More Deneb 45nm Benchmarks

The person that posted the Deneb Super Pi scores on the Chinese forum a few weeks ago has posted some more Deneb benchmarks: Cinebench; 3DMark Vantage; Fritz Chess; Sandra; Winrar; & Everest. Unlike the Super Pi results, which showed Deneb performing about 12% better than Agena, these latest tests show basically no improvement over Agena, with some tests actually showing markedly lower performance.

The tests were conducted at 2.3GHz & 3.27GHz

Cinebench

http://www.itocp.com/html/amd45nm/R10.jpg

By comparison, my Phenom 9850 @ 2.3GHz scores 2012 & 7330.

Fritz Chess

http://www.itocp.com/html/amd45nm/frz.jpg

By comparison, my Phenom @ 2.3GHz scores 5462

3DMark Vantage

http://www.itocp.com/html/amd45nm/3dmarks.jpg

I don't use Vista, but looking up Phenom scores on Futuremark's site shows that the 3.27GHz Deneb CPU score is about equal to a 2.9GHz Agena score.

Winrar

http://www.itocp.com/html/amd45nm/winrar.jpg

The poster mentions something about a "bug" in his comments on this test.

Everest

http://www.itocp.com/html/amd45nm/everest.jpg

Sandra

http://www.itocp.com/html/amd45nm/siss.jpg

Strange results for the most part, to say the least.

Source:

http://www.itocp.com/thread-12164-1-1.html

seems like the mem controller is still unfinished, but damn the 3rd level cache is fast :eek:

Interesting. Keep in mind this is a first revision chip so it most certainly is going o have bugs. I wouldn't expect too much from Deneb in terms of IPC though.

This is the first stepping out of the fab. I wouldn't be too concerned about it not performing well. Clearly there is a bug with the chip given the subpar memory performance.

lol, weird:ROTF:.

Prolly some bugs though, but the tester typed bug already before he started any bench as well:confused:. And also, did you do your tests with 1.8Ghz NB btw? Because Deneb will run 1.8Ghz NB but the Agena BE's run it at 2Ghz. Not much but it might matter.

Hornet331, fast!?
The L3 cache on the 9850 BE is way faster than that:

http://img403.imageshack.us/img403/4...ststockyh9.gif

and this surprises you guys?!?!? this IS amd we are talking about, the only thing they do good is gpu's so eveyone holding their breath for a goodamd cpu may as well exhale now ( not trying to troll, but c'mon seriously)

Quote:

---

Originally Posted by philbrown23

and this surprises you guys?!?!? this IS amd we are talking about, the only thing they do good is gpu's so eveyone holding their breath for a goodamd cpu may as well exhale now ( not trying to troll, but c'mon seriously)

---

that's nonsense. you're not trying to troll, but actually it's exactly what you're doing ;)

ont: even if this batch has still bugs the scores are extremely low compared to current phenoms.
even though, it's pointless to bench such an early ES if bugs are causing these low performance...

Quote:

---

Originally Posted by philbrown23

and this surprises you guys?!?!? this IS amd we are talking about, the only thing they do good is gpu's so eveyone holding their breath for a goodamd cpu may as well exhale now ( not trying to troll, but c'mon seriously)

---

WTF, get a clue.

Quote:

---

Originally Posted by doompc

Hornet331, fast!?
The L3 cache on the 9850 BE is way faster than that:

---

deneb has 200% more cache, so i call this fast. :yepp:

Quote:

---

Originally Posted by Hornet331

deneb has 200% more cache, so i call this fast. :yepp:

---

It's relatively fast yes. However if you see the RAM bandwith as well, I think there's an issue with the NB. A bit early to comment on it, thus far it seems to OC quite well though. I hope they get the little bugs/flaws fixed soon though.

I dont want another TLB fiasco, nor do I think AMD can actually afford to have that.

Quote:

---

Originally Posted by doompc

Hornet331, fast!?
The L3 cache on the 9850 BE is way faster than that:

http://img403.imageshack.us/img403/4...ststockyh9.gif

---

Looking at what a C2D can do those numbers are really :(

Looking at winrar and Everest numbers, TLB patch is applied.

My guess: due to it is a very early stepping, it suffers TLB bug too.

Quote:

---

Originally Posted by Hornet331

deneb has 200% more cache, so i call this fast. :yepp:

---

It's not fast at all, no matter how much cache is onboard. Agena's cache was already very slow. Nehalem has 8MB of L3 and its cache is a good 50% faster than the L3 of the 9850/9950, and that's for the slowest model (2.66GHz).

There is clearly something wrong with the NB of C0 Deneb. Normally there are going to be performance problems with the first revision of any chip.

Quote:

---

Originally Posted by PetNorth

Looking at winrar and Everest numbers, TLB patch is applied.

My guess: due to it is a very early stepping, it suffers TLB bug too.

---

QFT.



that is pointless benchmarking right there.

Quote:

---

Originally Posted by philbrown23

and this surprises you guys?!?!? this IS amd we are talking about, the only thing they do good is gpu's so eveyone holding their breath for a goodamd cpu may as well exhale now ( not trying to troll, but c'mon seriously)

---

So was your method the same back when AMD was on top and Intel could not compete? Were we just saposed to say "Oh well, we can never expect a good CPU from Intel again because well, its just Intel".

Good thing your not trolling:ROTF: http://i303.photobucket.com/albums/n...images_bmp.jpg

Quote:

---

Originally Posted by PetNorth

Looking at winrar and Everest numbers, TLB patch is applied.

My guess: due to it is a very early stepping, it suffers TLB bug too.

---

Well, my guess it is just a rough engineering sample and there are bugs to squash and BIOS opts to do... the data is sketchy at best and probably should not be taken seriously -- at least I hope not.

But AMD would need to be utterly stupid to propagate the TLB bug into their shrink.... this is inconceivable.


EDIT: The more I look at the data, the more I am convinced that the early engineering revision simply needs some work either this is an early stepping and not what will launch or the BIOS still needs to be worked over to tweak up the IMC. I put no faith in these numbers.

Jack

Quote:

---

Originally Posted by philbrown23

and this surprises you guys?!?!? this IS amd we are talking about, the only thing they do good is gpu's so eveyone holding their breath for a goodamd cpu may as well exhale now ( not trying to troll, but c'mon seriously)

---

Well -- I don't think you will find too many people here that will not admit that AMD had a rough couple of years and did not execute well, but AMD has been stellar in the past ... it is in the best interest of anyone who enjoys comuting as a hobby that AMD brings out a strong product this next round.

Jack

It's pretty obvious that they're not going to release a chip that's actually worse than the original. Unless they're doing a Prescott, although, I kinda doubt them to be that stupid. ;)

Either it's due to the early chip, or maybe the BIOS/motherboard recognizes it as an early stepping Barcelona for some reason and applies the TLB patch? :confused:

Either way I sure hope this will be a sweet alternative to Nehalem, I probably won't buy it myself (never say never though) but every Shanghai sold will fund the development of Bulldozer. :yepp::up:

nehalem has 8 virtual cores, it's going to be tough to beat in any modern benchmark that is fully multithreaded.

guys I used to use amd even when they sucked for years, like I said I wasn't trying to troll, but let's face it, at this point in time gpu's are what they do best, and as long as they stick with the same cpu diagram (wich is essentially what they are doing to cut the cost of changing to new technology) they will continue to fail. Sure they have 45 nm but they just took a regular amd cpu and shoved a smaller die in it, but if the diagram wasn't working (enthusiast wise) before then it wont now. Sure intel just shrunk their dies but they did it on proven to work technology.

Quote:

---

Originally Posted by philbrown23

guys I used to use amd even when they sucked for years, like I said I wasn't trying to troll, but let's face it, at this point in time gpu's are what they do best, and as long as they stick with the same cpu diagram (wich is essentially what they are doing to cut the cost of changing to new technology) they will continue to fail. Sure they have 45 nm but they just took a regular amd cpu and shoved a smaller die in it, but if the diagram wasn't working (enthusiast wise) before then it wont now. Sure intel just shrunk their dies but they did it on proven to work technology.

---

Except that the same very thing was said for ATI. No one is expecting a 4800 series here, but a solid 3800 would be nice, making enough improvements on the architecture to make things viable again. Sure, we could just ignore AMD and assume they are dead, but why? There isn't anything wrong with being interested in what AMD or even hoping that something might actually surprise us.

i wouldnt say "doomed", but they certainly havent had a shortage of horse crap products thats for sure lol

Obviously this thread is worthless.
Pre-production glitches at play.

