AMD Piledriver rumours ... and expert conjecture

Page 30 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
Reynod

Reynod

Administrator
Mar 26, 2007
14,921
16
42,215
We have had several requests for a sticky on AMD's yet to be released Piledriver architecture ... so here it is.

I want to make a few things clear though.

Post a question relevant to the topic, or information about the topic, or it will be deleted.

Post any negative personal comments about another user ... and they will be deleted.

Post flame baiting comments about the blue, red and green team and they will be deleted.

Enjoy ...
 
Sort by date Sort by votes
No L3 will depend on how RAM is used/accessed (IMC taking most of the logic for that) and the latency it will have. There's also QPI (Intel) and HT (AMD) that provide annoying ceilings for transfer and latency. At least AMD has been working on access latency with Llano and we'll see a better stepping with Trinity (there was a link as well, too lazy to look for it, lol).

And DDR4 won't improve latency overall AFAIK, but at least will bump the bandwidth quite a lot. I think there was an Article over XBits about it.

And also, getting bigger L2 or L3 won't always come as a boost overall. The arch/logic has to be made with certain cache sizes in mind. Ugh, it's murky waters in there for me, so I won't get my feet dirty, lol. Anyone willing to grab the baton? xD!

Cheers!
 
Upvote 0 Downvote
Sorry to keep going off topic but in my A+ class i was told L3 cache is 3 times faster then ram.
Click to expand...

+1
and I had to dig into my A+ books and find the info for myself, all I can say is this is the final word and it's fact.
period..
-----------------------

A cache is a chunk of faster memory that holds information for faster retrieval than would otherwise be possible.
One analogy would be to the cash you have in your pocket - it's faster and easier to access and spend than money in the bank.

In the same way, compared with CPU speed, system memory is slow. By caching data either on the CPU itself or in a special card with high speed memory,
your CPU won't be hobbled by slow system memory.

L3 cache can be far larger than L1 and L2, and even though it’s also slower, it’s still a lot faster than fetching from RAM.

Fetching instructions from cache is faster than calling upon system RAM, and a good cache design greatly improves system performance.
Cache design and strategy will be different on various motherboards and CPUs, but all else being equal, more cache is better.
Click to expand...

I usually are agree with the text and books, but why not test it and know the real performance?
 
Upvote 0 Downvote
in the annadtech bulldozer review they tested the latency of the ram access and the l3 access.

41721.png


41731.png
 
Upvote 0 Downvote
It's like performing 21 divided by 3 in your head. It is faster than using pen and paper (RAM), but there is a limit on how many sums you can do in your head (cache).
Click to expand...
Now that you put it that way, it makes a bit more sense to me.
So basically the CPU cache helps in reducing time for process vs RAM, and there aren't certain restrictions or constraints RAM would put on the CPU processing.
 
Upvote 0 Downvote
L3 Cache is for worst case scenario's, I know I've told many of the posters this before. L3 Cache in and of itself will not give you any performance improvement, it should never be used, ever. Of course that's in a perfect world where your prediction engine is working and everything you need is already located in L2 prior to it being needed. In a not so perfect world bad things do happen and sometimes mispredictions do occur. When a needed block of data isn't in L2 then your CPU must stall while the memory controller is asked to go fetch the data from RAM, this takes a very long time relative to the CPU. It's not just that L2 cache is multiple times faster, is the additional latency and bus protocols required. The data is eventually retrieved from memory and your CPU goes about it's business. If this is a rare occasion then it's no big deal, but if your predictor is constantly missing then and only then does L3 cache save the day.

To sum it up,
good prediction = no need for L3,
bad prediction = absolute need for L3.

The above statements about doing math in your head vs paper only applies to L2 cache not L3. L3 is just on-die system memory, and honestly its the one part of the CPU you can easily afford to trim if you need room.
 
Upvote 0 Downvote
and by downclocking my system, i got this
l1 5
l2 26
ram 245

config of downclock
fsb 150
cpu mt (multiplier) x8
nb x4
ram 400mhz (but @ 300mhz) 7-6-6-6-18-26-2t
ht x1

according to results (in k10)
it seems that latency of
Ram = 10x L2
L2 = 5x L1
Ram = 50x L1
 
Upvote 0 Downvote
using latency benchmark of pcwizard i got this result
latency of
l1 3
l2 15
ram 159 (of course it is taking time because it is ddr2 667mhz with cl5 and 1t)
Click to expand...

l1 5
l2 26
ram 245
Click to expand...

Thats about normal. L1 should be able to be accessed within 3-5 cycles, and the L2 historically has access times around 20 or so cycles. For RAM, 100+ cycles is the norm [though 245 is awfully slow...].

The point being, the L3 isn't significantly faster then accessing main memory, so its not that big a performance benifit.
 
Upvote 0 Downvote
well i have been reading S|A for 9 months now.
all of their news has been accurate, and original, which cant be said even for Tomshardware news.

and their analysis/rants turn out to be true.
Click to expand...


I think to be completely fair Charlie (semiaccurate) has good inside knowledge when it comes to AMD .... beyond that he is at the same level at the rest of the small sites.

Charlie writes well and his stuff is humorous ... he learned from the best ... Mike Magee.

Here Chris and his team have good inside knowledge for Intel in particular, and NVidia, and as Toms reviewers sign NDA's they receive ES and have time to benchmark products prior to release, so a good release review is up there on day one.

Our reviews are usually very comprehensive.

Toms editorial team follow the ND rules set out by the manufacturers and follow a strict code of conduct ... small sites don't generally.

You also need to realise Charlie's team have no benchmarking and review team, no news team and no coders sittig there tieing 6 different mirros together across the world.

They have no sister sites (Toms Guide / T Style / PCPresence) with a focus on light news and entertainment and electronic gadgets, and weird French idiosyncracies.

Toms also has argueably the largest and most active forum of this type in the world ... with 5 staff and 20 moderators.

Toms is ranked 1043 on the Alexa global website benchmark - http://www.alexa.com/siteinfo/tomshardware.com%3Cbr%20/%3E

SA is ranked 52,927 - http://www.alexa.com/siteinfo/semiaccurate.com

Personally I like Charlie and his team over there (I know Charlie and Copper) and I don't like users here bagging other hardware sites ... unless you want to link to the Ottolini incident where Charlie had to wear the bunny suit (he wore the other bunny suit) as he lost a bet. :)

Most of us have links to a dozen or so sites that we check regularly ... I even like to visit Fuad, AMDZone, and a number of others ... but I call this home.

When I have a criticism to make here, I pass it on dierect to the Editor, and I offer something constructive. I must admit the staff are keen to improve anything.

Thats why the site is ranked so high ... people keep coming here ... we try to give them what they want.


:)

 
Upvote 0 Downvote
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom