Hector Ruiz: K10 will put an end to the performance war
There will be no more tug of war . K10 will decisively finish Intel once and for all. Rahul Sood impressed by Israeli craft? Get ready for the AMD masters' true engineering skills.
HECTOR RUIZ: "In spite of all the hype and hoopla, there is no really such thing that Intel has leapfrogged AMD. It's quite the contrary. As a matter of fact, despite all the perceptions of Intel closing the gap, half the time they do a little bit better and the other half we do. And all of that will end with the introduction of Barcelona because it's such a significant jump in performance and quality."
K10 will be released by the end of 2Q07.
posted by Sharikou, Ph. D. @ 2/21/2007 04:27:00 PM
29 Comments:
http://www.edn.com/article/CA6418172.html
http://www.theinquirer.net/default.aspx?article=37775
AMD going BK.. LOL
K10 will be obsolete a quarter after its introduced as 45nm Penrym makes it road kill then Nehalem comes in 2008 and flattenst the road kill.
This comment has been removed by the author.
- Intel will launch a Clovertown LV (*QUAD* core / 65nm) at *50W* this quarter.
- AMD will launch an Athlon 3500+ EE (*SINGLE* core / 65nm) at *45W* this quarter.
- ATI R600 launch has been delayed again. The original release was planned in November last year already.
- "We see Michael Dell's return as CEO at Dell as a negative development with strong ties to Intel."
- "AMD faces a potential cash crisis amid rumors about a private-equity buyout for the microprocesor maker."
- AMD (NYSE) Last Trade : 14.56$
If K10 is late, too slow or if Intel launch is Penryn on time... AMD will then finally and for ever disappear from the surface of the globe.
If that happens, be ready to pay more, much much more, for your next CPU and your next GPU.
AMD is pretty much finished. All their products are vaporware. No one has seen R600 or K10 benchmarks.
Conroe frags AMD now. Penryn will frag K10. Nehalem will finish AMD once and for all.
K10 is just two quad core K8 with some L3 cache.
Considering nobody has seen either Penryn or K10 performance yet it would be a very brave man to bet either way.
Interestingly it is AMD doing all the talking now and Intel are quieter, whereas before Conroe it was the other way around, I think this shows who is plaing catchup and wh the pressure is on currently.
Nice to know that K10 is mid year and Penryn is 2007, I was worried a while ago both would be pushed back towards 2008.
lex -> AMD going BK.. LOL
This is not funny !
Morron -> GM, Ford and Chrysler going BK.. LOL
This is not funny either. :(
Don't you understand that we NEED AMD ?
In fact, Intel need AMD too !
Without AMD, Intel is likely to face antitrust charges.
Giant -> K10 is just two quad core K8 with some L3 cache.
No, K10 is more than that, much more.
Cenuijmu -> Considering nobody has seen either Penryn or K10 performance yet it would be a very brave man to bet either way.
Penryn is only a die shrink.
Result: Same perf at same MHz.
K10 is not only a die shrink.
Result: +40% avg, +70% FP.
(K10 vs Opteron at same MHz)
Am I brave enough ? ;)
wirmish
"lex -> AMD going BK.. LOL
This is not funny !"
It isn't and it isn't true either, neither will BK in any forseeable future.
wirmish
"Morron -> GM, Ford and Chrysler going BK.. LOL
This is not funny either. :("
That isn't funny but it might be true. Have you seen their marketshare shrinking when Japanese and EU cars are moving in?
wirmish
"No, K10 is more than that, much more."
Yes, it is. It is about as big enchancment as was C2D compared to Pentium M
wirmish
"Penryn is only a die shrink.
Result: Same perf at same MHz."
Besides dieshrink it'll also have a bit more cache and SSE4 instruction set. That'll improve performance in quite a few multimedia applications. Ray tracing will get quite a significant boost thanks to single cycle dot product. Barcelona will have SSE4 too.
wirmish
"K10 is not only a die shrink.
Result: +40% avg, +70% FP.
(K10 vs Opteron at same MHz)"
Could you describe what would give it such a boost? It mostly has the features C2 has and it doesn't add things that would boost IPC that much. I'd say it'll be as fast as C2 clock-to-clock. Depending on workload one would win the other, there won't be a clear winner. In absolute high-end, the one with higher clock speed will win.
Of cource if that is comparing highest performing dualcore vs highest performing quadcore then those numbers might be true to some extent. Also they might be valid when comparing performance per watt. Without benchmarks we won't know what exactly those numbers mean. For some reason AMD hasn't clearly said what they mean by these numbers.
