Phenom TLB patch performance

epsilon84

Distinguished
Oct 24, 2006
1,689
0
19,780
http://techreport.com/articles.x/13741/1

Also includes benchmark results between the ES samples for reviewers (with a 2GHz NB/L3), and the retail versions, with the slower 1.8GHz NB/L3.

All in all, there is a 14% performance penalty with the patch enabled, and 20% if you include synthetic memory benchmarks.

There is nothing else to say, except... DAMMIT.
 

Mandrake_

Distinguished
Oct 7, 2006
355
0
18,780
I think it's time Ruiz got his marching orders! So for those saying a Q6600 was 13% faster than a Phenom, the difference is now closer to 30%!
 

epsilon84

Distinguished
Oct 24, 2006
1,689
0
19,780
Maybe AMD should't release the patch it just makes things worse

If stability is paramount (ie. mission critical tasks) then the patch is a must. For most people, I would say a rare crash is better than an overall 14% performance hit.
 

Kamrooz

Distinguished
Feb 8, 2007
1,002
1
19,280
Truly a sad day for amd. With all the problems they've occurred, to slap this on top of the list is just disappointing. The B3 stepping which will resolve this issue will be shipped and available for purchase around March 2008 from what it seems. That's not a very good sign. If AMD can't pull in a large sum of profit by the end of 2008 (Nehalem Release), then, sadly to say, AMD might end up going under. Or downsizing significantly to become a smaller company that will lose AMD's current aim for the future.
 

Kamrooz

Distinguished
Feb 8, 2007
1,002
1
19,280
Behind atm I'd believe. They haven't done a comparison benchmark yet between the two, also too lazy to search for individual similar system benchmarks to compare. But it should be behind with this new bios patch.
 

Kamrooz

Distinguished
Feb 8, 2007
1,002
1
19,280
AMD really did not need this. With this stacking as well as a march issue release for the B3...There are going to be excess stock from these processors building up. With the march release, doesn't give much time for them to pull profit. If they don't get high clocked Phenoms in the 2.8 ghz range at least by the end of 2H/3H 2008...Sad to say, AMD might go under..With Nehalem due at Q4...Also with the word about it's massive performance gains, it's not looking good for AMD.
 

tstebbens

Distinguished
Dec 4, 2006
26
0
18,540
It is distressing in that it's such a large performance hit, but remember that these are server chips first and foremost. The benchmarks used were all end-user benchmarks. Who runs Office, Photoshop or Firefox on a server? It would have been better to include benchmarks for web and DB servers, etc.

If you really are using Opterons in an end-user environment (high-end workstations or gaming machines) then you've got way more money than sense anyway. You should be waiting for the Phenom FX chips in this case anyway or just wait 3 months more for the fixed silicon. 3 months is not going to kill anyone.

And no.... AMD will not go under. Bought out or merged maybe, but not bankrupt. They have way too much IP and other assets for this to happen.

Besides which, people seem to forget that AMD still have the 45nm switch over to make plus much higher clock speeds to release, and Intel have to integrate their memory controller on chip and switch to HT bus yet. With AMD's aggressive pricing policy they only need to cover their costs to stay in the game and still increase their market share which they are doing very well at the moment.

My 4x4 platform (memory, mobo and processors) cost half what an equivalent benchmarking Intel quad core plaform cost, plus I can have 8 processing cores in the future without upgrading the mobo or memory so in the long run, it's going to cost around 1/3 of the equivalent upgrade path from Intel. For the money I save, I can live without the extra few fps in games or an extra few seconds here and there transcoding videos or audio.
 

epsilon84

Distinguished
Oct 24, 2006
1,689
0
19,780


I'd love to know how you saved 1/2 the money on a comparable Intel platform when a Q6600/P35 board costs around $400. How much does an FX-74 / Asus QuadFX board cost? $700? $800?

I'm not even taking into account the power consumption here...

And upgrading to 8 cores? Didn't you read the news? AMD is dropping support for QuadFX... so much for that then...

 

zenmaster

Splendid
Feb 21, 2006
3,867
0
22,790


The telling one for that is the Memory Throughput test which tanked.
The Excellent Memory performance was the one main thing that really helped AMD.
A big hit there is a big problem.
 

homerdog

Distinguished
Apr 16, 2007
1,700
0
19,780
Talk about getting kicked while you're down. Now I see why AMD has decided to bring back K8. It might be slower than Core2, but at least it's not broken.
 

shadowmaster625

Distinguished
Mar 27, 2007
352
0
18,780
If AMD cant raise their IPC, they deserve to go under. A single phenom core needs to outperform a single athlon 64 core by at least 50%, considering its been 4 years. That's 12% a year. What they're basically saying is their IPC has improved by only 1% per year. That's horrid.
 

bfellow

Distinguished
Dec 22, 2006
779
0
18,980
Wow a 19.8% average drop in performance (Firefox sticks out with 57.1% drop) with the TLB patch and 13.9% without including the memory tests. That is truly disturbing
 

greenspoon

Distinguished
Nov 28, 2007
83
0
18,630



I thought AMD has lost $400M the last few quarters (3 or 4 quarters?). How is that covering costs? I am not understanding what you are saying I guess.
 

