Tuesday, September 1st 2009

AMD Preparing ''Thuban'' Desktop Six-Core Processor

AMD is planning to create a desktop implementation of its Opteron "Istanbul" monolithic six-core processor. Codenamed "Thuban" (named after a star in the Draco constellation, which means Dragon), the new processor will be based on the socket AM3 package for compatibility with existing and future desktop core logic. It features six cores, 9 MB of total cache (6 x 512 KB L2 + 6 MB L3). Thuban is aimed to make for AMD's high-end desktop processor, as the company prepares to face competition from a near-complete lineup of processors based on the Nehalem/Westmere architectures from Intel. It is expected to be the posterboy for AMD's "Leo" high-end consumer desktop platform that succeeds its current Dragon platform.

Some of the key components that make up AMD Leo platform are the upcoming AMD 890FX and 890GX chipset, companion SB800 series southbridge chips, and members of AMD's Evergreen family of DirectX 11 compliant graphics processors. On the software front, AMD will give its Fusion and Overdrive utilities some big updates. The SB800 series southbridge chips will feature native support for SATA 6 Gb/s and USB 3.0; connectivity is further enhanced by integrated Broadcom MAC Ethernet interfaces. While the Leo platform is expected to launch almost simultaneously with the 8-series chipsets, the six-core Thuban processor on the other hand comes later. It is due only in Q3 2010. Thuban will have come out an year after its enterprise implementation in the form of Opteron "Istanbul".
Sources: X-bit Labs, VR-Zone
Add your own comment

93 Comments on AMD Preparing ''Thuban'' Desktop Six-Core Processor

#76
Unregistered
TroubledWhat is kinda funny is i do all of that stuff with ease too...and i dont have a Core i7. Go figure. You make it sound like you need a Core i7 to do all of the many things that a computer can do....when you don't really. I didnt really spend all that much on my video cards either. The Radeon 4670 isn't considered top of the line...but good enough to play the games that i play...Crysis, Far Cry 2, Need For Speed, and all the many other games that i play...along with the Zipping and Unzipping of files, dvd encoding and the many other things I do.
Yeah, plus you have to consider memory too, such as 8gb is going to do better than 6gb like 12gb would be better than 8gb etc...It's not just the cpu or the video card
#77
Hayder_Master
btarunrYes, like ShadowFold said, the past two flagship desktop dies are named after stars: Agena (aka Beta Centauri from constellation Centaurus), and Deneb (aka Alpha Cygni from constellation Cygnus). The naming has something to do with K10 "Stars" design.
now i see AMD chose nice names for them cpu's , thanx for this cool info btarunr and ShadowFold :toast:
Posted on Reply
#78
eidairaman1
The Exiled Airman
This should be the Last of the K10 Arch AMD CPUs before they release their CPU built from the ground up. Since Shanghai is out in the Servers this is a derivative of that class.
Posted on Reply
#79
btarunr
Editor & Senior Moderator
TroubledOk...so a 5% increase with the 4mb Larger L3...but note...i did not say L3...i said L2. L3 runs at the speed of the NB...unlike the L2 which runs at HT Link Speed. What i am saying...is we have been at 512kb per core for a long time...it is time to bump that up a little bit. I am sure that you would see yet another boost in performance.
Well then look back and compare Athlon X2 Windosor-2M (1 MB L2 cache / core) to Windsor-1M (512 KB L2 /core). You'll find that the difference isn't much, and that with a mere 100 MHz increase in clock speed, whatever extra performance the additional cache provides is made up for. The boost in performance is not the cache's job. At the end of the day, your cache isn't crunching numbers. When there's a fast interconnect between the system and the processor, beyond a point, adding cache doesn't help. I think that point has already been reached.

And no. L2 cache is "full-speed" meaning it runs at CPU clock speed. Not "HT link speed". HT link actual clock-speed == NB clock speed.
Posted on Reply
#80
Troubled
btarunrWell then look back and compare Athlon X2 Windosor-2M (1 MB L2 cache / core) to Windsor-1M (512 KB L2 /core). You'll find that the difference isn't much, and that with a mere 100 MHz increase in clock speed, whatever extra performance the additional cache provides is made up for. The boost in performance is not the cache's job. At the end of the day, your cache isn't crunching numbers. When there's a fast interconnect between the system and the processor, beyond a point, adding cache doesn't help. I think that point has already been reached.