AMD has not put nearly enough cache on their CPUs. All of the supposed performance gain comes from having additional cache.
Intel will release 50W Quad core parts this quater. That is eight processing cores in 100W of power. AMD's fastest dual core parts use that much. AMD
R600 is still vaporware and when launched will be a good old paper launch. Nvidia has the goods now. It only takes one G80 GPU to crush X1950 Crossfire. G80 SLI is unstoppable.
Intel has a clear lead on all fronts now; Mobile, Desktop and Server. Penryn is coming this year. Nehalem is coming next year. This will be enough to crush AMD once and for all.
RICHARD: Linux has always been a stronghold for us.
But it has not been one for ATI. I hope they'll release enough information to make an open source driver for their RS482 (the closed source one crashes).
ihttp://forums.vr-zone.com/showthread.php?t=130290
Sharikou a present fer ya.
Ho Ho
"Could you describe what would give it such a boost? It mostly has the features C2 has and it doesn't add things that would boost IPC that much."
Simple math.
Go there : http://freespace.virgin.net/roy.longbottom/cpuspeed.htm
Look at the tables.
Compare P3 vs P-M vs C2D vs A64.
See the differences between the P-M and the C2D ?
K10 will have the same differences vs K8.
The K10 double : FPU, FP, SSE, BTB, prefetch, HT speed, ..., and add : SSE4a, OoO, L3 cache, ...
So +40% avg & +70% FP perf for the K10 (vs K8) is not out of reach.
giant
"All of the supposed performance gain comes from having additional cache"
Care to explain why is there non-existent performance difference between 2M and 4M L2 Core2Duos at same clock speed? Same question about 1M and 2M L2 Prescotts, 256k, 512k and 1M K8's.
Sharikou said...
There will be no more tug of war. K10 will decisively finish Intel once and for all. Rahul Sood impressed by Israeli craft? Get ready for the AMD masters' true engineering skills.
I agree completely. AMD's engineers have already crushed Intel's engineer to the point where they outsourced Israeli engineers. Israeli engineers cannot compete with American engineers for long.
The Israeli engineers need to focus on their core competencies: Making bombs and killing Palestinians. Making exploding CPUs is not a long term profitable market.
Sharokou is awfully quiet since the last earnings report.
Sharikou quiet..
What no more BK statements after that last earning annoucement...Opps AMD sucking red.. LOL
What no more flaming laptops? I guess once all the batterys got recalled funny how that stopped.
What no benchmarks yet on K10.. funny you are a quarter away and nothing. What is AMD afraid of? INTEL got a K9 dog catcher waiting in the wings? Oh I guess they do. Penrym on 45nm pluss clear head room on the architecture to the sky and Nehalem on 2008...
Sharikou is eating a lot of crow these days..
I'd like if we could you (sharikou) to -not- post some of the more annoying comments, like those from the r tard Giant who keeps throwing words around without the slightest clue what the f he's talking about, or what the words means.
really sharikou.. no reason to make the reading irritating on purpose so please just ... don't post them ><
Here's some interesting information
http://www.hkepc.com/bbs/news.php?tid=746541&starttime=1170259200&endtime=1172678400
Sharikou: K10 will decisively finish Intel once and for all
But Hector said K10 ramp only starts Q2'08. And I remember you kept saying Intel BK in Q2'08. Not that most of us took you seriously, but don't you think you owe your disciples (i.e., penix, etc) some explanation. Of course as soon as you're done eating crow.
K numbers i think are an obsileet nomlicature but enthusists will still use them. This is great amd creameing intel. INTEL HAS ONLY A TEN% LEAD IN IPC SO AMD CAN EASELY CREAM INTEL.
----
May you please join MadModMikes Form Rubyworks.net
hmmm,
I do not think Hector Ruiz said it like that.
The majority of the gains in performance with Barcelona are by adding two more cores so it's a quad core and adding extra cache. The only other thing noteworthy is 128bit SSE. This was probably reversed engineered from Intel since AMD does not have real skills to develop CPUs.
It seems R600 has been delayed again to Q2. I predict that by the time it is out they will be facing G90, the Geforce 9.
Intel announced a new 50W LV quad core CPU coming out this quater. 3Ghz Quad Core CPU with 1600mhz FSB coming as well to greet AMD. The canelake server platform for 4P servers and above coming in Q3'07 to take away all of AMD's server revenues. Then 45nm Xeons in Q4'07 for AMD. Then Nehalem next year. Quad frags for AMD!
"Intel announced a new 50W LV quad core CPU coming out this quater."
It's running as slow as 1.6GHz, with uncertain price (i.e., uncertain demain/availability). It seems just the tail of fabrication distribution. I doubt it'd have a big impact on the market.