Kamrooz

Distinguished
Feb 8, 2007
1,002
1
19,280


If affects both Desktop and Servers, Phenom is desktop while Barcelona is the server line. the TLB effects the B2 steppings of each line. So unfortunately it does matter for desktop users.

http://techreport.com/discussions.x/13724

If you really are using Opterons in an end-user environment (high-end workstations or gaming machines) then you've got way more money than sense anyway. You should be waiting for the Phenom FX chips in this case anyway or just wait 3 months more for the fixed silicon. 3 months is not going to kill anyone.

It will kill AMD. At march we should see the B3 stepping if it isn't delayed. There is also plans to release a 2.6 ghz Phenom in this time frame also. If this gets delayed again, it puts AMD in big trouble. They may be gaining marketshare, but it's mostly lower dual core X2's through OEM's/Brand name desktop channel.

And no.... AMD will not go under. Bought out or merged maybe, but not bankrupt. They have way too much IP and other assets for this to happen.

That is true, even though I stated going under, AMD really won't die, it will downsize drastically, become a privatized company, or be bought out/merged. Regardless though, it will somewhat feel as if they will go under...We won't see many new products for quite a while as AMD would rebuild itself from the ground up.

Besides which, people seem to forget that AMD still have the 45nm switch over to make plus much higher clock speeds to release, and Intel have to integrate their memory controller on chip and switch to HT bus yet. With AMD's aggressive pricing policy they only need to cover their costs to stay in the game and still increase their market share which they are doing very well at the moment.

AMD is having huge issues at 65nm..With their SOI process, it makes it extremely difficult to even get down to 45nm. Heck, they are having trouble with it at 65nm. The difference though, Intel already has working Nehalem's, the performance increase is said to be massive compare to what we've seen in the pasts Netburst to Core transition. AMD MUST make profit, not break even. They have to attempt to pull in some profit before Nehalem is released for R&D funding. AMD's only chance of recovering without being bought/merged/privatized, would be to pull in some profit from X2/Barcelona before Nehalem is released at the end of 2008. Also keep in mind intel has been ahead of the gun lately in releases, chances are they might do the same with Nehalem, to put more pressure on AMD. March may seem close, but it is indeed far away as well, if they have issues getting faster clocked Barcelona/Phenom models out as the time ticks away to Nehalem...It literally, does not look good for AMD.
 

gallag

Distinguished
May 3, 2006
127
0
18,680


bs, 4x4 was the most expencive quad solution in history and the most power hungry and the slowest.

[pointing and laughing mode on] Good luck with the whole "8 processing cores in the future without upgrading the mobo or memory" thing as AMD don't seem to be interested in you 4x4 suckers anymore lol. I mean who was stupid enough to buy 4x4 lololol hahahahaha na na naaa na na naa you bought 4x4 [pointing and laughing mode off]
 

wolverinero79

Distinguished
Jul 11, 2001
1,127
0
19,280
This kind of paints a sad picture for the troubled company. Even if they fix both the bug and return performance to the point where it was pre-patch with the next stepping revision, they still have a product that is not competative (and by the time they get that fixed revision, we may have Nehalem benchmarks). Ouch!
 

speedbird

Distinguished
Apr 19, 2007
547
0
18,990
AMD phenom suffered delay after delay, but when they actually finally release the processor it is unstable! :sarcastic: Then to add a further insult a patch is avaliable to completely ruin performance.

This is humiliating for AMD, there's no other word for it. I would rather have had further delays and a product that actually worked properly on launch.

Shame on AMD :pfff:
 

BaronMatrix

Splendid
Dec 14, 2005
6,655
0
25,790
I'm still waiting for people to do real tests with ganging and unganging the HT3 links. Some Chinese site did it and got a nice increase. Also, one with the L3 disabled would tell a lot about the bug and how effective the L3 actually is.
 

epsilon84

Distinguished
Oct 24, 2006
1,689
0
19,780


I've sure you've seen this site already Baron... perhaps you just chose to ignore these numbers.

http://www.erenumerique.fr/test_processeur_amd_phenom_x4_9600_drole_de_rel_ve_-art-1804-11.html

Ganged vs unganged doesn't have a big difference on overall performance, do you have a link to the Chinese site?

 
BM - Too bad Many of the top Exec do not share your optimism.

Ref Insider trading for AMD for 01/03/07 -> 11/27/07:
1, repeat ONE buyer (3600+ shares)
6 Sellers for 163,800 shares. (Mottol Topfer sold 100,000 on 07/24/07)

Normally spelling is not my forte, sorry but I'm just a dumb ET.