I've been having a high memory access latency problem with a Sempron LE-1150. The following email conversations I've had with AMD will fill you in on the problem. It's been a month since these conversations happened and I have tried a different Sempron LE-1150 and the same results (well almost, 282ns vs 285ns). Being a sempron, benchmarks are hard to come by. I've found older 90nm sempron results where the latency was similar to 90nm athlons as expected, but I haven't found anything on the 65nm LE-1*** semprons. Anybody got any ideas?
1: Second message I sent to them. I never got a reply or anything to the first one... maybe I entered my email wrong?
1: Second message I sent to them. I never got a reply or anything to the first one... maybe I entered my email wrong?
2: AMD repliesI built a system using an AMD Sempron LE-1150 (2GHz) with 1GB (2x 512MB) DDR2-800 running at DDR2-667 5-5-5-15-20-2T on an ASRock ALiveNF6G-VSTA. The memory latency is horrible, 285ns in Sisoft Sandra random memory latency test (http://img162.imageshack.us/img162/6277/le1150memorylatencyvq5.png). I picked the ALiveNF6G-VSTA because I built another system with it and was very happy with it. That system used an AMD Athlon64 3200+ (2GHz) with 2GB (2x 1GB) DDR2-800 5-5-5-15-20-2T. In Sandra it does around 100ns at stock speeds. After lots of switching of the RAM and CPU in each motherboard, I have ruled out the RAM as the problem. When using the 3200+ in either mobo with either set of RAM I always get under 125ns in sandra, even when setting up the 3200+ to be as much like the LE-1150 as possible (only differences being voltage, L2 cache, and orleans vs sparta core). It's hard to believe there would be such a huge difference in the 3200+ and the LE-1150. I've come to believe either this is normal for a LE-1150, or there is a problem with the memory controller on my LE-1150, or maybe the LE-1150 and ALiveNF6G-VSTA don't quite work so well together, even though it is on the supported list. I don't think the mobo is the problem however I had the same problems with this CPU in an MSI motherboard which is part of the reason why I ended up switching to the ASRock thinking it was the MSI board. I can't test again with the MSI board due to a bad BIOS flash (trashed the bootblock and all) and it's soldered on and I haven't had the time to remove it and add a socket. I don't have any other AM2 boards to test with.
I contacted ASRock and got a pretty generic broken english reply that said nothing. I contacted AMD 2008-03-03, but haven't heard anything yet. I seem to not be the only person who has ran into this problem as I found this (http://forums.pcper.com/showthread.php?t=450060) while searching on google.
3: Wheeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee!Dear <edited>,
Thanks for contacting Amd technical support.
If you compare the cpus you can see that there other difference:
-system bus 2000mhz vs 1600 mhz
-frequency 2200Mhz vs 2000Mhz
In addiction L2 cache 256kb vs 512kb.
So all these features can determine the different value of the latency timer and ,finally, you have to think that sempron family cpu was always a less performance cpu than athlon family.
Giuseppe, AMD ETSc
4: I think he really read it this time!Dear Giuseppe,
Thanks for the canned response and your lack of reading comprehension.
Not only did you fail to address the problem, you are also incorrect
in what you did reply to.
> If you compare the cpus you can see that there other difference:
> -system bus 2000mhz vs 1600 mhz
While almost technically correct (it's really 1000MHz vs 800MHz), if
you would have read what I wrote, you would have noticed that I stated
in my testing I set the Athlon 64 to 800MHz HT. That said, this
doesn't matter. As you should know, what with working at AMD, that all
recent AMD cpu have the memory controler ON CHIP. This means the CPU
talks directly to the RAM at the speed of the RAM. There is nothing in
the middle like the old days of memory controlers being on the
> -frequency 2200Mhz vs 2000Mhz
We are talking Socket AM2 CPU's here, not 754. The AM2 Athlon 64 3200+
is 2000MHz. I clearly stated in what I wrote that the Athlon 64 I was
compairing to was running at 2GHz. I also stated that I was using the
same model motherboard for both CPU's, and that I had tried both CPU's
in both motherboards. Obviously I'm not sticking a Socket 754 CPU into
an AM2 socket, and I am unaware of any adapters that allow 754 CPU's
to fit into AM2 sockets, so from the context it was clear I was
talking about socket AM2 CPU's.