I think they already started to shrink Deneb before the TLB bug was made known, or fix. So it was technically done in late 07/very early 08.

Why don't someone disable it through the MSR paths?

I just noticed something strange about today's new Deneb benchmarks. Although I didn't post them, the Deneb benchmarker also ran Super Pi again, in addition to Cinebench, etc. I didn't post the Super Pi results because I assumed that they were consistent with the results from 3 weeks ago, which showed Deneb beating Agena by about 12%. That is what gave rise to the heightened expectations for Deneb in other tests, such as Cinebench, in the first place.

But today's Super Pi results are not consistent with those from 3 weeks ago. They are about 12% slower.

Today's Deneb Super Pi results (2.3GHz & 3.27GHz)
http://www.itocp.com/html/amd45nm/superpis.jpg

In comparison, my Phenom 9850 @ 2.3GHz scored 32.672, an almost identical score to Deneb.

Deneb Super Pi results from 3 weeks ago

2.8GHz
http://www.itocp.com/attachments/mon....jpg.thumb.jpg

3.0GHz
http://www.itocp.com/attachments/mon....jpg.thumb.jpg

In comparison my Phenom 9850 scored 28.101, and 26.406, @ 2.8GHz, and 3.0GHz, respectively, both of which scores are about 12% slower than the Deneb results at those two speeds from 3 weeks ago.

The actual Deneb chip tested appears to be the same in today's tests and those 3 weeks ago, and the same motherboard was used (MSI K9A2 Platinum), so I can only assume that something in the testing environment changed to cause the performance hit. Perhaps Cool & Quiet was enabled for today's tests, which Anandtech has shown can have a dramatic negative effect on Phenom performance, or maybe Vista SP1, which was used in today's tests and has been reported to force the TLB patch on, to the detriment of even B3 Agenas, is responsible.

Whatever the cause, the poor Deneb scores from today were not produced in the same operating environment as the one that produced the promising Super Pi scores 3 weeks ago.

Deneb Super Pi results from 3 weeks ago:

http://www.itocp.com/thread-11103-1-1.html###zoom

Can someone make a compare with a current Phenom but with TLB fix enabled?

Clueless reviewer....
Those old ones look much more accurate.
Let´s wait for final product.

Like Yokozuna already said,the original thread poster(reviewer) said this in his post at itocp:

Quote:

---

Test:

1，由于主板对新款处理器支持有问题，存在一部分bug影响部分软件性能测试，所以测试数据只供参考，并非 最终上市产品性能。 1, the board of the new processor support problems, there part of the software bug affected performance testing, test data for reference only, is not listed on the final product performance.

2，测试加入对CPU 超频测试，着重测试超频前后的处理器发热量。 2, testing the addition of the CPU overclocking tests, focus on testing before and after overclocking the processor heat.

---

One look at those synthetic(everest) tests and all is clear that this ES is crippled.Be it Vista SP1 "TLB auto on" feature,or a ES related "feature",results don't line up with the previous World Record in SPI1M,for example(it's the only result from the previous runs where the ES showed major improvement over Agena).

Also keep in mind that comparing your own aganea results vs Deneb, he has CL 5 sticks around which arent that great.

Quote:

---

Originally Posted by informal

Like Yokozuna already said,the original thread poster(reviewer) said this in his post at itocp:


One look at those synthetic(everest) tests and all is clear that this ES is crippled.Be it Vista SP1 "TLB auto on" feature,or a ES related "feature",results don't line up with the previous World Record in SPI1M,for example(it's the only result from the previous runs where the ES showed major improvement over Agena).

---

yes, im read preview test with translator from chinese to czech :). I was happy to corectly test and ...:mad:. One good finding from preview is good temperatures! We must waiting on final sample product:cool:

Quote:

---

Originally Posted by R101

Can someone make a compare with a current Phenom but with TLB fix enabled?

---

What comparision do you actually want? Clock for clock? I cannot hit those clocks with a 'TLB fixed' phenom 9600 BE, my 9600 can only do 2.7 GHz for short periods.

SP1M no patch @ 2.3 Ghz = 36.312
SP1M with patch @ 2.3 GHz = 38.203

I.e. about a 5% hit.

Looks like some people like to make fools out of them selves:ROTF:.

