K10 Scores starting to surface

[informal]

Based on what? Hope?

Hmm you are one intel only dude,aren't you?

Here you go,knock yourself out :
http://forums.anandtech.com/messagev...90639#27132633

Originally posted by: Viditor
There appears to be some problems in those reports...

1. It reports as an Opty 2332, but also as a Phenom/Agena (the 2332 is a socket 1207, the Agena is a socket AM2+)
2. It's only a stepping 1 chip...

The latest Barcelona chips are B02 steppings with one more to go. Believe me, the reason we did not post any numbers at Computex or since then is the simple fact that the CPU/boards/BIOS have undergone dramatic changes over the course of the summer. If you have an earlier stepping there is a very good chance that HT and the secondary cache is disabled, this will affect the benchmarks dramatically. We expect to see final stepping chips and board revisions early next week, until then, it is all speculation for the most part.

The one caveat that I will add, this chip really does not get into a groove until you get over 2.4GHz and then it scales incredibly well. Also, the first RD790 boards we have will undergo another spin so any Phenom results with those boards are subject to interpretation depending on whether you like AMD or not.

[Periander6]

I still say these results are not the representative of final scores.

I say they will be. Slide 16 here was a dead giveaway:

http://www.sunmicrosystems.se/virtua...f_Nordlund.pdf

That's not "up to 15% IPC gains" against C2D, it's 15% against K8.

Add to that keeping a tight lid on benchmarks when there was no reason to do so if they were good, the change in emphasis from performance to performance/watt, the Henri Richards resignation, low launch clockspeeds, the focus on Specfp_rate (which is really little more than a memory bandwidth bench), the rumors out of Computex, and the "simulated" benchmark scores.

Really, the writing has been on the wall for months, just like it was for the R600 and for many of the same reasons. The only thing keeping the hype alive were the AMD forum boosters and their relentless efforts to shout down anyone who pointed out inconvenient facts.

[[XC] gomeler]

I just have 2 hopes, these K10 results are somehow skewed a bit due to a buggy processor and that Opteron prices won't be through the roof so us normal people can buy one to play with.

[AliG]

as long as the actual phenom cpus won't be that buggy when they debut and produce better scores, I'll be happy, but the servers are going to be gipped, unless if the super pi and cenebench results don't display the true processing power of the k10 cpus

[informal]

I say they will be. Slide 16 here was a dead giveaway:

http://www.sunmicrosystems.se/virtua...f_Nordlund.pdf

That's not "up to 15% IPC gains" against C2D, it's 15% against K8.

Add to that keeping a tight lid on benchmarks when there was no reason to do so if they were good, the change in emphasis from performance to performance/watt, the Henri Richards resignation, low launch clockspeeds, the focus on Specfp_rate (which is really little more than a memory bandwidth bench), the rumors out of Computex, and the "simulated" benchmark scores.

Really, the writing has been on the wall for months, just like it was for the R600 and for many of the same reasons. The only thing keeping the hype alive were the AMD forum boosters and their relentless efforts to shout down anyone who pointed out inconvenient facts.

Read mu previous post..

[Carfax]

Haha Informal, I was just about to post that

[informal]

Beat you by 1 minute .Still you did post some very useful info

[kl0012]

Read mu previous post..

there is a very good chance that HT and the secondary cache is disabled

That is stupid explanation. Barcelona can't work with "disabled HT". And try to disable L2 - you will see slide show instead of your windows. Any way, you can see in CPU-Z that it wasn't disabled.

[railer]

Well if these SPI scores are real it will mean one thing that K10 will suck at real games. Most games are single threaded now and those 8 cores are useless in anything but in 3d06 which is really far from actual game performance. I'm really disappointed with K10. I hopped for a way better. AMD should build a good performing single CPU and then add those cores. And Collaer is a reputable source, 99% it wont be any different on the sept 10. Let hope in 2009 AMD strike back.
Intel pwn's right now in prices and performance

[w0mbat]

u compare superpi and games? lol!

[XeRo]

u compare superpi and games? lol!

Superpi is a game. It's just like to some people that 3dmark is a game.

[railer]

u compare superpi and games? lol!

Il give you a hint for every 1 sek decrease in PI my fps in FEAR increase 3%.
I don't compare its just an indication that at single threaded apps i don't seee much difference between K8 and K10.

[oldblue]

... Specfp_rate (which is really little more than a memory bandwidth bench)

I've seen this belief pop up a few other times on this forum, and it's a little strange. Specfp_rate tests how quickly a system runs 17 different real-world floating-point codes under full load. (Specint_rate is similar, but for integer codes). Of course it becomes a memory bandwidth benchmark if you don't have enough memory bandwidth. That's true about every benchmark that doesn't fit in the cache.

If you look at the results (www.spec.org), you'd see that systems with the same memory bandwidth but faster processors tend to do better.

It's probably better to think of it as a scalability benchmark. If your system scales well, you want to show off specfp_rate and specint_rate. If it doesn't, you want to stick with single-threaded benchmarks.

[PetNorth]

As kl0012 says, I don't think HT or cache are disabled in these B0/B1 stepings. But I think that the new (or part of them) features are disabled or something.

I say this because when I see to these results, I conclude that they are like if it would be a K8 (for example, I've run Cinebench 10 x64 with my X2 at 2.0 and 1cpu gives me 1905 -this K10, 1896-. Equal result. The same thing I can say about those pov-ray benchs). And with new K10 features (like 128 SSE etc) this isn't possible at all.