And no. L2 cache is "full-speed" meaning it runs at CPU clock speed. Not "HT link speed". HT link actual clock-speed == NB clock speed.
I think you are referring to the Athlon X2 6000+ 2mb L2 Windsor which is 90nm core, Vs the Athlon X2 6000+ 1mb L2 Brisbane which is a 65nm core. Now...we all know that the 65nm core is a much better core and with the 100Mhz boost in not only the core speed, but also the cache speed that is what made up for the difference. Neither one of these processors feature an L3 Cache.

Now, let's bring this up to today's processors with L3, which as you have already stated run at the speed of the Northbridge. My Northbridge runs at 2.3ghz...this means that my L3 Cache only runs at 2.3Ghz. My Core Speeds are at 3.6Ghz, which means that my L2 Caches are also running at 3.6Ghz which i considerably faster. So...as I have said, having more of a faster L2 Cache would be a bit better.

I am not saying that Cache is going to make all the difference in the world, there are many other factors that are involved with a good processor...but every little bit helps. 5% may not seem like a lot, but when you are looking to squeeze every little ounce of power out of your processor....5% helps a lot.
Posted on Reply
#81
Yukikaze
This is very interesting, but this does look like it is going to be quite late, unless it is very cheap (relatively speaking).

What is the ETA on the 6-core Nehalems, currently ? That thing is my next chip, unless I lose my job until then...
Posted on Reply
#82
btarunr
Editor & Senior Moderator
TroubledI think you are referring to the Athlon X2 6000+ 2mb L2 Windsor which is 90nm core, Vs the Athlon X2 6000+ 1mb L2 Brisbane which is a 65nm core. Now...we all know that the 65nm core is a much better core and with the 100Mhz boost in not only the core speed, but also the cache speed that is what made up for the difference. Neither one of these processors feature an L3 Cache.
No, Brisbane is an optical die shrink of Windsor-1M, no changes.
TroubledNow, let's bring this up to today's processors with L3, which as you have already stated run at the speed of the Northbridge. My Northbridge runs at 2.3ghz...this means that my L3 Cache only runs at 2.3Ghz. My Core Speeds are at 3.6Ghz, which means that my L2 Caches are also running at 3.6Ghz which i considerably faster. So...as I have said, having more of a faster L2 Cache would be a bit better.