Anyhow, does anyone actually know what cooling is used because I looked at the temps and found them to be damn good. But I also saw the temp sensor on the CPU and lots of paste so I thought he was going on good watercooling or even TEC watercooling:confused:.

But if this is the same Deneb as a few weeks before, something went really wrong indeed. But if you look at the different results, the newer one is in Vista and the older one in XP. So an eventual TLB fix might indeed be applied in the newer one.

nice cpu with promising performance. a nice step forward by the amd. i think this is not enough to beat a nehalem but what i see is amd is doing the right moves maybe not this step but later ones can make another jump like athlon64

Quote:

---

Originally Posted by Yokozuna

But today's Super Pi results are not consistent with those from 3 weeks ago. They are about 12% slower.

---

It isn’t healthy to believe all you read on the internet.

This right here is a prime example of people not knowing what they're doing when they benchmark. Most likely running Vista Sp1 32bit. Which means the TLB fix that was written into Sp1 is also effecting revision Cx cores as well. Great job on that cluster :banana::banana::banana::banana: MS. On the other hand, with Official SP3 on WinXp my performance has actually gone up on my 9850. Hitting right around 2,000KB/sec on WinRar at stock, which is about 100KB faster than previous. Other guess would likely be a beta bios on the board. Probably still using a pre hardware release bios, like I've seen a lot of the chinese sites do. Definitely TLB fix goin on somewhere though. I remember what mem and L3 latency numbers looked like on my 9600be with the TLB fix, and those look close to the same.

Quote:

---

Originally Posted by informal

Like Yokozuna already said,the original thread poster(reviewer) said this in his post at itocp:


One look at those synthetic(everest) tests and all is clear that this ES is crippled.Be it Vista SP1 "TLB auto on" feature,or a ES related "feature",results don't line up with the previous World Record in SPI1M,for example(it's the only result from the previous runs where the ES showed major improvement over Agena).

---

Proper translation should be kind of like...

1) Motherboard used for testing may have problems, or softwares used for testing might contain bugs. Benchmark data is only for reference, and does not reflect the actual performance of the final retail product.

2) In the overclocking tests, we focused on taking the temperatures of the processor with and without being overclocked.

There is no verdict on this, so maybe it's the TLB auto on feature.

Eh? Like now it looks like crap but there is a high chance it will improve with the new 45nm process etc.

Not much to say beyond that.

Perkam

Maybe AMD is using their old technique where they didn’t inform the full speed of new processors. They were always keeping a low profile. The ATI 48xx was one example where this worked. When they have informing or bragging about new products the reaction hasn’t been good.
Also it may be cheaper to skip the marketing ;)

Quote:

---

Originally Posted by gosh

Maybe AMD is using their old technique where they didn’t inform the full speed of new processors. They were always keeping a low profile. The ATI 48xx was one example where this worked. When they have informing or bragging about new products the reaction hasn’t been good.
Also it may be cheaper to skip the marketing ;)

---

It's possibly, but I don't believe it's likely. The reason this worked so well for the HD4k series was because the most money in that market is from the midrange, so aiming to best cater to that market made sense. However, on the CPU side, the general perception most people have is that Phenom isn't competitive. If AMD can't change this image, they're going to be limited to selling low margin budget CPUs for the forseeable future (as far as desktop sales go).

If they had something that was capable of trumping Nehalem, I think they'd be pretty vocal about it right now. But then again, you never know since they could be a little wary considering what happened with the Phenom launch.

Quote:

---

Originally Posted by resar

If they had something that was capable of trumping Nehalem

---

And if Nehalem isn’t that good that everybody thinks it is? When Intel is going to show the advantages of Nehalem they need to use same type of software that runs very well on Phenom/Deneb. Maybe Nehalem will perform about 10% better or something like that. But the price is much higher. Then you have a similar situation like ATI vs nVidia. The 4870 isn’t quite as good as GTX 280 but it is much cheaper. And it is selling much better.

Also if Deneb will OC to more than 4 GHz but Nehalem isn't going to OC that much, then Intel may have som problems

4 GHz ?
http://www.expreview.com/news/hard/2...8981d9645.html

The only thing nehalem will be doing for me when it releases is to lower the prices of yorkfields, I'd consider amd if they can match the q9xxx on what matters most for me (rendering and not benchmarks) and if denebs can oc decently.

