AMD Phenom TLB Patch Benchmarked and Explained

[Apoptosis]

AMD Phenom TLB Patch Benchmarked and Explained

When AMD was briefing the media about Phenom at their launch event they mentioned to the media that a fix was in the works for one of the hundreds of erratums that the Phenom processor has. All processors have glitches and they are documented and given a erratum number. Read on to see why erratum 298 made the news and if it is anything to worry about!

Using the PCMark Vantage test suite the difference between having the TLB patch workaround enabled and disabled was night and day. The system didn't feel any slower on the test bench, but the results showed differently. The performance difference ranged from 2% to 11.7% depending on the testing. It's obvious that the TLB workaround impacts performance.

Article Title: AMD Phenom TLB Patch Benchmarked and Explained
Article URL: http://www.legitreviews.com/article/618/1/

[DMB2000uk]

Do you think the fixed silicon will have this performance hit too?

Dan

[Apoptosis]

Do you think the fixed silicon will have this performance hit too?

Dan

I've asked AMD and I'll let you know their response. I don't think their new stepping is off the line yet, so they may not be able to give an honest answer with 100% certainty.

or I would be wrong... AMD PR said this in an e-mail just now...

new rev of silicon will not have a performance hit.

Looks like they are certain. If that is the case just a small bunch of 9500/9600 Phenom CPU's will have this erratum, which is great.

[DMB2000uk]

Well that is definately good news then.

Dan

[leexgx]

be nice if you could test an crash as i do use VMware some times

not seen any reviews in makeing the pc lock up with this TLB problem

allso on page 2 i think the Price is linking to an 9600 not an 9900

[Apoptosis]

be nice if you could test an crash as i do use VMware some times

not seen any reviews in makeing the pc lock up with this TLB problem

allso on page 2 i think the Price is linking to an 9600 not an 9900

As I've stated in the review we were unable to make it crash (even with VMware running), so it's not like we can make it crash. The reason the pricing link for the article is a 9600 is because a 9900 is not out yet. Ads are what keep the site running, so when a product isn't available on the market we use the next closest thing. The 9900 was delayed due partly to this issue, but when it comes out it will be a new stepping and have numerous erratum fixed.

Back to the crash... It is rare. AMD never did find this erratum. It was discovered when a customer found it that they were able to reproduce it. AMD hasn't said what exactly has to be done to cause the erratum, but since no one has found it, it goes to show how rare it really is. It's been six weeks and no one that I am aware of in the public or journalists have been able to get erratum 298. Honesty, it's all not going to matter in a month or so as a new stepping will be shipping and the problem is long gone.

[leexgx]

guess this TLB bug was mass spam for all review web sites

but as you stated this is AMD fault for Not telling users that thay can Disable it Via AMD overdrive tool and the crash is very rare even to the point it Never happen as like you sayed as well no one has made it crash (not seen any web sites with it doing so)

9600 Black edition

In particular, it should eliminate the notorious “ERRATUM 298” bug in the TLB block of L3 cache that can theoretically cause the system hang under heavy workload and in case virtualization technology is activated.

most motherboards have virtualization disable by default
(my asus m2n32-sli delux option is disabled {has any one had problems with bios 1501 running the HT link speed @ 200 and hideing 2 HT link speed options under chip set as i had to drop back to 1402 to resove the problem})

amd latey been to quite about when the Phenom was comming most likey due to the fact it was going to be slower then an q6600 but as long as the price point is good all should be ok

think AMD should scrap all single core CPUs now and lower the price of the dual core cpus thay are like only £10 from an single core now any way, Vista is an CPU and ram hog and single core cpus suck with vista

[A_Pickle]

Look, guys. This is ridiculous. Both you AND Tech Report have provided users with, unfortunately, useless TLB erratum benchmarks. You're great sites, but come on -- I could really care less about what Sandra or PCMark say I'm going to get. Fire up that copy of 3D Studio Max and whip out the stopwatch, or get some Paint.NET PDNBench action in, or Photoshop, or Premiere... I mean come on. Is it that much to ask for a demonstration using... you know... REAL programs? What if I wanted to buy one of these, and I'm fairly concerned about this TLB erratum and the effects it might have on my typical workload of Premiere, Office, 3D Studio Max, Paint.NET, Half-Life 2 and Call of Duty?

Synthetic benchmarks are useless. Please provide us with some real numbers from real applications that people use in day-to-day work, or don't provide anything at all.

[Apoptosis]

Look, guys. This is ridiculous. Both you AND Tech Report have provided users with, unfortunately, useless TLB erratum benchmarks. You're great sites, but come on -- I could really care less about what Sandra or PCMark say I'm going to get. Fire up that copy of 3D Studio Max and whip out the stopwatch, or get some Paint.NET PDNBench action in, or Photoshop, or Premiere... I mean come on. Is it that much to ask for a demonstration using... you know... REAL programs? What if I wanted to buy one of these, and I'm fairly concerned about this TLB erratum and the effects it might have on my typical workload of Premiere, Office, 3D Studio Max, Paint.NET, Half-Life 2 and Call of Duty?

Synthetic benchmarks are useless. Please provide us with some real numbers from real applications that people use in day-to-day work, or don't provide anything at all.

It was a tough call to post anything, but we spent so much time getting the bios patch, answers from AMD and benchmarking that we would have been stupid not to. My reservations about posting it were that new processor revisions down the road wouldn't have this erratum and the article would be pointless. What the article did point out was that the bug was minor and hard to replicate, a bios fix was produced, AMD's OverDrive can disable the workaround and that future processors would not have the erratum. If you real LR a bunch you'll notice that this article didn't really include many benchmarks... It just goes to show the article was directed more at getting information across to our readers rather than showing 10 benchmarking applications that don't matter to anyone in the months to come. I ran all the benchmarks, like in Cinebench R10 32-bit the score drppped 4 seconds with the patch... just didn't think it added to the take home message from the article. It's not like that article was to help you pick out which AMD processor is right or you. Hope you can see where I am coming from a little.

[wwarlock]

I for one like the post....

It gave me information I needed and helped me out. Thank you Apoptosis for doing the write up.