"3Ghz Quad Core CPU with 1600mhz FSB coming as well to greet AMD."
Which shows that Intel knows current 1066Mhz FSB is not enough for quad core. Stop saying otherwise, fanboys!
"The canelake server platform for 4P servers and above coming in Q3'07 to take away all of AMD's server revenues."
For server and workstation applications, Core 2 simply doesn't have as much lead over K8. This is especially so when number of processors increase. Intel's 4P is going from zero to become available - which doesn't make frag-ability at all.
"Then 45nm Xeons in Q4'07 for AMD. Then Nehalem next year. Quad frags for AMD!"
That would be good if 45nm brings lower production cost so that Intel can lower price further to press on AMD. On performance side, Xeon @45nm simply won't have much.
OTOH, Nehalem seems to be the first thing worth buying in a long string of Intel's recent releases. It finally does away the ancient FSB, even though it could've used a more industry-standard hypertransport. But a Datacenter upgraded to Core 2 Xeon in late 2006 would again find a total change of platform 1.5 years later. No wonder Intel's been quiet about the supposedly big release/change in 2008.
abstein
"I doubt it'd have a big impact on the market."
Being able to cram around 768 cores into one rack is good enough reason for some people. Having half that with slightly faster clock speed is not nearly as good.
abstein
"Which shows that Intel knows current 1066Mhz FSB is not enough for quad core. Stop saying otherwise, fanboys!"
It is good enough for most purpouses but having a faster FSB always helps. If with nothing else then by improving latency.
"Being able to cram around 768 cores into one rack is good enough reason for some people. Having half that with slightly faster clock speed is not nearly as good."
No, you were not listening.
It doesn't matter if someone's putting them in a datacenter or a supercomputer. The point is those 50W chips are slow, and they won't make much impact on the overall market. As I said, their availability, pricing, and demand are all uncertain, that's because these chips are the tail of the manufacturing normal distribution.
Besides, I don't think it's plausible to use those 50W in high-density servers. First, they use FB-DIMM which are quite power hungry themselves, almost as much as the processors (50W) in this case. Second, to get good speedup on multi-core, beefy northbridge chipsets are required. Third, what is the upgrade path for such a server/datacenter? CSI is coming out soon and will obsolete the current Intel platform. Unless there's a roadmap showing Core 2 quad 50W passing 2GHz, it's really a waste to invest that much on the 50W 1.6GHz today.
"It is good enough for most purpouses but having a faster FSB always helps."
Yes, FSB is good enough for most notebook/desktop apps and certainly benchmarks. It is not good for multi-socket workstations that perform high throughput scientific calculations, or servers that demand high IO/memory bandwidth.
abinstein
"Besides, I don't think it's plausible to use those 50W in high-density servers. First, they use FB-DIMM which are quite power hungry themselves, almost as much as the processors (50W) in this case"
Again, get your "facts" straight. One stick of performance targeted FBD takes around 16W. You can use a bit slower FBD modules and get by with around 10W per slot. IIRC, DDR1 was 10W per stick also.
Now could you show your sources for FBD power usage. I got my data from here(PDF)
Also FBD+CPU power usage is still less than any other solution that gives you as much computing power.
abinstein
"Second, to get good speedup on multi-core, beefy northbridge chipsets are required."
For 2P per rack you don't need much.
abinstein
"Third, what is the upgrade path for such a server/datacenter?"
Its the best there currently is. Also going to those quadcores might be an upgrade to some older CPU's, perhaps some slow dualcore Xeons.
abinstein
"Unless there's a roadmap showing Core 2 quad 50W passing 2GHz, it's really a waste to invest that much on the 50W 1.6GHz today"
Why not? Is 25% CPU performance really that important?
abinstein
"It is not good for multi-socket workstations that perform high throughput scientific calculations, or servers that demand high IO/memory bandwidth"
But for everything else it is good enough, not everyone need lots of bandwidth. Mostly having lots of computing power is enough. One such thing might rendering. One rack like that would do the job of several racks using fraction of the power.
SpecFP is not a real-world usage scenario. If it is then why is Spec's Java benchmark dominated by FSB-using CPU's in 2P? Isn't it simulating a large business application?
Also, who wants to bet that we'll have 2.66GHz LV quadcores in at most 1.5 years?
"Hector Ruiz: K10 will put an end to the performance war."
because AMD can't compete anymore? LOL.
This comment has been removed by the author.
"Hector Ruiz: K10 will put an end to the performance war"
Because after K10, AMD will be BK.
Post a Comment
<< Home