Quote:

---

Originally Posted by gosh

And if Nehalem isn’t that good that everybody thinks it is? When Intel is going to show the advantages of Nehalem they need to use same type of software that runs very well on Phenom/Deneb.

---

I hadn't noticed that. Of all the benchmarks used in the Anandtech preview of Nehalem, Phenom was only competitive with Kentsfield on the x264 test.

Quote:

---

Maybe Nehalem will perform about 10% better or something like that. But the price is much higher. Then you have a similar situation like ATI vs nVidia. The 4870 isn’t quite as good as GTX 280 but it is much cheaper. And it is selling much better.

---

More likely what will happen will be repeat of the last two years, where the launch of the E6400 effectively eliminated AMD's market for >$200 processors and the July 2007 price cut of the Q6600 made Phenom dead on arrival. It's probable the sub $300 Nehalem will do the same.

Quote:

---

Also if Deneb will OC to more than 4 GHz but Nehalem isn't going to OC that much, then Intel may have som problems

---

It may take a 4GHz Deneb to compete with a low-end Nehalem. Besides there's always Yorkfield as backup, which has already shown itself to be a very impressive die shrink in terms of performance and power usage. AMD needs to do an even better die shrink just to be competitive with what Intel has now.

Quote:

---

Originally Posted by accord99

I hadn't noticed that. Of all the benchmarks used in the Anandtech preview of Nehalem, Phenom was only competitive with Kentsfield on the x264 test.

---

wait to real tests are out

Quote:

---

Originally Posted by accord99

I hadn't noticed that. Of all the benchmarks used in the Anandtech preview of Nehalem, Phenom was only competitive with Kentsfield on the x264 test.

---

Something interesting about Anand's Nehalem preview is that the types of tests he did match up near perfectly with this:

http://images.anandtech.com/reviews/...G/image002.gif

I'm thinking that the performance improvement is being inflated a bit in his preview because of HT, but I will reserve final judgement until the final chips and more thorough reviews are out.

Quote:

---

Originally Posted by resar

I'm thinking that the performance improvement is being inflated a bit in his preview because of HT, but I will reserve final judgement until the final chips and more thorough reviews are out.

---

Inflated? It's the processor perfomance, nothing more nothing less. I don't trust Anand numbers, just saying HT is a feature and should be included in the benchmarks as an enhancement to the already high core perfomance. What if DivX sees an improvement? When you have the CPU in your hands you'll use HT and you'll see that boost too, so it's fair to include in in the reviews.

Quote:

---

Originally Posted by STaRGaZeR

Inflated? It's the processor perfomance, nothing more nothing less. I don't trust Anand numbers, just saying HT is a feature and should be included in the benchmarks as an enhancement to the already high core perfomance. What if DivX sees an improvement? When you have the CPU in your hands you'll use HT and you'll see that boost too, so it's fair to include in in the reviews.

---

Perhaps it was a poor choice of words because I agree with what you're saying. It's entirely fair to include it in the review. What I meant was that beyond the applications that show a substantial benefit from HT, the improvement may not be nearly as much. So by selecting only benchmarks that benefit a lot from HT, the results can be spun to make it look like it's much more of an improvement than it actually is.

The TLB fix is on ~that's up to 30% or so decrease in performance.
The mem latency says it all.

Quote:

---

Originally Posted by resar

Perhaps it was a poor choice of words because I agree with what you're saying. It's entirely fair to include it in the review. What I meant was that beyond the applications that show a substantial benefit from HT, the improvement may not be nearly as much. So by selecting only benchmarks that benefit a lot from HT, the results can be spun to make it look like it's much more of an improvement than it actually is.

---

You show Pentium4 HT? Nehalem is SMT. Not hyperthreading. Also Core 2 arch that Nehalem uses is alot better for SMT or even HT.

Quote:

---

Originally Posted by Shintai

You show Pentium4 HT? Nehalem is SMT. Not hyperthreading. Also Core 2 arch that Nehalem uses is alot better for SMT or even HT.

---

now tell me what is the difference between HT and SMT???

i thought HT is just a intel trademark for smt

Quote:

---

Originally Posted by generics_user

now tell me what is the difference between HT and SMT???

i thought HT is just a intel trademark for smt

---

The implemtation is quite different. If a tread stalls the entire pipeline would reload, and thus the second thread aswell. Plus the P4 was a long pipeline and not so underutilized as Core 2. And nehalem is like half the pipeline length.

To think the 2 implementations mimmick one another is quite foolish. Also benchmarks sofar already shown huge effects of SMT on Nehalem, unlike the P4. So if you think SMT on nehalem yields the same performance as on the P4. You are either blind or way too deep into fanboyism.

And its funny we have so many newly created users that write like that.

http://www.xtremesystems.org/forums/...archid=7532913

also on most of the cases i liked ht on my old p4 2.6 yes it had bugs but actually after selling my 2.6ht and get a 3500+ i missed the performance of 2.6 on desktop multitasking a lot which i believe runs smoother then a 3500+ because of ht.

Quote:

---

Originally Posted by PetNorth

Looking at winrar and Everest numbers, TLB patch is applied.

My guess: due to it is a very early stepping, it suffers TLB bug too.

---

TLB patch is applied - Vista SP1 does it for all Phenoms whether they B2 or B3 or even 45nm
45 nm Phenoms are not cpus with tlb errata - but Vista thinks they are :ROTF:

A man who conducted these tests is not very smart. He even doesn't know the way to remove the patch - AMD Overdrive will help him

I think we had enough off topic (Nehalem,intel) in this thread.
Time to post something on topic.

I searched for TLB issues with Vista SP1 and came up with this great post:
http://www.hardwarecanucks.com/forum...html#linkbacks

Here you can see what effect SP1 had on stock 9600BE chip(Not OCed) under Vista.The thread goes into the only way of disabling the patch (done via sam2008's tool).
I won't post 3dmark numbers since we have none for Deneb system,but you can all see them in the thread i linked.
I will post Winrar scores though,since it COULD tell us a lot about the new Deneb chip and its performance under Winrar(it's obvious that a Multithreaded score of 518 for 2.3Ghz is with TLB patch ON,since the chip should score a lot more,above 1500 at least)

System used in the linked thread:

Quote:

---

* AMD Phenom 9600BE,
* Asus M3A32-MVP Deluxe,
* 4GB of Ram
* ATI Radeon 3870 512MB in Crossfire.
OS:Vista 32bit SP1=>TLB patch always ON

---

From the thread i linked:

WinRar 3.71 SP1 w/ TLB locked
http://images.hardwarecanucks.com/im...TLB/WinRar.jpg

WinRar 3.71 SP1 w/ TLB disabled by Application
http://images.hardwarecanucks.com/im...BDisabled).jpg

For comparison purposes,i found another score under Vista SP1 for TLB patched 9600BE chip over at Elitebastards review here.Note they used Vista SP1 64bit,so maybe that's why their score is a bit(16%) higher than the one from a user who's thread i just linked in the begging of the post.Also note that their TLB disabled results confirm that they didn't actually disable the TLB patch entirely,since the score is still too low,in the area of sub 400pts .This confirms the theory that Vista SP1 somehow overrides the BIOS option and sets the patch to Enabled all over again after boot up,but disables the CnQ which results in slightly better score in the end.Also,the AMD Overdrive tool can't override the TLB patch that Vista applies in SP1,but disables the CnQ.<-Read the thread about it @hardwarecanucks forum here
The Elitebastards actually scored 252pts so i will use this number as well as it correlates with the result(290pts) from the thread i linked.

For a Deneb's Winrar score @ 2.3Ghz (and under Vista SP1),we can use the only resource we have so far,the itocp thread :

System used:

Quote:

---

Deneb@2.3Ghz
Memory 2x1GB
MSI K9A2 board
Vista SP1(32bit)

---

http://www.itocp.com/html/amd45nm/winrar.jpg
Left score(518pts) is a multithread enabled score for default 2.3GHz Deneb.
Now compare this 518pts score @2.3Ghz for Deneb to ~280pts/250pts for Agena @ 2.3Ghz.The difference is obvious.Even if we go by a higher score from elitebastards(~340pts,MT enabled), Deneb is still 518/340=~50% faster at least(518/290=~1.8x at most).This is quite an improvement.It could be and evidence that L3 cache and other improvements are working great for Winrar,even though it's done under sub optimal settings(TLB patch).Also it could be a testing glitch(on the itocp's user side)...The result of 518 is evidently a MT score(the MT option is ticked) and is done with Vista's SP1 TLB patch forcing,which is quite obvious(a quad K10 chip is in the 1500pts region for 2.3Ghz clock).

So all in all,those itocp.com numbers are not so worthless after all ,since we have had 9600BE results for 6 or so months and we can compare Deneb's (Winrar at least) to these.I don't claim these are representative of final Deneb Vs Agena scores,just wanted to see what's the difference between the two under Vista's TLB fix plagued Service Pack 1.
Note:Agena's results in Winrar done under Windows SP1/2 are not directly comparable in any way to Vista SP1 scores,since TLB BIOS options actually work there(in XP Sp1/2) as well as AOD.

Quote:

---

Originally Posted by MAS

TLB patch is applied - Vista SP1 does it for all Phenoms whether they B2 or B3 or even 45nm
45 nm Phenoms are not cpus with tlb errata - but Vista thinks they are :ROTF:

A man who conducted these tests is not very smart. He even doesn't know the way to remove the patch - AMD Overdrive will help him

---

The ITOCP Deneb chip is C0, a 2007 spin (they announced 45 experimental in late 07, and why not believe them?). It is VERY likely that that revision has TLB.

I dunno when C1 was spun, but C2 is flawless as it's a 2008+ spin.

Quote:

---

Originally Posted by informal

Left score(518pts) is a multithread enabled score for default 2.3GHz Deneb.
Now compare this 518pts score @2.3Ghz for Deneb to ~280pts/250pts for Agena @ 2.3Ghz.The difference is obvious.Even if we go by a higher score from elitebastards(~340pts,MT enabled), Deneb is still 518/340=~50% faster at least(518/290=~1.8x at most).This is quite an improvement.It could be and evidence that L3 cache and other improvements are working great for Winrar,even though it's done under sub optimal settings(TLB patch).Also it could be a testing glitch(on the itocp's user side)...The result of 518 is evidently a MT score(the MT option is ticked) and is done with Vista's SP1 TLB patch forcing,which is quite obvious(a quad K10 chip is in the 1500pts region for 2.3Ghz clock).

---

saying deneb is 50-80% faster then agena with this results brave.
Noone know how the TLB patch affects agena, sure it is enbled, but does it affects it in the same exact way?
I doubt that, you can't draw any conclusions from that. You only generate a hype with saying deneb is 80% faster then agena with TLB enabled, and later that might only translate into a 10% increase...

I'll rather wait for real results then putting my hope into such "skewed" comparisons.

Quote:

---

Originally Posted by Hornet331

saying deneb is 50-80% faster then agena with this results brave.
Noone know how the TLB patch affects agena, sure it is enbled, but does it affects it in the same exact way?
I doubt that, you can't draw any conclusions from that. You only generate a hype with saying deneb is 80% faster then agena with TLB enabled, and later that might only translate into a 10% increase...

I'll rather wait for real results then putting my hope into such "skewed" comparisons.

---

Selective quoting much?
Read the last paragraph ,its 5cm below the quote you put.:rolleyes:

Quote:

---

So all in all,those itocp.com numbers are not so worthless after all ,since we have had 9600BE results for 6 or so months and we can compare Deneb's (Winrar at least) to these.I don't claim these are representative of final Deneb Vs Agena scores,just wanted to see what's the difference between the two under Vista's TLB fix plagued Service Pack 1.
Note:Agena's results in Winrar done under Windows SP1/2 are not directly comparable in any way to Vista SP1 scores,since TLB BIOS options actually work there(in XP Sp1/2) as well as AOD.

---

Anyways,new 45nm Fusion(45nm Deneb dual core+ 40nm RV810 graphics) from TGDaily:
http://www.tgdaily.com/content/view/38703/135/

Anyhow, nice write up informal! Althougb I do understand they wont be final numbers, it's surely an impressive difference.

And hornet, I dont think that TLB fix would be all that much differently applied on a Deneb than on an Agena tbh. In the end they just disable certain parts of a CPU and not like a random feature;)

OT, Those WinRar results while impressive are higher than I expect from the first batch of 45nm AMD processors. Probably 20% improvement clock for clock at best but if it's better than it's all good.

SHOW ME THE...Retail ;)

Perkam

Quote:

---

Originally Posted by perkam

SHOW ME THE...Retail ;)

Perkam

---

Seconded. :p We're seeing lots and lots of different results with ES chips. Blames are shifted everywhere. And theories are drawn a lot.

Internal war?