> In addiction L2 cache 256kb vs 512kb.
While true the Sempron LE-1150 has only half the L2 cache of the
Athlon 64 3200+, that doesn't play a significant enough role to
account for the outrageous latency I am seeing. Near 300ns is probably
worse than what we saw back in the day with FPM SIMMs. Don't quote me
on that, I have done no testing to back it up, but it is just to
further my point just how crazy the latency I am see with this CPU is.
> So all these features can determine the different value of the latency timer
> and ,finally, you have to think that sempron family cpu was always a less
> performance cpu than athlon family.
True, the Sempron is the budget AMD CPU line, but the performance
differency in the memory access latency wouldn't be that significantly
different. Especially not between a 2GHz Socket AM2 Athlon 64 3200+
and a Sempron LE-1150 which are very similar (be it that they do have
different cores, L2 cache, and hypertransport speeds).
I'd like to take this time to apologize to whoever had to read this,
since in all probability, it won't be "Giuseppe". I would like it if
someone at AMD would take a more indept look at this and come up with
a more concise answer. It really does look like some CPU's shipped may
have memory controler problems as I have shown with my CPU and
"Cataligh" who made the post here
(http://forums.pcper.com/showthread.php?t=450060) who has the same
problem, with an AMD Athlon 64 X2 4000+ (kind of makes it hard to
believe my problem is just because it's a sempron). For the record, if
I run the RAM with the Sempron LE-1150 at DDR2-800 4-4-4-12-16-2T, I
get the same numbers as "Cataligh" (in the 190ns). I'm only running at
DDR2-667 5-5-5-15-20-2T to avoid people jumping to the conclusion that
this outrageous latency is because of running too tight timings or
running the RAM at DDR2-800 when the Sempron only officially supports
5: I can be nice... no really... see!Dear <edited>,
Thanks for contacting Amd tech support.
-While almost technically correct (it's really 1000MHz vs 800MHz),
It's true but you should know that HT uses both edge of clock cicle so the real frequency becomes 800Mhz x2 or 1000Mhz x2.mMoreover as you says these cpu are both memery controller on board and they comunicate with memory directly so it's normal that a higher HT is faster in access memory than a lower. Do you agree?
-frequency 2200Mhz vs 2000Mhz ,
sorry this was a my fault
-- addiction L2 cache 256kb vs 512kb
yes you're right.Cache is important but probably don't justify a latency so high.
Finally i don't know if the problem you have is a problem of your cpu.It should be interesting check another Sempron LE-1150 in your system and see what happens.
For example a few day ago I can olny tell you that i have received another email where the customer complains the difference between real and setting frequency of FSB(10% lower) in his system with sempron 3400+. You know that in elettronic field,sometimes, it happens. So you could test another cpu (same model) in your system we could know if this is a real issue of your cpu or not.
6: Gene joins the party...Dear Giuseppe,
I thank you for your more thorough reply this time.
> It's true but you should know that HT uses both edge of clock cicle so the
> real frequency becomes 800Mhz x2 or 1000Mhz x2.
Not quite. While a Double Data Rate bus performs twice as much work
per a clock cycle, making it effectively twice the clock speed, it's
still not technically twice the clock speed. It's just the same as the
Quad Data Rate bus intel uses, it effectively quadruples the clock
speed, but the clock speed is still technically 1/4 the effective
speed. That said, this isn't of any importance, and I was just being
needlessly anal because of the canned reply I got.
> Moreover as you says these
> cpu are both memery controller on board and they comunicate with memory
> directly so it's normal that a higher HT is faster in access memory than a
> lower. Do you agree?
I do not agree. The memory controler on AMD CPU's comunicate to the
RAM directly at the speed of the RAM and not at the speed of
> Finally i don't know if the problem you have is a problem of your cpu.It
> should be interesting check another Sempron LE-1150 in your system and see
> what happens.