So certainly, I think that Gary Key form anadtech, is on the way, but he fails in the reasons. Of course no HT or Cache disabled (without cache, it would be an absolute turtle), but like I said I think new K10 features (or part of them) are disabled in these steppings and thats the reason why we don't see basically improvements in relation to K8, in heavy optimized SSE apps, like Cinebench.

IMHO, we'll have to wait till commercial step (it seems to be B2 or BA) to know how this thing really perform.

[awdrifter]

Damn, I might as well build a quadcore now, I was expecting K10 to be at least as fast as Kenfield, but I guess this is another R600. If these benches are true, AMD might as well just keep the K8 arch and try to get it to 3.5+ghz on air. It would yield about the same performance as the 3ghz K10.

[Mad_Man]

talking about "disabled HT"... noone noticed that in every cpu-z it says bus speed 200mhz, rated fsb speed 200mhz ? should not HT link be much, much more? at 200mhz it will be a real bottleneck.
just a though...

[Sparky]

Bus speed is the HTT. HT is different than the bus speed/FSB/HTT.

[n91htmare]

I just don't think it's an early sample this late in the game .

IF anything, I think we can attribute the low performance to the mobo or something.

[duploxxx]

Well if these SPI scores are real it will mean one thing that K10 will suck at real games. Most games are single threaded now and those 8 cores are useless in anything but in 3d06 which is really far from actual game performance. I'm really disappointed with K10. I hopped for a way better. AMD should build a good performing single CPU and then add those cores. And Collaer is a reputable source, 99% it wont be any different on the sept 10. Let hope in 2009 AMD strike back.
Intel pwn's right now in prices and performance

i think its about time school is going to start again.... so you guys can learn something.

the benches are from a 2 socket opteron socket and as additional info the memory is running 667 speeds cas5. what superpi perfromance do you expect.
the performance can't be compared with the phenom. try to bench a k8 with memory 667 cas5 and then change it to 1066 cas5 like the demo system AMD is showing. even at the same speed it will be a huge performance difference.

looking and comparing the cpu-z from Coolaler to mine it is a fact that his sample is a dvt or early bird from oem. so it will be the chip that launches within a few days. All features are enabled, K10 is no INT miracle it will be as fast as core and no penryn is not faster then conroe, it's just when the program is affected by cache or not it will be faster clock/clock and the magic sse4. k10 will increase the gap on fpu and IO even more not to mention virtualization. How well it scales will be depending on the new stepping coming in oktober and you'll see how well it does when people get phenom es in there hands.

informal's information about the bios is correct, it is changing alot, systems with b0 had no CPU options in BIOS available except changing HT speed.

[Flambo]

If this score is truly representative of K10, then AMD must be going backward because even my 2.67 GHz 3800+ X2 can do SuperPi in 32 seconds! Of course,my 3.124 GHz Opteron 175 can do it in about 28 seconds. I cannot really believe that this will be what the shipping chips will be like.

[Mad_Man]

If this score is truly representative of K10, then AMD must be going backward because even my 2.67 GHz 3800+ X2 can do SuperPi in 32 seconds! Of course,my 3.124 GHz Opteron 175 can do it in about 28 seconds. I cannot really believe that this will be what the shipping chips will be like.

and my windsor at 3.17Ghz can do it in 24.8s
whats the purpose of your post?
OCed cpu&ram against stock cpu with ddr2 @667 cl5 ....

sPi aint real world...

[terrace215]

and no penryn is not faster then conroe, it's just when the program is affected by cache or not it will be faster clock/clock and the magic sse4.

You're forgetting some other Penryn improvements. Take a look at throughput and latency improvements to various SSEx operations, for instance. (Super-shuffle engine)

But in most apps, most of what Penryn brings will be higher speeds & lower power. Assuming AMD provides any pressure at all for the higher-speeds part...

In terms of a 2-socket comparison, one near-term boost for Clovertown (and Yorkfield) will be the transition from Blackford --> Stoakley. Much better snoop-filter & throughput up 25%... should give the spec_rate benchmarks and performance-related apps a nice boost.

[savantu]

I say they will be. Slide 16 here was a dead giveaway:

http://www.sunmicrosystems.se/virtua...f_Nordlund.pdf

That's not "up to 15% IPC gains" against C2D, it's 15% against K8.

Add to that keeping a tight lid on benchmarks when there was no reason to do so if they were good, the change in emphasis from performance to performance/watt, the Henri Richards resignation, low launch clockspeeds, the focus on Specfp_rate (which is really little more than a memory bandwidth bench), the rumors out of Computex, and the "simulated" benchmark scores.

Really, the writing has been on the wall for months, just like it was for the R600 and for many of the same reasons. The only thing keeping the hype alive were the AMD forum boosters and their relentless efforts to shout down anyone who pointed out inconvenient facts.

Excellent post .

[[XC] flat-four]

Bus speed is the HTT. HT is different than the bus speed/FSB/HTT.

Shouldn't CPUZ show "HT Link" link instead of "Rated FSB" like usual AMD cpus? If the HT link is actually running at 200 instead of 1000+, this would be a huge bottleneck as Mad Man said.

[informal]

Isn't it very convinient this info comes from the intel exclusive place(coolaler forums),and that it is coming only a day after Theo's post??It could be that Theo's results were in some way inflated(since the score is really hard to believe),but this avalanche of "poor K10 results" comes after he wrote this:

This statement warrants at least three hatemails from Intel's R&D lads, but all that we will disclose here are results we have in our possession