.40GHz Clawhammer vs 2.40GHz San Diego, or G70 vs G71 under same frequency) But as you correctly pointed out, the truth is Yorkfield will have a bigger L2, higher FSB and higher clocks, as well as new platform supporting DDR3. With all those combined, yes the performance gain should be 5~10%, or possibly even more than that depending on how DDR3 evolves over time. (and say 3 years from now when apps actually make use of SSE4, the gap will only widen)

 

[Amaroque]

Inreresting discussion....

If AMD had Intels mfg technology, the X2's would be way over 3 GHz rite now (just a thought). A joint effort (AMD Intel) would be super sweet, but I wouldn't want to see the prices! :Q I know that would never happen, but it's fun to think about the tech.

I have C2D systems, and X2 systems, and I like both platforms alot (I'm a DC junkie). X2 is lagging behind the C2D... but for who? CPU's are so far ahead of current software (for the average user) rite now anyway.

My 2¢... I think Barcelona, and Penryn are way ahead of their time for 99% of users.

 

[Idontcare]

Originally posted by: Amaroque
Inreresting discussion....

If AMD had Intels mfg technology, the X2's would be way over 3 GHz rite now (just a thought). A joint effort (AMD Intel) would be super sweet, but I wouldn't want to see the prices! :Q I know that would never happen, but it's fun to think about the tech.

I have C2D systems, and X2 systems, and I like both platforms alot (I'm a DC junkie). X2 is lagging behind the C2D... but for who? CPU's are so far ahead of current software (for the average user) rite now anyway.

My 2¢... I think Barcelona, and Penryn are way ahead of their time for 99% of users.

I doubt any sensible extrapolation can be made between Intel's bulk Si CMOS mfg capabilities and AMD's SOI CMOS mfg capabilities.

Whether Intel could make a 3GHz X2 (in SOI or in bulk Si) is a truly not an answerable question.

 

[Amaroque]

I know it's not an answerable question. I was just kind of thinking out loud... lol!