I am not saying that Cache is going to make all the difference in the world, there are many other factors that are involved with a good processor...but every little bit helps. 5% may not seem like a lot, but when you are looking to squeeze every little ounce of power out of your processor....5% helps a lot.
Once again, you're getting into the trivial "more/faster cache is better" argument. It really isn't for this architecture. Faster L3 cache would mean higher CPU/CPU-NB VID to support that speed. TDP takes a hit. Again, the impact of L3 cache is limited. AMD felt that 6 MB, 48-way cache is good enough for this architecture.
Posted on Reply
#83
Troubled
btarunrNo, Brisbane is an optical die shrink of Windsor-1M, no changes.
The Athlon X2 6000+ came in Brisbane which was the 1M version, and the Windsor was the 2M version. The Cores were different on these processors.
btarunrOnce again, you're getting into the trivial "more/faster cache is better" argument. It really isn't for this architecture. Faster L3 cache would mean higher CPU/CPU-NB VID to support that speed. TDP takes a hit. Again, the impact of L3 cache is limited. AMD felt that 6 MB, 48-way cache is good enough for this architecture.
In case you haven't noticed, the Nahalem series is stomping the Deneb series. Maybe AMD doesn't have it "right" yet. As Intel prepares the 32nm Westmere AMD is still trying to perfect the 45nm Deneb. AMD is behind in the game. They need to do something to catch up and take the lead. There are several points to improve on, one of which could very well be in the cache...along with many other points of improvement such as the IMC which could support triple or quad channel, Chipsets which could operate more efficiently and several other factors which go into a high performance system.
Posted on Reply
#84
btarunr
Editor & Senior Moderator
TroubledThe Athlon X2 6000+ came in Brisbane which was the 1M version, and the Windsor was the 2M version. The Cores were different on these processors.
Right, and the Brisbane chip was clocked a mere 100 MHz higher than the Windsor 2M. So all it took is a 100 MHz clock speed increase. Brisbane is an optical shrink of Windsor 1M. Brisbane is not one bit faster than Windsor-1M at constant clock-speed, architecturally, or in any other way.
TroubledIn case you haven't noticed, the Nahalem series is stomping the Deneb series. Maybe AMD doesn't have it "right" yet. As Intel prepares the 32nm Westmere AMD is still trying to perfect the 45nm Deneb. AMD is behind in the game. They need to do something to catch up and take the lead. There are several points to improve on, one of which could very well be in the cache...along with many other points of improvement such as the IMC which could support triple or quad channel, Chipsets which could operate more efficiently and several other factors which go into a high performance system.
And the answer to Nehalem is not bumping up cache/interconnect (your original argument). My educated guess is that turning that L3 cache into 12 MB, 1 MB L2 cache per core, and bumping HyperTransport speed isn't going to take Deneb any closer to Nehalem. It's only going to up transistor counts and die size by 2 times, and some 200W TDP.
Posted on Reply
#85
Troubled
btarunrRight, and the Brisbane chip was clocked a mere 100 MHz higher than the Windsor 2M. So all it took is a 100 MHz clock speed increase. Brisbane is an optical shrink of Windsor 1M. It's not one bit faster than Windsor-1M at constant clock-speed.
Based on that theory, the Kuma 7850 is slower than the Athlon X2 6000+...yet outperforms the 6000+ when overclocked to 3.0ghz (by simply bumping the CPU multiplier up by 1.)
Posted on Reply
#86
btarunr
Editor & Senior Moderator
TroubledBased on that theory, the Kuma 7850 is slower than the Athlon X2 6000+...yet outperforms the 6000+ when overclocked to 3.0ghz (by simply bumping the CPU multiplier up by 1.)
Right, so you see, that 2 MB L3 cache and 3600 MT/s HyperTransport (on Kuma) is all the more useless now. It's on par with Windsor 2M clock for clock.
Posted on Reply
#87
Troubled
btarunrRight, so you see, that 2 MB L3 cache and 3600 MT/s HyperTransport (on Kuma) is all the more useless now. It's on par with Windsor 2M clock for clock.
Which explains exactly why the 7850 @ Stock 2.8Ghz outperforms the 6000+ (even overclocked to 3.4Ghz) in every benchmark. More important to note is the Phenom II X2 550 which is clocked at 3.1 also outperforms the 6000+ in every benchmark. So where are you getting that Cache doesn't matter?
Posted on Reply
#88
btarunr
Editor & Senior Moderator
TroubledSo where are you getting that Cache doesn't matter?
From AMD stating that its 200% increase in L3 cache provides 5% performance uplift. In context of X2 550, you'll also see how a 5~8% overclock for an Athlon II X2 250 (that lacks L3 cache) places it on par with Phenom II X2 550. So really, the L3 cache while helpful, isn't something that you can keep increasing to get big performance gains beyond a point.
Posted on Reply
#89
Troubled
btarunrFrom AMD stating that its 200% increase in L3 cache provides 5% performance uplift. In context of X2 550, you'll also see how a 5~8% overclock for an Athlon II X2 250 (that lacks L3 cache) places it on par with Phenom II X2 550. So really, the L3 cache while helpful, isn't something that you can keep increasing to get big performance gains beyond a point.
Not sure where you got that information. But an Athlon II 250 Overclocked to 3.9Ghz barely meets par with the Phenom II 550 @ 3.1ghz. Which last time i checked is a 30% overclock which would require a liquid cooling system. Evened up clocks...the Phenom II 550 outperforms the Athlon II 250 with ease. Now the Athlon II 250 is however on par with the Athlon X2 7850 at the same clock rate.

