rp4440 TPM figures

[Martha Mueller]

You are such a treasure! Thank you very, very much. I have them now.

[Ted Buis]

I don't think the rp4440 TPM figures are true TPC-C results as they are not posted on the www.tpc.org web site. The official tests must be audited and are very expensive to conduct, so they are not done for every system. Likely the ones supplied are estimated. Maybe Cheryl can confirm?

[Ted Buis]

With the new pa-8800 processor the rp4440-8 has as roughly the performance potential than the rp7410-8, but does have fewer I/O slots, and does not support hard partitioning (never will) or vPars at least at the moment and possibly never. There are also some differences in bus bandwidths and latency that can be a factor. The 8-slot replacement for the rp7410 is the rp7420-16, which certainly redefines the mid-range along with the rp8420-32. With the dual core pa-8800 everything doubled. Some slight differences in the chip design to note: the pa-8800 has 1.5MB L1 cache per core and a 32MB L2 cache shared by both cores on the chip. The PA-8700 has 2.25MB L1 cache, so on a per core basis, there can be differences in performance that are not purely due to the clock rate difference.
The bottom line is that the 1GHz pa-8800 rp4440-8 8 CPU system should be close to the
peformance of the and the 875MHz pa-8700+ rp7410-8 8 CPU system, but it could vary by application. However the rp7420-16 is very likely to beat the rp7410-8 with the same number of CPU cores and still has plenty of room for 2X CPU growth.

[rick jones]

Indeed there are a number of rules for disclosure of TPC benchmark results, most of them rather stringent, particularly when it comes to "public" disclosure (non-NDA) of results that have not been audited if not published on www.tpc.org. I suspect the gory details are burried somewhere on www.tpc.org.

This forum would I suspect constitute a "public" disclosure, as I suspect would a 1800 number.

[Martha Mueller]

I can understand that. But it would be nice if we could have something that had a disclaimer, so we could simply use HP's word on their various products to compare different HP models. I can see where we would want auditors for offical results used to compare different vendor's equipment.

[Ted Buis]

From www.specbench.org or SPECint_rate2000.
(CINT2000 rates) Higher is better.
system #CPUs base peak
rp7410-8 w/875 8 (2 cells) 53.1 55.8
rp4440-8 w/1GHz 8 70.3 73.2
rp7420-16 w/1GHz 8 (1 cell) 68.7 71.7
rp7420-16 w/1GHz 8 (2 cells) 67.8 70.8
The 2 cell rp7420-16 may improve with 11i v2 for pa-risc if it has cell local memory support as does 11i v2 for Integrity Servers and has the potential for 2X the I/O of a one cell solution. Note that there is NO I/O in this benchmark at all, and as the SPECint relatively code is small, it can often fit in cache. Remember also that the rp7xxx products support vPars and nPars, and the rp4440 does not. The rp4440 does not have as many I/O slots and cannot be upgraded or expanded as far as the rp7xxx products, as it only has 4 sockets for processor modules.

[Martha Mueller]

I appreciate the extra information. That kind of help is wonderful to know, and very hard to get. It's tough to get past the "sales speak" to try to distill down what features the various models have. It's always harder to find what is missing rather than what is advertised....sort of like the old "Do you want air in those tires?".

[Ted Buis]

Just remember that your "mileage may vary". Different applications stress systems at different points, so performance really depends on the application. Also, some customers are using vPars/nPars, and some are not, so that will change the value of the rp74x0 relative to the rp4440-8.

[Emanuele Guarneri]

Hi Adam,
is it possible for you to send me the TPM figures for 4440. Now I need it too.
thank u so much.
EManuele

[Adam Noble]

rp4440, 1 GigHz:
1 cpu: 38,800
2 cpu: 66,400
4 cpu: 112,700
8 cpu: 191,300

rp4440, 800 MHz
1 cpu: 35,600
2 cpu: 61,000
4 cpu: 103,400
8 cpu: 175,500

Here you go!