> For example a few day ago I can olny tell you that i have received another
> email where the customer complains the difference between real and setting
> frequency of FSB(10% lower) in his system with sempron 3400+. You know that
> in elettronic field,sometimes, it happens. So you could test another cpu
> (same model) in your system we could know if this is a real issue of your
> cpu or not.
I do agree, the only way I can be 100% sure is to test with one (or
more, to reduce the theoretical possibility of getting another CPU
with the same possible fault) Sempron LE-1150. That said I am pretty
close to 100% sure as it stands now. Having tested in three
motherboards, two of the same model by ASRock and one by MSI, It's
pretty safe to say it is probably the CPU.
7: I replyDear <edited>,
Thank you for contacting AMD Global Customer Care.
Have you tried any other brand of RAM aside from what you have now? Please be aware that the LE-1150 is a very entry level processor; it has various core differences than the Athlon 3200+ which included memory controller implementation and speeds, including latency speeds. The latency you see if perfectly normal and can also be credited to RAM, processor and motherboard combination. If you use faster RAM than what you currently have, you will most definitely see lower (but not by much due to the limitations of the processor) latency speeds.
If you have any questions, please contact us at 408-749-3060 (US CPU Support) or 44-1276-803299 (EU CPU Support).
AMD Global Customer Care
8: The return of Giuseppe!Dear Gene (or whoever reads this)
If you read through the conversation that transpired between Giuseppe
and myself, you would have found all of your questions and thoughts
had already been addressed. I recommend you re-read through our
conversation, but for simplicity I will also address them here.
> Have you tried any other brand of RAM aside from what you have now?
As stated in the first message I had tried two different sets of RAM.
Since that time I have now also tried a third set of RAM.
> be aware that the LE-1150 is a very entry level processor; it has various
> core differences than the Athlon 3200+ which included memory controller
> implementation and speeds, including latency speeds.
This has also been addressed many times. While true there are
differences between an Athlon 64 3200+ and a Sempron LE-1150, none of
them can account for or give justifiable reason for the outrageous
access latency I am seeing. I have also given an example of where this
same problem has occurred with an Athlon 64 X2, pretty much nullifying
the reasoning that this only happens with budget Semprons.
> The latency you see if
> perfectly normal and can also be credited to RAM, processor and motherboard
I'll grant you while it is possible for the RAM and Motherboard to
also play a roll in this, the fact that I have tested with three
motherboards, and three sets of RAM pretty much points to the CPU. If
you are trying to claim that an access latency of 285ns is "perfectly
normal", I will have to say that I have lost faith in AMD and will
never buy another AMD product again. However I know this isn't
"perfectly normal", so I'm not worried about that. After all, I am an
AMD fan, and my first computer I owned was powered by an AMD 8086. So
I've been using AMD products for a long time, and would like to
continue to. I still have that computer, and it still works, and I bet
that even with the 8086 and 384KB it probably has a better access
latency than 285ns (again as with a joke about FPM SIMMs in an earlier
message, I've done no testing to prove this, so just take it as it is,
a joke about how outrageous 285ns is).
> If you use faster RAM than what you currently have, you will
> most definitely see lower (but not by much due to the limitations of the
> processor) latency speeds.
As I stated in another message, I am only running at DDR2-667
5-5-5-15-20-2T to avoid people jumping to the conclusion that this
outrageous latency is because of running too tight timings or running
the RAM at DDR2-800 when the Sempron only officially supports
DDR2-667. I can run the RAM with the Sempron LE-1150 at DDR2-800
4-4-4-12-16-2T, and while I get a lower access latency time (as would
be expected), it's in the 190ns, which is still outrageous.
> If you have any questions, please contact us at 408-749-3060 (US CPU
> Support) or 44-1276-803299 (EU CPU Support).
I would have probably done that by now, but I do so hate talking on
thanks for contacting Amd tech support.
"I do agree, the only way I can be 100% sure is to test with one (or
> more, to reduce the theoretical possibility of getting another CPU
> with the same possible fault) SempronLE-1150."
According to your last statement i think that now it's important test another cpu SempronLE-1150 in your system to verifity if your cpu is fault or not.
Let me know.
Giuseppe, AMD ETSC