I will give you this much. At higher overclocks (approaching 4Ghz), the L3 cache doesn't make as much of a difference in such tasks as Video Encoding, raring and unraring, etc. But when it comes to Gaming performance, it seems to make a rather good size difference. 10fps may not seem like a big deal...but it could be the difference in a playable game and a paperweight in a box.

Now if you will notice at my first suggestion...it was to bump up the L2 a little. not the L3.
Posted on Reply
#90
btarunr
Editor & Senior Moderator
TroubledNot sure where you got that information. But an Athlon II 250 Overclocked to 3.9Ghz barely meets par with the Phenom II 550 @ 3.1ghz. Which last time i checked is a 30% overclock which would require a liquid cooling system. Evened up clocks...the Phenom II 550 outperforms the Athlon II 250 with ease. Now the Athlon II 250 is however on par with the Athlon X2 7850 at the same clock rate.
That's simply not true, I won't spend time on that.
TroubledI will give you this much. At higher overclocks (approaching 4Ghz), the L3 cache doesn't make as much of a difference in such tasks as Video Encoding, raring and unraring, etc. But when it comes to Gaming performance, it seems to make a rather good size difference. 10fps may not seem like a big deal...but it could be the difference in a playable game and a paperweight in a box.

Now if you will notice at my first suggestion...it was to bump up the L2 a little. not the L3.
Go back to posts revolving around the 512 KB vs. 1 MB L2 cache. Won't repeat again.

I'll conclude saying that more cache isn't going to help AMD's cause. It doesn't provide the kind of performance that makes upping transistor counts by 100s of millions of transistors worth it, or feasible. As it stands, AMD Deneb has 758M transistors vs. a lesser transistor count of 731M on Intel Bloomfield, and Bloomfield emerges the superior core. Development is due on different fronts than cache.
Posted on Reply
#91
Wile E
Power User
TroubledI think you are referring to the Athlon X2 6000+ 2mb L2 Windsor which is 90nm core, Vs the Athlon X2 6000+ 1mb L2 Brisbane which is a 65nm core. Now...we all know that the 65nm core is a much better core and with the 100Mhz boost in not only the core speed, but also the cache speed that is what made up for the difference.
Actually, the Windsor is faster clock for clock than Brisbane. Brisbane's cache runs with a higher latency than Windsor. All you need to do is look thru the AMD OCing threads on this site from that time to see the F3 Windsors walked on ALL brisbanes at the same clocks, not to mention clocked as high as Brisbane in most cases. 65nm was a bad process all around for AMD. Leak problems not to mention the slower cache. That was true for Phenom I as well.

I went thru 4 brisbanes, all of them were ass compared to my F3 Windsors.
Posted on Reply
#92
Yukikaze
Cache increases have extremely diminishing returns as the cache size goes up. I can find you some slides I have on a study conducted comparing different L2 cache sizes and performance, and doubling the cache size only nets you a few scant percent of performance beyond a certain point (Somewhere around the 6-7Mb mark) while driving the cost of the chip and the TDP higher and higher due to the transistor count going up and up.

As for increasing L2: The Core 2 Quads have insane L2 cache sizes of up to 12Mb, while the i7 only has a small 4x256Kb L2 cache (And of course, the 8Mb L3 cache). The Phenom II X4 has 4x512Kb of L2 cache (And the 6Mb L3). Out of the three architectures, the i7 seems to be the best one. Based on this, I highly doubt that increasing the L2 cache on the Phenom IIs is really going to do very much, except for making the CPU more power-demanding and more expensive.

AMD isn't in the position of making their products more expensive, especially if it offers no real gain.
Posted on Reply
#93
swaaye
Also this CPU has the same amount of L3 cache as an X4. That means that it will be somewhat less efficient than the X4 CPUs.
Posted on Reply
Add your own comment
Apr 30th, 2024 06:19 EDT change timezone

New Forum Posts

Popular Reviews

Controversial News Posts